{
  "arm_map": {
    "cg-ask-fz-009": "declared"
  },
  "docker_images": {
    "cbm": [
      "bench-cbm:latest@sha256:7956959468dbe6a7ac4c0f2c4a3170aacef4f3bae5397c55378adc0e8c2094d8",
      "bench-cbm:latest@sha256:be5b5e16f84b72002cad0738d49e3ea3a9885797af6f758844a7efc0e70461c0"
    ],
    "codegraph": [
      "bench-codegraph:latest@sha256:22397b9c7030c6b62b6bd1d32acec433be23baece17ec3243c45fb4fc4a7ef3b",
      "bench-codegraph:latest@sha256:d8d443436bdfd2ba3fe44dde5fe3380bfe7f3f50da7be5d13cd39970cc917132"
    ],
    "codex": [
      "bench-codex:latest@sha256:b0f8c410feb52327826936bf88a87c1ad67375e6a44e6e0acc81937d553fb31b",
      "bench-codex:latest@sha256:d33cb29bbfbcddc95f57c364104fdb8410261934a255f955c6e8d5405fbb0fa5"
    ],
    "cold": [
      "bench-cold:latest@sha256:11d4f976d7a2c9004e2ba863e9832ff8833da24a216b91577fd345b9e9b55afd",
      "bench-cold:latest@sha256:6d6b255d2b5ca888313fb74ce7fcba3794cc4fbe24316179f1809a58829dbb7c"
    ],
    "cursor": [
      "bench-cursor:latest@sha256:2099eda34b442e67234af03c3e40470983f02121154e87d5d662bccd165e9292"
    ],
    "gitnexus": [
      "bench-gitnexus:latest@sha256:18e6e748f657b8826739c03a03b237a647d01aa2dfc80937edfd45799e9034a9",
      "bench-gitnexus:latest@sha256:c801d8074cc25e20ec2cc153f7c8eb6cbc9bef7d1c48ffba936e379a2c6b655a"
    ],
    "graphify": [
      "bench-graphify:latest@sha256:be790a7f4910b25780a377531f5ab27a6ff596de0fa0dda52c0501c6a1d39e22",
      "bench-graphify:latest@sha256:e6cfefa3a9719bac693e87a1dd120007be0b29d2a23141d5bdedefd3f29f38c4"
    ],
    "serena": [
      "bench-serena:latest@sha256:94899b440b263db1121d6dcbc87a67569c127b28c211d8db794231079ce798b8",
      "bench-serena:latest@sha256:e4490457a72e73d237b181596d9fae5003d32b3f68e162d56bd0629c64bf0c6f"
    ]
  },
  "epoch": {
    "id": "E004",
    "label": "Epoch 004 — line-4",
    "no": 4,
    "notes": "Indexer v1.4.7 — Corograph v0.11.4 — Dev v0.4.16"
  },
  "export_tool": {
    "name": "tabench export public",
    "schema": "public-run/1",
    "version": "0.1.29"
  },
  "freeze_map": {
    "fz-009": "declared"
  },
  "freezes": {
    "fz-009": {
      "instrument": {
        "Corograph": {
          "engine_version": "0.11.8",
          "module": "v0.11.8",
          "replace": "../ta-indexer/pkg/Corograph",
          "surface_content_hash": "dc074c6eb47fc8f6",
          "surface_min_engine": "0.11.8",
          "surface_source": "embedded",
          "surface_version": "2.4.2"
        },
        "examiner": {
          "commit": "994aec27df526222bc53da219a819e2d1c083a37",
          "go": "go1.26.1",
          "version": "0.1.13"
        },
        "indexer": {
          "ir_meta_version": "1.0",
          "ir_schema_version": "1.0",
          "module": "v1.4.74",
          "replace": "../ta-indexer",
          "version": "1.4.74"
        },
        "modules": {
          "connectrpc.com/connect": "v1.19.1",
          "github.com/Corograph/ta-dev": "v0.0.0 =\u003e ../ta-dev",
          "github.com/Corograph/ta-indexer": "v1.4.74 =\u003e ../ta-indexer",
          "github.com/Corograph/ta-indexer/pkg/Corograph": "v0.11.8 =\u003e ../ta-indexer/pkg/Corograph",
          "github.com/Corograph/ta-proto": "v0.0.7 =\u003e ../ta-proto",
          "github.com/google/uuid": "v1.6.0",
          "github.com/scip-code/scip/bindings/go/scip": "v0.7.1",
          "github.com/sourcegraph/beaut": "v0.0.0-20240611013027-627e4c25335a",
          "golang.org/x/mod": "v0.35.0",
          "golang.org/x/net": "v0.53.0",
          "golang.org/x/sync": "v0.20.0",
          "golang.org/x/text": "v0.36.0",
          "golang.org/x/tools": "v0.44.0",
          "google.golang.org/protobuf": "v1.36.11"
        },
        "ta_dev": {
          "module": "v0.0.0",
          "replace": "../ta-dev",
          "version": "0.4.17"
        },
        "ta_proto": {
          "module": "v0.0.7",
          "replace": "../ta-proto"
        }
      },
      "label": "fz-009",
      "minted_at": "2026-08-04T00:25:35Z"
    }
  },
  "generated_at": "2026-09-29T20:46:08Z",
  "repo_map": {
    "memos": "usememos-memos"
  },
  "repositories": {
    "usememos-memos": {
      "batteries": [
        {
          "battery_key": "memos",
          "content_revision_id": "c66b913000d39be11929de2eed3c57d6",
          "denominator": 30,
          "exemplar_kinds": {
            "correct": 57,
            "incorrect": 60,
            "partial": 29
          },
          "exemplars": 146,
          "exemplars_adjudicated_by": [
            "memos-agent"
          ],
          "exemplars_adjudicated_from": "2026-08-05T04:15:07Z",
          "exemplars_adjudicated_to": "2026-08-05T04:38:05Z",
          "exemplars_verbatim_in_answers": 15,
          "gold_revised_questions": 3,
          "gold_rows": 33,
          "questions": 30,
          "title": "battery-1 (30q)"
        },
        {
          "battery_key": "memos-hard",
          "content_revision_id": "c66b913000d39be11929de2eed3c57d6",
          "denominator": 20,
          "exemplar_kinds": {
            "correct": 67,
            "incorrect": 59,
            "partial": 41
          },
          "exemplars": 167,
          "exemplars_adjudicated_by": [
            "memos-agent"
          ],
          "exemplars_adjudicated_from": "2026-08-05T01:54:29Z",
          "exemplars_adjudicated_to": "2026-08-06T23:19:29Z",
          "exemplars_verbatim_in_answers": 0,
          "gold_revised_questions": 3,
          "gold_rows": 23,
          "questions": 20,
          "title": "battery-2 (20q hard)"
        }
      ],
      "commit_sha": "2d01420c23372d5708904ec78e03485edb44d49b",
      "repo_key": "memos",
      "source_url": "https://github.com/usememos/memos"
    }
  },
  "runs": [
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "7f875fab4731097865450fd84bf631afba83fb572ad9252859e8f52bb64ba483",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "25ed200cb6e17e61b706c1b3f5f53850b323565e9aea56026621553ca74123cb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b790d50cd824d569b7c1e4466fad97bb94e31e682dd6142bf34f8e67c17cbc8a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "4451c53693c07cd6536978c0f9e2d60c570c9613364553e181e6023a647d07e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "659be018d6b862200d5e28690b5e480a771bbce9db2ea76091485ce0834c1ff3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "4ada3240bfc088371a507c1bf01fc2abed25d13742fb5f6cdaf3be196a6a6aff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "10c231c85c77a107c83b185cd28f0fb62e4059939e8476a8b5fe5e4db4e91da3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "18151a567dcf404056afb9a0938628c679625b40c1970be647eb83291f03de39",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ddef8783ae4a46af5aa91962b49a6da5dd2497ea4bdc08ba23abe9f3fd7dee3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "4c183bf8112b5ef007742bdc7674d930929ede206992c63054c6c0f55b90fbec",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "09dfacbffd093d067f4e18c0840a03af8367077998cd82572d47b5aa0cd0ad99",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "9344e84316f8041eaf6fc871b7da17fdf302f6e274cd20e4ad674fdc1b8d2a8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "2ebd05fd3f260a81398a066c2367523c406a1d1b136a521a09bc82fa850f9aaa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e3028c6b6ee5d37fc06bbe6a264c0ada928cd120e4ce7dc4a29ef8aa8d50d0a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "75429048f9687179e7df610b90ed119d548521470dcc5420a99b6313986b4a90",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "5733e7a9fa7de12d42cb1e97d569a2d29187a17c267fbf3f963720925126683f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "377ead560661dece10c433dc5fb11fb0a3daf526ba4db287548d94c56e04023e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "2339f31ef72ae001d61f130c678e0dbf616df9ea0cb5657e658a020d4498d683",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "01e01d58b0b5f527d305d9f06d41d2e357c83818a681d5d0cad57ba79116951f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "0d74448adabe67b5d320f99a4733624105cc8d248c96c48f54bee445d66de955",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "2508abd952946a18e17563e32e5c06e5a2fd37df262c820b37b9b93b1b3323fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "74c21ab218828f677dc9869ebed24c3610c73fb8a81078ade433b9675808e101",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d9e2e5f97d230c35a73769f13ff49d955d0319a62045107cb04cb2f0f1857212",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "8a0d616839dc006d59fa273090f9704de18edc87a689de88202ce29963e92214",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "e841a555f42a64883314c56ce3e799a6a50c19a1a5d398fbf918f9a5cbbc3cdc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "43e8c319e78d95486c021bc38852f34c053d03830657c0a9095617e58923df8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "f94df156e51c7abe12d6d86005e8cecbe230ee67a19717c2ac83e57d4ba4ec8c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "e26dde51de7ab155f1235ea93a99b5d701aba3c8b1314a38ddcb58316f56906e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "a1d24007d92ffebec9fe95f19cba850de54ba925cd8e621ed4db0c7853f661c4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cbm/claude-fable-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "f8be201a52d47e0b97144b04317889c3e8a07b338aee526783ff65ac222dfe96",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "9d88f1bd9d62d7a5a652e1320b0ac3e6f256c6570f9211dcf144922cf43223a0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "83916150ef8e4f62f9f5b1269e604ef53c61a491d178dcf3fe3e9964ca20debf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "088e3146bdcfc3460bc21e01d701b7a6be22acd5ba44ea3a1c7581816f4f8b0a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e0ae007fdcabb325331ae4738503ff3cc86b307ee42cf7eb007551f14fb1dd88",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "04cb007072ce59bb2052e53e3dc5b41ae3ece57fec3cf8398dda2d43e47e97f0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "040e03f3ed3b0d354de4082bb7b6a146047771be923a16308fc9ae2840e073ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a1c889257c496fba0a8009c1ccc0ab8e265df28d5b6e197cba464c7bab8494c0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "87b818d74b7d9099c0b3196fd6c491ae0a98141fb11d8cf7181a34e365058348",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "ee1c3d074ca1107377f3c21cdcb464faab8300709b8abaf664c01c8a5baafaed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e03b18c00d5e95ab5341f14cbe4f963c5d6980f1c94727ce04b66fe53c20ea13",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "d93febb80af43f1df421ef72006fef6ffb362de2d41d5ccf1d0c1de9b0158108",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "37987e5068899fa2fc5c01941aae8c10526d27028d0c695751b0a35c0fa563a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "517086bc854f359532fdf0950153d95da98b253c04071abaebd6fba9125afea7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "f4320f91577bb785c14cabc1536aad6b93e0b96a4e13b4ac2f441469acde61f3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "ed38d070b821807e29ba15786b76d3bd78b3525d180085ccfa1f71f4eebf4f7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "be88a3f98da58f5fe47fae46f21ae09688c3a2b39da4bc84a1919be4f7f7e08b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "72f842f33c99cf01df6337aeb25498e9f75b92e1ef3440b5fa00dd60661b7d45",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "4f1e32650347ca3d9c5cff3575fe49dbd9959f118b24648301e7e83653ade8ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "bf11657a376f3e08696feb605f3b28aed9d8b3197b2750cf95fe3d5d8e7fe123",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "e6198687f51769d5ba3500ba3354c21f2753e33186ee1480f320edddbf2b246b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "cc1544836bcaf9c290b5cdde16191d9347b50a2467b9f5600d3238a4b27e1cf1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "90dc64586fec98c5e46a8297f053e5348afa35e489fc914b7311267b7788ed08",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "88007cda38b4c85f083e855eeb0f497296aa4da7cff9702681dede6154a0d1f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "cc85af974631eca2934a6dd981047234ff5fb6fe1800fa3123a0cb23fedb3480",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "bb3ce028eb35e9fa94cebb54e5ac291631c8df158d4403268988217882cbfc19",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "7de092a3ce6d1bfdc3bf782f41878c2c6ee93e2ad09340d88ea36be0305e84c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "87e77b4d169d461e6f59da9fd7fd7ad5c97fd83350cecaa77ce8eff29633e447",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "159ca574fe0cb5500d45015fd71c668068199474925b2f73f111d9774593f8a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "76f65df5e96eae70b5c46802a36d8ca1f6ed0f3aa8206222c46d79d3f5f6aed2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "1279f173ea948285ca6d56cfa4a3be29aa01c593942974052d6be5205b2a2fae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "7b38ba1240518f1ed9ed26935ddaf39db520e2be2b8add6028cd078406d78165",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "551bcc5f39c59988601834d2dc71ff3a0d99d20846fdf0830f083f4110b02efb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "08f1ab67f0a6f94df30541c0c3ffc765c6bacd299a853d06ddec8a990fa75d25",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "61458341eba9de741f91e96257f5c59e04aba9c1a270c89dff698f07b8724111",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "3e5d39ddbd18830935931d98d0e3ab4868f58d5028f1471ba1da90fc4a75baf2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cbm/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "0d63ce44e69b3681553daaa800e6286693d0feb7254d55dfd396f08524db3b46",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "297f1846317233f47e735b12088c26f70c0af2747468405754d24bebd8956524",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "6764bee154b6caf57e9fa3246f650bbb65fd938ecfb7f8937d0e1dcd9820d58f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "51a75305eb11ff62bb7786f7f1a3daeee9f8e125a5bf59072c22dbd34bab4689",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "8b7cf3ce08c7edc539929a0336f673f8755b4818d0cbfe15db8273c0312b0026",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "2ce5b3daf62a7306a2a79aeaa1dec8bd62f166c191b97f5685bf2d18f04db95e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "90a7f32116ca89a4be219ee56f2c14e761aa94b6a970be4b79d64b71f64186ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "71711910271f19900bddd2a093adad30efa9b1594236e05a5675435c94525902",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "74f5f1f7177e0f3fea4f52a3297d541c734979f70cf153c5492dfe93d606f4a7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "73ec79e1bc1835727e078d0bd425c0cdaa5de2d89fbd53f806dedf629cd7676c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "adc2d813eecf3dbc9f36b05a4ffa73aaab32b8321e051a4e681c5f5b08c52023",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "76ffc03341a6c42f87ae0ca8b518db7f3e976176bcf8a9fb8873a64c466da17d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "0099da725f2eb569ca6a5b24251bd701df3d46259bd1f016a46e8e46d486ee6d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d5768eeb55f18ec7657c19d8cf49cff02b73fc8d011b96ea51f95296ff2d5a0f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "1f9304a35b8c13a5e3de9e1b3c8d59abb99b40bdc2954b506e94225e0b64b72d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "fd10f0012dbe972aeded04b6c66907039ee8c30005bc3e6b353df471237627a7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "51c8fc2c60c500141178d23d01dc1aaa7db32d86e4bc4a1b573953ab7cdf06f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "dbcfb5e7d47fd53db5d6e55f4f26915cf5b32146319bedcd34246e74f33388bb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "9cdaf874661e4f2e98594d929eae6d7cab85a3792e101589ae7cd86f6e13c009",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "8890d25f76deb081aa36c81927f44866891d9842c8975ed2ea33b793154d130f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "71ac0a7dea1ddf4a1bd1381d06645e74245af238d5858feb1e78bba32be63bd2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "12792055a7dea28945ec1f2e6121c37c8847e3364f80333707e2c53e3d45f68b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "3e0b4f996791536806260d2e508bd3e8d388d308dce21506a3d8b274197a4795",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "7704cd2b4f44c85e7a74545a07d88a524f86a9e508c6612db9ccbea3e11872af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "12304e4164c886e1875e0a8778edea072d7b758e42e5e7c3840cd2f39f62834f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "afcb00ae495df640a53956cac5dc6ad9d4aa056555927d139f34a698f82123c1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ef6dbc99cf0a4fb7edd07a8b0223ebbe0168f73324b33a21271b51cfdac572c8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "a6ed377addd738d3210d40cb16007f663986fa6a2f53aa5888d99fffcbd503d6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "7e44d2f7e2c1a053ab192129cc844d1c30c0b25bd6a4a75aa07558fe9d72d8db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "f304364cc1cd81b39b6055f9947798ec689d47b55f1998665da97305d3c0339a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "4f735eb7654e944975945e76436f29916ee831b32f916b6bfaae4b39c776d813",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "ca1d043773cc60484b8910a65f14185176732a4c124ab1049cac3ffbafc7456d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "656fb9fad9ee23f7205c7787cef2c5c1ee764ea64575edf0907d5eb71ebeb3c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "5b8bb2e87cfb61b351b438f47dc7a1fd61be42700d18a6a756c761582c3f777d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "b4ee519c08ea5a1959d282bd2429c61ec4b61d66cf1541ba74603e68507ae19d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "bddb4cbc39ed92bdfec21304bfbdeea8242405caff928458ee35dd3f26e60e56",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "c5122e4645f4a844a6e7c203382a6ef6a2da53edcd41ebc3f59875ece34d7a0a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "0d13086d2d62bcd78fc8c319c6eac63f7e329307d022549a5a2725462a1a2f08",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "ecd3530a76a347ca4f35062ccc36379ff4ebdb516e8b0c5b323e4de9bc182f07",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "904eea33f00cf00cd40421ff5c6a4b0a8b3d759e9cbb8b1fe85632a6627aedc7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "2e96d3a1ff13677636f62c3b1954ac9b9c5c7dd9c4548a76a524972f4c631d35",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "2c71ee290b119618b02ccece88fb7d0a6f07fea581e24f96975181ee90aa038a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "1c025f23dc05bc33176f7eda14e881bdc225e39fbebd85b61f4fb2a395065d57",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "6abe38a343611849a181660de47c1bce73d8ec369a144866a98454a89ddff501",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "71f8aaa140e7d7baf3a0841dca48be09e173c417c91880763e0ef63950800f6d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7cc6b581e3460a182155698d53d0468a1ca1deb8d3542bbb54839e3bcd7f4991",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "d0592b2fe6bc6ed62dc5c30c94ace477d474e35054ebddd39fdbbf6e9e277009",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "657d62d0940dbcedb3ad08081ef14248d86dbafed84b371ca61310385d43ac77",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "853381e5b3ae02b038afffd07de9f1af51ce86425e825c3d278b913be7535b6e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "3f467d7b07cad50c0b7c141896172909eb7b9be0226f115bfd304f99d1d596a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "6a37f80fb13ae39676a15303ab51b15de3c7ceea9a671928bcf89ca24693f992",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "4298d772f81a874eca1d2fa005bc8912dfb7c7a404d9005f827a1d54696fcf55",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "87f7505adc88e3a997f445ba74e2bf4df45f7d6f3933daf084a7304590ecd76b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "ae4294bdb5b2c11fd15f43ac069d83148eaa6569393648b04085af3883cbe19b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "d1f59b36887fae80df15f95a69ea661bfc76855d509525672de5ef01b3f45a48",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "2e47fd822da9c7a4d0f2d57f699cf3d66409cada1f26c74d25ee4b1ef14a311d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "3f8c32d136b11dc4f29f72c3f4fcf3907c8a51ca77a600be8d08c364ecce27b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "ab17fb4d4ddcdbd4b5ae10fcea9687c28cb595d2a7f417036c6f5afd61bb72ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "447be00fd00dca7adc7ee6f2735a3f165d831f5cd0b589a0a8d46a1b8b4ac845",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "1b803fe0f38aa38f1ea6c897c8c511b585422e0fe353fa9cafdedfe3f0ba0f99",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cbm/claude-opus-4-6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c2bb92286d18300c16efe2a00061d1b1e006fe6addd6fabc64654da4adf1e584",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "a3a2a173e9005480353e5b6fa5494f843bedd9e1a6b995c18f67733891661653",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "dc75493f9b005d6a226441b43ac44e46cd3297f0b8cf2b88498bf954b497ee8c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "62a04e989715e1de293fcc10b6786c944588c8e03bd0fbd927f99b006e8d3c41",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "388f06ad72a5d585d83a43da5d26f888a1fc18308eb5139e09f39a786e54cd90",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "a23ca44c1ec00c278fd1c9018be34222758310278cb6fafecb6c030e7c7ab338",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "2544d426953a6b86d34074b8aa882894c3ebc0c5dad2adf79f291c39fd7f1c20",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c0bd582d212b3d34b220175d343a7b652e17e75ba68b84e3387596df2dbb0192",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "202acff12f3a026c98abfb0a18f10992bab410280b3e928ff996389e9bba0cbf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "c5f6ec43c8c477769aef7dfb68006c6c4703a2fd39c9c9cef114c9776c11c1eb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "f87503071b62bd89ba6c3e8451528e779b79b8d642ffd0e6b4beb60a0bb88ad7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "d2eb1d0de869111f80f61510b5d1f677f269703f892e7c50da5c078b5b5dfadc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a01a224dd217cd4a010a5cc8c47bf4eb24287f732c6c9924772341fdec848980",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "2bae5f21efda04f9f93f8e2b9a3d9cf4dcf3111af5c13616353ab45bf5f6481c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "1d1a2df7ae30306f6981105cb3cf00e4cf3bad762b53a2f9f5aa420d550215cf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "f0339791add0be42a06a3f42328987d7b7e022ba335103020251d74d005e1690",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "daf5b63558109011edfd76cc04dd767d2f841c4cc273841434d2453c4be1a73a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "876b48d7be56ed0cac9cbc5070aface389337945292813aa92f07d479cb4f404",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "37ae13e92b94c3dce3aeb919e3735f717539d69659cfddc5d5ff350dcc7ab721",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "3347c95c8930891bfcff89eb9fbdb8a653c4b76b8faa4e64ef7e93984cea8e1e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "0b9b6d4a9fc62819d3d74489e6eccecfb16b416294b311d85b1f73dfc6bf0b23",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "528d2c13731c7d2cf5756e234f732e796b170a49a943b07100ef70be314ba728",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "795cc3f07b562956491afdd886d3430de2c45162856bfbf8c161fa1dca86d938",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "53a36603c4fdef857dc3c27554cccb2f58cf6b5d480be7e7d0c2bd679d9e63f3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "3a6da9e2db7ea14820d21ac5ddad2745a8abd5592444ac3eee40d2d401afce03",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "23940b864be34384b8a1fa59addacbdadf86f1b6942fb68ebed63a7115f07ae0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "c2389426dbfc760f91f5b668d75f231f4ffd0dcff0c7431bd713b4ee457dcbc2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "cc3c7cadc1f5f76c1d9aa46495b0e9c1681d1c3be45bb464da71ade13bf9c3b7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3b16f6e433e0296d10155401b58f297661537937b7a8e21dc800827bf1f25dcc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "73e16f13a0792b344f763c97599ea4f4c24d8b303cc4b9cd316a4a2c77bd2667",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "8bc2d532087cbe7b74e04b28ce0a8c76ec0ef29bf522948ea1a0892fb7552c2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "97ab4f72bb6f1329adc0eb36c2064fdf591b05cd32c6eacdf7332d87baf7d007",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "cc6988b1577ad8ae18f7467873961f7848597ffb3cd02d1e4dab70f1053aa301",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "cf3c4d99f5b3ed712babeaa03b8712fbc22033bcf6585f38f0e50b58e9914b2d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "0f616eaa6f8b007d0993ca0cf4092fc7a995126be12f41f1c434478a85a7d995",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep15",
      "rep": 15,
      "run_json_sha256": "b5f7a269d99964b54301764eecabd3075d8ec1bd3c388d41d2312c7db12b42bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep16",
      "rep": 16,
      "run_json_sha256": "0c154687aadfbea780a6fa9faccb8535e9b18c2e83b4e281352e31dd128343e9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|16",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep17",
      "rep": 17,
      "run_json_sha256": "2d1bd7859eab2e2f6e9ca6f6a95b96914b537f9df57148e51400f5cdef50f2f0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|17",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep18",
      "rep": 18,
      "run_json_sha256": "276952c5b36aacf911cbcd43971ccb017b3920080d960c6c046898c4f852dcde",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|18",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep19",
      "rep": 19,
      "run_json_sha256": "6b3eb0531b6608b9157e855499f8c355c0fd04f6179d87b7c6cd79ac2e805cfe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|19",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos-hard/rep20",
      "rep": 20,
      "run_json_sha256": "120ec31edd6b6033709c4a1b19cb28f7f84ea4152873c878c5e448e3605ea48e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|20",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "69b1fc37bdf9ac14f0c953617bd47442051c48b9fd78603749a9557c738298df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "200161be8b3a271a6162630d78c1f75ff9b03a366181ab50661b17864f239ece",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "fb014dba588f9b5f376ba4e2243edf3aa3de6118acbecc281d558d518e749f6b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "14466ea5405972abfc19b283a7265f794770c1a623a41a8f2dd51a97e76c26be",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "7f28a1d114ce20feed5a09b7b83f82693aedb1a2d1198b8f9c422dec213be4ef",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "bdd4a6e90f4ffb0a9c4fda3a8d07fbb8bc8f98a8bb0d98ea39c91cf4780fb515",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "36ae778e9fda3f51b315b40dac621e5c0a1d88007ff10916d583e5ea74a19c53",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "c173e62d0945a0e9c339786065788ad63b26176c093d8b5b476708b9cc3d8076",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "78a8a22619751b7f8d3eb6c210d647ded5e3f6917fc67f6c1c3ee7d70725f221",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cbm/claude-opus-4-8/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "b41e75480d06510ff5536839b4cbdc42cee7eedbd68c492abdb4d5188270ea54",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "b159e5834e96a038b64b408a5564e808edc0ce640c37d8a60e16aeac8efbb427",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "d162c74a390360f02685cf5ea19c645db24bf237990bd4a6036e4fef6085dd00",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b83a76c64cfe6fb89fa0c19a2a9f5654bd1b3636e7bbd6953eae941fe813a694",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ecc3b0ef3c8fcaccb05374e8603d813de0848ce6dbffc708e416298bf32d6b65",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "e5eab64863788214a8e32528025e27b97aa1443eaba4999c8cdd2a66025f326f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ea2975f0dad504aec9c2e447ac9b47e24db33a7414dfb1e4ee84a8248a315c0e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "00f2055c89c67bb8cf0a942336ba9b060a22a44c2051f309fa8689353ca3ffa1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "395918392d2cc6efe0acae6c99c38b2b6aaba6cb8d18ae380943af9cf92852ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "61ac2c4245d8ab6eba60208f7a7f28ad3b2a39dede9a5b4cb0d7382e0474c061",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "3aeb742ed426513b7bec2f6ce19aac430cd9a419c41dfe5b09883b7ae92104f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "1ffab36d1ef5d5969fee1d32abf01dd693f489437e0b2ea9d060a4b00231f524",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "21884ec25533c0672fe6a335c8627c57aec1f021dfb122c0b58c947d731c71fb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "398da0e51c7fc6283401f63900c705bdf6e12140a955515c46c28ae69c608036",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "eff07808cf1a364f1a2531e35a5127abc40423e62c48e8e68a4f91c8f5733de2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "92865557f41e05d8c196877669dbe42dcb124168cd07549f2d5f4c807a92e5d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "9660dfdb9ad830c44b98187a2c871fbacc017f5d5a4c2601f9e6e6fd6b604cd2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "94160d8db1fc05db98cd9cca516aba0deefccf6fb555e387c06f82a9da7f3aa9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "c2843f8566bbede7852c00f626b3e4c4ce69a2586815df4b1c07a07bbd871824",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "ca3d4369d1aa5fe1b24474bb07e5ca7ce740c800b45b49b3636abdd8218fb17e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "3efae4a11b2307012c995d01d4dfd5d45845d2656ba2e139db4ea15effb8bde5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "a6e7ac300e83e33313676f4fc6cb2f2653c5a75d070b5067586dfc68116f610a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "4bd7d8fed984c2c1e4607a6d0841af5a6230cef9c512a2668f066458fd38e5d1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "d29ddfff2b43eb82f8fbbf5e5f53d9e0aa31e251c6d7da17682334bf65949e04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "4ccdca5cb750fce179d44352a38af153f2a542f46b5916adc7f6638d505c6681",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "ade0a021c1b99bacae6a28b900ab1723b3271b5c1cfdb960a4c9f46673a2bf8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "4f0e9de9a683e596b1c9eb3bb5b54368e150d832f0d257a4b7f28cc650866bee",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2992737e17aa7fe010b60894840cc485466963adee5957525abe89d9e5d5db28",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "dae1f95d73c22ff44e194d0c5ba87a7c18ede3b4fd5cd6acb02e01f1d2b59e06",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "57404363c144812e2450d8c216e7e6e3cc44e659984c878045952faf92c02f64",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "99a069f731e22c01aa0e60089144809ca603bc9fee40215fe830341e312129e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "f2715cdf43a9320afacb75aa3d41df862c57300ebfb75539e3acd786ef52c8ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "46299986b5936ed3247de0ada440413f971f0b5b3819b481de2079872eed7381",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "1f011588c52f56434c9f3fba2deb3fb89646151e6d0b128f77a0f1952d5f1fac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "68334d655b0fd98d0550f6d51fcb8ae2bfc1aa779454279c7e01c7cf7f179096",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "4a645a259e54886efa0bb4675fcda536d77f93151d510a52f9ed5b3684a18dac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "1246c59418b7536068dc9425dd5158cbe815cd9c73225c26220b419c30bb0430",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "219cca2b2db783146188e7a93268b695d914275f74f90501452a8310978de59b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "f42aa02e9d09ff6014da038b78968f7c50bc5074d6b6281612524fb81166e7d0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "02b92fc8254a3ef8c04e5335adf36402cac0559790103b31f87ea542c14aaf5e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "da6ba45edffb1056db9999875d5ba8e949b4d7d9455facd0506237cbe9571f5e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "546e6af23b0aec024baa2358106720b58cee45e5b5220dabec601fcc37991af8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b6ffe73d482da5fe4fa75100d49ef1c3d9bd4c5c2b8fb7da984b30f91ee22a40",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ce9e519b56b3ddcb7519e5994df825338f3d660e4ac3bb8442a80d22da12e960",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "b8220514dec924765cf85ebc6820debebc1f915661fb19a7ed8f1c3b685ea38e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "f70ea5546aa31339511d677a2fb0ea5a75902091da8e9ca671a7eed585b64bcc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "eea652dfacfc87c490ceb44b82ecb086685175870d1513ba405b284ff1d69c9c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "1e6d27cac317d21f588a065882994983b996f58a966d52e3b46bb06fa0f44894",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "96ca9f935025b84dace0b418e263cc6758ef95ca34c482407e223510f8e14c62",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "bcb0aad4c41cc0665d9497705385a21c08757d127afa4c72181484d1f8573417",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "2455d46b997a55ef1a246f5021b5aa4058bd5f13ffa5ef75f8746aee00988e44",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "88396328835f160caa702f65c2a83b4d19435f9e99bd1c0e11f62d07159f2aa8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "d12d3cc9acfce437882cbae20db6b9efc7e8ff236a07fb11fd6e4d6d015f21ef",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "663b76a625aafaf544bc24ff3a0fb94d0a01f57c62889d9449c1628b3cb8554b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "ad251f9f3ce6f88885994ca3f94478a59c36c6ac979d1fef246b0ba3ddfb3d79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "282683c0b6e0cd9c8df56a132a4b87c37b512341dfc880e8707b72704b27f133",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "33ecaad089330f9f14cc44c26efb7cc758075be2841fa585afdfe60e4159020b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "9d0722995b048e119e717f468dd39c8a1451bbc78542f092602f8e8b74abe293",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "49a4241a044497796285ed4762089283a07a8d9cd00421fce03c2933773f16e1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "853f6dcabb6f883ec9c70fa5e6c705b7fb58a167cf76e1f40df3e74ee1be0733",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cbm/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "d5ac6edf7cb1eee955e1d80adc1b20ceaf7bdb0cf283e231fb8922bb11ef671f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "727ae6dc81f2a38601f25225c006b404ee3e26effab114e4b830fbfde21985bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b977ac9e45c323bdc57012754e830a5333be6d5f6e82219626ae894ae330e820",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "d90ca17f074455d0cf4fb60ca0879adeb4ad164d976bc19a65b12825d03f3679",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "9ec07e6dcdeca9564f16ff245aca3dab42b66fced2d9be17b8a796346f794b27",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c14aa0daea5a20f6fac084993ad5c8c70d07db194dd546132efdfa37b65d0632",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "0ea523451c3d91e97abfbf4dfb6952576c7d9110d45847e0173067da75c9d80a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "f75ff449e13eb1dd9eedb9dc3b30b681aba0bfcc5d02b366c327610de6304022",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "8ba377c781150f652ce96bbbf59869dbcb13db78120cafdb0a9ca8a4c846b333",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "4b2df40e297ef9634a4b44be210acedcbd35178c1da6b0167ea91a87daae8a49",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "1cd005a03022272a4c6b604cf0702a6995495fd439dd2a1404f5ab89f66b8ed2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "60ead3f38ab9bd02c37b71e36d8732fb350a7146d877ece6aeb28d139c99abd9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "0cdacad8badb2381bb7146107e24d339f643b0e4bf11b8f75c4c8d1bb7bde8ff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "8eace2ef560bf65e7d3caa4ee9857c9ad5459567a5fb67d9998ad20dcb3da5e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "4335fccf4241296aa8c50854999e615a0092cf6e14c4bd9f6e0d271f8c5fe0d2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "cd31de4bc4f05ce01aff1ccdeece9c8831858707985ec3b24bbd388180d497fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "a3d573d0cb0bf4c68b906b6b1d09571e6f80b4d97c726972c97deed6b6a1ba75",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "c9f661b5aa1ca925b93fac41d2cdd74c4284563849c6d0db8208c4243346ef09",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "951de2897816ffb00df80a01438f26c5205f34cac5971284a11b1a7aa4bc954d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "affdc4fd71a8f675d96d0cf0c86210ebb6c27627ebe0397af8572e2ef2822d85",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a2166901b95007d837d0a3efe20fcfe910154a50296e8fe2d96dd693c28fe532",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "197b3bcc00ab7ede5a7f76e881b334aa9d26faff2a9014a05a6139bf9590eec3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "85b9b6c507816e63cf212b6307bacdbdbdc0f87905998d16bc905082fde66fcd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "5d48e24bef6dab1202271af28a3eb626c4c26be75d5f72d7010d31a08479c72e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "63ea553521a1f91b2277699a28ec6e46b30e2f77079fa784266d3c5bb3e5437b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "aa421b697f6cb65f297db3d050f136b29a4ff7feac22eaa7a276afa5c24f2738",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "936ed937add3a3636eef41f9a5ba961477c8991083fa05edb068648a44dcad82",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "6534edefcb76b7a62f86b72f0797e70c0b772163af5c2af12613fd7cab9a08c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3a14f84277e977a688c71345f150d27a485dcbc88ec46931c91267bbc8b102ff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "9e58eb8f6eea53a5a82e46658b7bb0d76c2916d64052a7e5235b43574f1bfde2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "c5a65247ba37222337b5504805ea28b9f4630086b2b285ac833576227e3464d5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "75ebbc37a05b8094d9b7224de7a7ee4b9b4ed3ef4b9ccf3fd2289c62d7db7cc8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "91e2b4a8651bc682be3fecb537bad6cbfd16a767d159efc8b4a85b48c6ff8e89",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d2ff56e7b8f49e5d0b7078dab93b44db3d3eee756e7fa62717f88212c49bc3f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "fbb668099052a51ac9b7ea4411bd9861e10fb5d1526791d458d8437758cde7aa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "a3796ee714e9ce7af02d71191f2906c9de148387c4269c496abf9aca8e8edde4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "816f24bb9cfac5ddbc960caf719f21891f24a7b487d7353cee357a63e107718c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "66103a69dde854ef50cb55fbca87311097c40ef34aaae22995880702243c1126",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "19ea22a515ceee4f7a0959b4252066d600204ec5fe7b4ee73040d9e045641050",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "e65693ec6efd65837b454b74f624a2c0cf5747b830730e923171a2034ed4c467",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cbm",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cbm/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "8baa3e0cac0839ba9309aaedff48a5d8af8c0cc7a45a04dcda35343b3009bc5f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cbm|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "548c3eedd2d855dd27752d4d4ec0d67bbca1c4bf1ac877ae4533fee3f775ede1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "950d9fb28eeda1756a3a3bc1f4d7a2a5061544273dfbe82acd566d2ec1cbf3f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "dacf4d8d02ee9fbf57d2bf23d9b93672cae9128ea617d18c2c28280d7598661b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "c658ecd63926c27b68de075604914e5d8d47e9705e3c7aadd29ef7c039ff5cca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "d43749ed253c5e031f2547482010343d164c6cf08d03fc0bf01d9f10018d70bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "934b64358f3da1f4624c5ee9f9ece5436f35cbfefcc9de9eb4f67339dca614e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "7897da61862ad698a00a1230bb08299101bacee76833b75268dc5e93c3d5ec81",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b5af2e0151a9f5600e90edf7a6559a40c4d587f1e0126a02622ad152f2bf0cc8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "56515b6efc826507a3359ec94b2fb6073804c20e51ef3c472b61d63854a716ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "6722126ed66668dd44885dae2f3bf9239bf022832950878057bfc72a169d3fc9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "9fb97ae1ddf5783da32d09789c8a8a62c5ae2d3e9ed6e1eefff4ac5ab6b2d5f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "d5325ba6de79ba64beaa56f8ca564a33567e04e42f7f3ec5b4262b6d56f7b8f6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "4d1dea11bab850ee412056c3bc47425d023e6b4f81114002c4fc590cad4c9f56",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ed711b4f858cf5ca161dfa2b5f5e323890d250229dfef8f807486cb2be323d83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "cb2e50f66e482c0e29de518736e8baa3fb6705efb17bded3851b2776882ba9af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "4baad3c5c5eb7d3e69b95644981e52dc4059f47ae913f8306c571991c16e5606",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "c92c564df93e7f2573bba3c972b73713809db62a60d2b6ff7b158f6d836c01b8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "3e5dc836af0381d8b40d04a1abd3d5a9ebd047d22a50427dc23ac05eaeee79f9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "d27b64ac1909869105af71c8eaead5ce56b396fe22f92f5ab1721d23bfa39c51",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "6803f1beee5b9fb1f9355c3681f1885561420259e2c93db2bdc4b01b7fb47237",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "413128ff42ce8804129d4634c47fe00352cc4dcdfcb2f6215c7164a6426fc334",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "9498119293831354fb9842aaef83e8607429daf1685b3c33690dc32b68a0991e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cg-ask-fz-009/claude-fable-5/ZeroShotExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "38a26b671f606edb304d3303346c61b981abe7eb48dea93c83b98991e5226c1a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "23d05f789283e6429f46c778b8c765495c015102de8676b0778359f28f6a65c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f8de17b4387ac272beea9bf38738e9fab91428fa05e144a13d347421f0b3ffc7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "5111e746a6448fa6cc87af57f4fa1dc362bff79230adc0f34d748cabe1d6026a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d1dc5119d6509588b5353069141d953a7bbedfd64b7de82aa471ad3fbd453e91",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "ff9ec5e6ea3a9b0c71b315506e8de2be243f41b624063ca809c3795e9e5b6ddf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "23c7e0eb1f8c58bb7ffbdecc6231549b4c8b8f21c2bc01654004b8b56dcd8718",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d0d7dcff8d8cdc2256dcb986f8dec7826841dcaf7b93e8c81a6a6f8fe365c792",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "6e2fa0e183adacfe5b5b67f94047e0392199313cee1e2448aa83790a563f3f50",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "5f7330ced932148201b74c9f11d03d75c955e13a9ca24448bc2c2f17e401b210",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "faa5b7fa34e3250709644f68f68e85021b13438e256bc11b77a1ccaef19a40fe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "cd65c5be94056c9e02d47ac352926499fab8dc46b0ae3ca46158faead752d65f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "ab3e709a25714b97355c0c75f6739e8bfae34f7fbac8b319445e14a1345fdfa2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "3f1468596eff11e0245156b5ba7b7143227f0ffd9d25bcdd829bd5c315403e82",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "467d2bac38c12e83a18a8fb5354d2bad372775c8976d99f2909b91c12af046ee",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "573d3c0ad26c29bd29c42d03a0e2eaa3b379988b83f5ccf9c57a058c6ccbb368",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "0c3aa24739f2157b3f57935d4d18eb02dbf9c72417de8093c4896172c381ea3d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "2e8b4fb2d89dcb8d9f5dbe3b925d39819082c3a50c9e9cee1959551e548ccf5d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "2da654c225522d733525fb23946e169c8272650f9461bf6048828bccef045d04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "b55674931222d7dc64eedc33dcb8762f06fe57cedf610669284b2e5bc2436455",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "a1ee91e7870a4e20bdea980cd43d410334c926b002caee590b8d348d7e332e5c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "ffb49170791337de3ab78e31abcdae63032843071e0a4c2c7ec8a668c10e418e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "52b7db75d98e522b1a21d078bd5aee91dd60dd3c77ec531a0c50a800ea5e50fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep12",
      "rep": 12,
      "run_json_sha256": "1bd1ea3ea8794850f8e9ade3841f909c1de2cafc5dc9933b2f20cdd80c7708f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|12",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep13",
      "rep": 13,
      "run_json_sha256": "034f292df51126667bbb3f427bfa922fd73247e0814bb09034b14813ed6eef5a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|13",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/HumanExam/memos/rep14",
      "rep": 14,
      "run_json_sha256": "5494f66d4ac1236a0ae8fb1a02f952d8223f35bb2d37ad1a2bfce316960ac255",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|14",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "fed3f7b27bff5eae771f84f8e3666ebf743ecad0b2fab1dcf73c02715dd2e67a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "0f4121fa974c09b34555822d121d985599903bc9291dad989ed7b14110232d90",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "b7d59640b7035e92f745442cf78fbfb631807266849e2a09f10e028da13bece1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "81cd951bfd77ae71ff026ad1dfc273b2e0656f3e5eb61ceaa8c2d149841de74e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "fd2766b9620df273158d9a5380bd57595612c934baca0b05cddb7e2236db3fa1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "67a1c3587176cf98200528b90c1fa0da45b5ed14e935aa726a8450e78881ff56",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "1f1d088e8bc9c05b11e0f653782c38cb154b57b6559892921d8d09007b2d1996",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "26112a7b3d019ae70cb3f82708449731b8e7c27da8f34e4268d1f875f8b8566d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "7432a9940702637a60d0c3d8cf68b5431540d9db4914360eb5bb5f71c479fb2c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "2378a6a0df690134abcfdb772f1a8d164ad043ff88831f79753079ce70795308",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "ccc067350059b46af952d14a1623ef50b962230f02b8c5b70f03729fc34b8892",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ea469dd56566631120952fed513a8dd80c3d4d802fcf860c9aa2c594042d54df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "2464e4fa938645957fb019995b1b0db9ea369c56153f90838e5f29bfd4d66cde",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "45fae373a99375f32315265e7c5e27785823afb2f9387e3e2af40cc18051b292",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "5dc57f411289fc3077af9f61d1b64087c863ed624788ab77f5dc0d51f8b46638",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "0946db13cb9276ec1c0105424ec6d443e4484a3c4a970e365f5b3823cfbc2a2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "dfa86d1619a3696698b8eee844621dca8c92a30b7b85e08fc95c34a7b54933ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "e57ddf846553c6f1aae2340096d15eb2d363ed30ae3df1975b226f64107ed74d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "490256d34c6df2f947aa5eb8c8c275bc75bf427223a204400cbd7bce6aa144d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "d553ca5ba449c374e6e1d5f6a60aeb27349823b7814e7ee2485cc9ce57521d4a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "9850a51fb9f43b9e9306d8d687f332d6a814fced0bf94f776ad9949075a7e19f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "a3516fe2de173d62a161ab6b127f008c388ab7b22ca0a0c2a724cddb5c8d83f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "2ec135800fa23fa4d66326eb836c6bdf5ffbc5530e14df3d495e7e8fc3ea657f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "15a8627fb349c2ec1722e7aae86d0d244c41ed20eb1bb1d4f1d66b1f0ef43b32",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c8e4cc3fa91b6168c9f3f95b3025fdc1520662aca308c2afff4d2bbeb5a364dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "c137ff3fda3dba24aeda1c5aa08f06a8d476a2f3eb7f47bb26b72503d52949b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "90d4c7578e96477ace8465368bde926b10516bf45a2af365cf8ff8da2fa28b01",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "d213d9c3d84e387318230f3ff7487d220bf0d8bb39e84e8feff5bcadd1b46c56",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "c6fc8caf49978ec4d0d0c6386af2b949b3121ee9fe14a3085512e5f982f52481",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "16c3a909065d5d97383303a8368c4d612fc24442b161970468bcfe035cb38e07",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "d56d2d7b2a33ecc00fcc6b3f141465efefa0a021106294a30fb34255eb617b8b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "d6988af6a031ba510bd1907129e9e338f9f8364f839283f27301346db0502421",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "6641980bf49faf37e037e31581c039aa828ed4ca07432a38f8f5597d6d90fc72",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "da806a506243315cc652723a9a78d624644a04229dfc37230db2b2b2a719f08c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "9a6604ad8c2324b30a754818425e42ecc09ae6e6433d281793b63635f8bd09b7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "af8a968dc1321a6e04cd88099a61deca9d3434ff9e46dab6ee990aa961c10a33",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "e027ce1a9f1a3cd143b4e11c07890fc1b7a0aa49b01ea88e03f75bb832886820",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "ba37a20bf5b96f347cc047b8bf0f666a09abb4b1ed10c282abd993aadb759cf7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "5b431f34018ea819315bb3a02e05254e595c8762a72aa8a37b26b6b6e25e8dbb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cg-ask-fz-009/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c1e446125d6511fb8bb49a1cc52647b03544ad01afa2156b49edce98d43db801",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "10694465bc7fb19a3c220274eb86426522ea7eb195609bcd110fcc8cfdd4126f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "cde82e40a2138d3a75a506c4afe6fc18882dd12c7deccd5628e6d7c0d54163b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "43d0ec0b01b7271a0c8a790dc21e24e6768ccbb72848c6fca0f7aaa2078a4d4d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "b9a3233fe80a03cd3917d562859f15e6022b758fbbea1f3d7d6fefcd86af6b07",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "76d90533f9257c70c260bfb51f1b948dc16666d143e49978e95e083c7aa8fc27",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "f8426ebf00e60e40d99c4c3e2aac4967411a44de5c0ef0be0e445cd3ba0f7511",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "f939c994c7ee863005854383990f243d9ddbb791161847122c4b2b4c0bfa0ffd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "286b34efa794b70e869574eb313e04c56378c8a10110f03069e9b2eb8fb991df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "b1b8f2b643ac271aa5198a9392c9897fd318eb1e7f0ed9e14b2d0f79f8c0891b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "5e66f333670a426c0fdd55ae90147afca0d128891dfc634d46df7f924787fce7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "efa826c6e6fcd69b47b6bb1ad4b8621c314f762d5e82a0a7a50f2dbac7e105f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "5e868f423b1520692bbf798798e9f55df4d0f83e523f4c077aade954c332c660",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d079ace0dda80decd245e0e8df1c82faa65975e8cdc24872a020aaab9b77c30e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "fdaad11c65d85044e155f14b58473574fcc86f23b02bb1b3c32d712cbeae0ffe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "4e305a0175f0442457274b283600a13cd20191fadd7748969bc608426e821e39",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "938f11b14eca39f62b5f5a828f85d264d34abb883699a82b9798814d414dfedf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "f8a563f7578826fb2e3de36d01595638d48d644290244608a94a1d12a1622cbf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "ad0af83208e1577200207de5ed285b8965fb79a8e222b375906dd575a7a2b8e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "ac85e5e38c93b10a1afcebff444e2dcb120759d87d5548ed071225b6d6409241",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "05d79fa825bf29cf027316b1ac7b9bbad53629dc446e73af14d997cb3dfa769b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "bf7322bd8b8b615c1aeba18693d1ce6cbf74e76aa28bc52b48dfd90f5aee5481",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "fc7b27dbfeb5bbb7ac76f383daebc6badc47ae02ab3848c80002191056a32c1b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "49d276e272427759a2bfb87d75ea541ff639094a65472cc70c18b26341d90983",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "93d93c6b686f838f342a40b4d9846979f9de35a24bb711d856012543c0ee897b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "1f55869c3a63cebdb03faa3514a2805a4af6b47636ab699dc1a95e7e3595074b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "83f974cf48116dcd10ed4c9e95f240cb0a881c7b0e625517d744e445633d3080",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "54e9c440a6ecf848f63d5a033999f38effbdc0e608da20573324ab80bc504536",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3e5e1a70a70d1c2150e7a67181187d79c5719a74c9e70ada1adc20d1e0adcbbc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "442134475b4113fa995e7bb19b8b5d7bff5b0c369db93a9cf10b552e80112301",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "9a48c1795595a171ef584073c3359b66473e0d87691098eeb3c66f3af0571587",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "19210c1f404f152515b59e9b3aac8c28672eed74740bfd3c481e35edb665397e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "cee93792354524f6f1adfd0b82b45569edaa939120acd134609146662f3f089d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "7298f263f13ef2420babf97b165c02bc9ae8b8178bb7a495ac602e3adf0e1e3c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "691205519b2cda87b036a342a6781a7f4eb3b46909f33493929136bccddada54",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "7f2e5a8eb95a6c95ac5c402dd0d4e88aaf12a7a8060dc51039463d7849b3bd7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "d0f474c2f3d0cf74b18a295958c63e275085c1d6d7eb2f701fc6ebf546d47112",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "84ffda593c78fb1d4dbf6a146f72a6df568c995b7c4eacf2ed17f73a16d12d3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "f71931b85dd27308d6f371a7db15e4c39bae0220db84317e678271f1bd321827",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "3bf6903502bb4dd922874b574b1a1b9a8d510283a3e72afef1e8330090d2ca4b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "eeb3559dabf56847f665f2cc77d6c96acf200518d51ddcfbe77271aa8f3d3e9c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "36b2c1dfed19e6423d2069cba36f603106b53bb794f797e2bf93da61f40b20c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "c7fd53818d9568f08644c594af21f1a4df305aa5d7e72b93c51f018f6622c4e1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b5f330843d64992899d6fe3b16dde2b489e7919684bcb5b4d0b27115dc941e28",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "e6887ef3f4e8e8818b5267d59108a673adf7fc18ed5b5b6e737e1eb754dde5fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "a49f0544b1b5873b62fb1b5df68353b729a79bd224258cc5be88c57a6a69ea88",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ea3892b0a666d9a42b59564db2e0863d51294014f6cd45b81166e02bd8eb6064",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "a2097cbe76c4fc81f3d36625c3e59b1509aba8d2927f7f2783237f512c30d922",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "311d9c4fbacf2cea08270f09816814ea501bad2d37a2c3d59cdd07f10063e8f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "844f130e8c09d32406de2ba2414cae0c4287c25c58d68f7af3c66de7e2a1a823",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "1a8a77bf5e5ee4ec92cbcd0b5debe33f93dc4747311ebe4fb48f1c746525da0a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "dc875928b457898e03b06e26f6c644c9a51f6e64ed5cf9406613a2a909bf875b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "aa16348b7f29198a7e1c9d1184fea93f8d9c1c07a20f0e198eb80b28b5089eb6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "4defdea2b6f141efaea3a2b6469228dbad030a6b0423557f79a7dd5a0b67e5d5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "8d2274675492fa3dba71aac06fb00417db2c62908ac7af87dbb291bb27a9b6b4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "3854f4e5880645aaff9abab3f48cd4a856d1a92d193f7e2fa31d895739793edb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "fc1d1aee17735d012f1c167db21019bb364a0cae96652d122ecbeb6c0c9afa01",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "b0dd871b6e08d2ca04a1a1128335b48a0f95946f176d62be55c43e225d882ccd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "6c743b5339866001ce842f2244f2a8da1036def56de5315eeffb5cb72a604a1a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "5c5d811feed364b4ccaa3e22562b899a20cbcb60139ad0b3f5067e5763f6c878",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cg-ask-fz-009/claude-opus-4-6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "e2e55dce85ab26a1580a7e0a9f5f1eff413160d77e5abc0093d0d7c318d2f569",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "45da24e6a362938942e146838725f1b999d5cfbbe00986dddb35d960e93ea831",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "9abafdfe25eace31efae18dab6c680e6eeff9677bb645283737f7f2f9d6d2732",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ee76b94261f5a83ebc9dba9481a0e09d7cbe923a3cb8a5113171eb3f011237f9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "25a7bbe8593e2f649070888923e2c8d0538c0b8d783ba31d3f3b90fe76c87ae6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "d5c44a17b5ed3fcb018a6b018e671cd81a194fb2811cadbf4247ced0587cfdc3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "209cbae61cf10add680a7b42cfbb7c6452ca09f2c951cbe69d7b4f663cb0971d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "56d1d821c282f7181dbe03e809dd13620fdf480f87d8649ca9d3eb8ec64c4adb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "aa168f5dc56d32e3b3075fa201e2d4576756f2ff69f8867c8fab674fbae139d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "742bec13afcf8ad1c3d2c62a54acb749ab15916230f7c27a835c8dc74ab36283",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "075b223a6b72a37cf9bc8c5e48c64666cdcd44d4c3511e138d9841f88096c24b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "c34a7d6e3697fa72eac64eeaf1021eb9cbaeb67e4ae8f0832ce4c001949d2624",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "54269647a7874258e2573fb158f5adb0ec86839cad2a7e278a6c682a6aa6ad2d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "1d12adc0770f8cd967089b4a32ecac4ee565e32e43523ca7976cc9aeb8f8a08c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "47b40ccf59fdfadf6eb6311e39a05104ce0701512431686598aeb8bbf835a4a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos-hard/rep15",
      "rep": 15,
      "run_json_sha256": "0e27838e153115a9a79d9577e0741f7aea2970b96e6ad0e1491b3c489cda4747",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|ffc82068051276c8ecade67885ff45af|15",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "bb42b5a83a10f6fbab1931376f78e764f7de0f157d9313bd2f9ffeeb2091f9b8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "82746e67d4a6cb461a40a2b8e967b5d825be98a0b8937dd27295206a5751331c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "6bafc9d35a7b8ef305f39a367d9620cc479a13150f4425597bf3dbd0cc13d083",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "430c1d7e562b16e7907f0a3bfe573741898aab055825cc59c3dbc0a677dc3dd2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "310821c22f351ede226d96dc2d2a336072e98d3bd839e3be370ac4d25666b996",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "790c5201e720d4134166f13150a8b9ebde488d9079dbdb254f7c3322e0200d3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "580331622b7f735397aebb9b4cadb52a1f694b5223989dcf6219827c7d0d5f6a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "72db769236d9ca6531359b78fd01e9c54a4e6826478e85db54201c737882f4b3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "1fbf93645c77e0e54e52b9cf8d3e5224e0a2d0e752ea8704edd1e1bcb364a7f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "1cabdf767b7e789f072d3bae6e4d0d4ba6145a635bffac2b7d1dd3319e3887fc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/HumanExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "cb6e8d10409b49ef632b42da70931687a8e5917b4e22bce45fc02cea682f97dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "9b67ce9037dbea0070d8b930fbd5c0d65e12d75c64c960bc39a888f7e163e6a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "adc74888730eb8450e07d7f6551b00619d7073dccd6374aec9f910457335b75a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3dfa6fcb38d2b8ad18e2185d66a90a89b68d8fae28b40cc601678dd980d940cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "f9d315cc5050dd56813d85b143eaa870e15872e01b42e874a1001336879c836b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "b0ae214291907cfd26c7fd5194e67dcfda76a99be82132a3d10da8c8b634478e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "510f0f80feae3a9d21e2902ba4594cb0c2317cbb059e208fe8bbd61d16850c04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "56a9e78bea864eb34df7a11a5b8a5cdbf847b4dd4d80f21fc6e9405d0d6b02b7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "c76f7055ebcdf6426d89bcf948cd8a0df1bc147f6ec11034f6f2c70dec3cdefb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "0ca8cea7a97393dce052b8252c0312131564816f5f86c1d8e407be72a5718a33",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "981b1bbf3c7df7d66ecdee5da63375348b281af837183dc1b22501b182c790a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "754fb4a2a369817aad4ead6001ae80b92458271f253068e7edcfb6be941e451c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "9b9d0dc8dd788a6cc777f69dd679533dcd5fe9a2fd83014bf1f7e1f8227235f9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "3e47211c9f6e6956c35a96abd73b46e573b66b99470a71bc2ef274955ca7ef8e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "6b92c0ae0b398ec6525e634227c81699d6ea1ee10dcb05fe0723a6998344f9e1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "c3da92aa8dc47f2542089e9a211f1d74415251db054eeff9d293fb639cefcff7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "1eb3a31b86169b927a88362b8e3a68663adc15f7ff01820b5cb9bb91600beec8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "cf95f99409f265dd4c99866d34b62f14324b306127f6eced3d55393e11213621",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "9262f2d1ca2278f0f3911692744bef7d602a0a146a7099fecfbda7887921f910",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "988f4e2d70158c9c97ef22be7cca57e1e3c0c3129ab2d5153fe50a8040e4bbbc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "6b40d2475b5a1cebb8c526ec0551219782d6cc4d5d392adbca97173d3a5ad74c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "6cef76814d88d5aa3ed97a035c745f22e798ac29e027feb03b67256dd1823919",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b6a52edf333c39abd4bb6230567a511895c97ccbdaea09e31200c36f0654adeb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "14ae036fbbc69fd86600fce76cdaf0a37f404f6151d9d690bb8e17908b5309c8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "e5d6ee849100de5e69dd8bc166ccfe5779a8a931b19c9a4e80140bf01b03099a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "102d38cc3b83aeb0bf53805c035508b2d6cc88f0b9d96abe909ed0e9395d03e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "f362376a9eaa997aae0589f8dee6e5fbe28b1796321791dc3a044cd85296b391",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "f170946e5945a9823de6fa8ca78d5f767568e394ef84384a117c6938d409b37c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3832cf707486c63980edf86675b21e1677dc08ccb014309005482d33e3ec0e07",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "84754508f445fffd17f8c58d6c8fd58ac7e5b9fe964a2e06b1d26523f2e9d338",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "05872839509ea9643a3f9390537d4930bc5b502844e470bc10ba682df95bab4d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "7a9eaca84529662c792904f960eb233b3752435fe7697125eafaf77d66c76d1c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "4b549f118e9ebd384cd1052b9015e0ab2113fd03b44fd9b53d10aa19582719d7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "e5ebbc57e38320ac6e3cca6acb27928b25ad8868d4d48d212b287b2d82b43cc9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "481399dacf7b7017fcf7c294f36bac68e7645f19d2a853e7ac9e5e0cb973c428",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "2c4373cdf98c28caa1418bf61db0be51a08136aca7cb6ca5280cb2ac3064aeb3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "ea31c01cffe5ba0b6dec9db70ac1fb14c9e9e9d0816fc36a26dfc0b27aafde40",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "51748b56c6fb4a9da12c6b79e067ccf0a242ed9275f74cb1119a48ebbe2babcb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "8b7c09b20a06c8aa866a8a720d6586795a8d3302fe6c8f24fe3a3ec7f7ae8d42",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "b1c240a836da84c9398245ea11a04288039c03e82b9360af9a7a33c4d1603d9f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cg-ask-fz-009/claude-opus-4-8/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "3cea5e522a38a75e0fe86bceb3f2d85553e4096dff4e416c6e6063ab2acd434f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "156bbe63ac93520755c2999448303bbcce7d3164fa1173306f9e89e399722115",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "9e0d51640f743c18227a1485a7d487c63a10e3836ca1e92bf7512fb6db3cf614",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "c884ce80cb258236b2a42dd99060ec63fbf4c6810d90ec017ac91e4a11f7ca84",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d560bbb145fd485d9430d046b9bfdc96318e9cb33cb82d70a27ff642a18b3f16",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "da32d55948c1b68a054578e7ec74b83b5960396729db7d009d8feeefe2fa4377",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "258ee06fe2cc426fca4828614cf6cedbf053eb24ef1f0f2c189a4c1bcb51c442",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "4aa55faf41d5612bc0e7a205678788c9e44d395f5d647cc0ca573b27365b1524",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "211e9ed647d0b714ad5dd3a17cd22d1ff8ff39fa2a3dd0137e1e84c104416f66",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "4c705e51639439c509eb24359b2d2c191a21aaf6c3bc1db3d5a8fc1fde178728",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "a3c9260219ccf9c7ee1aaf9b3698dc1e793d29e3e5f1caad8b9bf9633786d352",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "352314e00ce0229c09a6e550bc628151ea1a408d0adc79f0ab0ef9c9e403bd58",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "5607065bc302965b801e135b4782c8d3f46698f31479e08ebb8d422577ab0c61",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": false
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "58d690cfd2ef8acdbc571e603a8edcb028b634b77d38373e4305269bf5cc79b5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "c226305f693030d6bd6416cb86aec5fd5adfecd68c61f3d1f6ac818273251870",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "6e2e5d46f09a41fe1252272b3a51e1b1a7d222f0bebec54d21f797b16e62fdb2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f8ad9adf6b3f3306a99015589fc98aca9d3b279f1551552c44e6d1351ce39c27",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "46d95a2685a1a6c3f4d0904577a96551c221b03ad59462d8d7751cfe9caba03e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "d906817405380ef2a784e46e003b017bb5b0d8030adb13819ff7fa095e81f794",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "cdb4844cd43f1a48fdb0edae349028234517961382ba60e818aec1ee65c6571a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "e63e786374dab9b5999e28391e316c22f5cf270f54288a7ed59291c5e2ed60e4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/HumanExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "8c42939f8441b4ac79983cfeb37df392aaa8fa1f36ec508df7e0474e86e83a23",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "30dcdcdeb3dcd582108487e785a56f628a1d083157f591586193717cc43e7d4e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b438dff9a0e6401dfa44c86292ee88539894d02663e2825d893ac22677225df0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "c2ace5d1d6d279cbd2695ddf88840d5d2abafebc2404ca1e8c2948e416696b16",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "b32152ca2f6395a93829c5413a4c0a0e523666d8fc36011375cfb4bd2c0de99d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "6f776d8344cff63ea9d1b2a2df4fec4e65a444b7742d7c51337c8ed6e2fe8122",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "196327921e3e6b4bff760f615b536602f0f9a803b1cfab7302357281251acff9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "6916bbbe4c2a91ce1a38ef9fd60c3b8f03dfba085a8fc5dd1c358ad3eb612f12",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "7e08a297a8e6bdcb2f111dca8bc55d8f67cc51457e46de6f34041d1a31e6ee8b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "cc4b283b1c2acc450d06ea921f461ee98b656ea9a847e86bff68d7a36826e260",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "601194e925e23d2ebe5d329eb2c5b7150cf5fd8042cb82f8c82fd48203c07dd6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "1cb727f283dd0f49f98c2fa13242ca0090ee64e3c44d79347be1fb3ff966ccd9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "7258981766cd86ba8ac417fc4efba289f0659bb7ebee59273951e7e9936109ba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "a005cbbbd364474f3e4c6e095ec95a9a0d892d5dc27da7c933d7b0bfc0e71d40",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "1a3b15fbce6c9e7bbace51a8aaae7fa94ecedee39a4bb341945b8c3b816153b2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "b2cb3ae1f4365a67eddb82d030b4a0b8f279d0feec6c3488ad3862422978e83f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "ffae9607d0e894f9de5cae505679436d3bdd497387188153185fe714f2d912ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "3b6e062c49df704c7fe61e82ecfd397ae712374b7fdce7648c30ab4cd387abf8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "759c26cd1ac10a723c023b720003b5e44ba5c73d31a890d0248446a5870f75dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "555d1afff2ee219fc84b76defa5fcdc5ca06a7fd01c5b956c2091cf71a1532f3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a01cbb3bb7790b95ae018ab7e957e964c20087a3f655c191f59fa6bb5715356e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "bde9aeaed3aa06dad272192829bec3b1c88f59d84c3c476057125d07e4a42f8d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "de82eec3dab9436a99f0ee607608a9bad70379fa20245bf253c15d469edcea9e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ff916f0b04cafa3c3361d96b3dd4ee27df2f2a578e6a43f0b1aa1892b1779ab6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ffdc1f336c528fba15eb4c97871291bbd06063027baaea377a3e296fea360c90",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "89cbba8cb2a425da26c94b0d228cb80ae189b347e1fe0b80c432cfa78bafed0e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e161fe63126f14984b051ab0c6f446cc17f6457c332920947f1ff6b31aa6f27a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "71ae65723c52801d7ad6168ed6b746703c2bda2b998c68349aa4f5a05526d9db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "56106991a674c7a0fcad7d9fb7775887d159feec5245b6989e05fabb7ea80183",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "34fd051a9483a340837ec8606e7c26d18d7de1087a135773df6bb4d360abd317",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "492f259a876e49490c0b356588456cdebb4de1a102d42b64eca6c3a4ff06ad2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "bc56033fec7c32c9e2993023af8753d27572143a793906f6f35e5f18a4c38bb5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "616bcd5f6f178cfdecbaed0156813de204733d18c65bbfe4bee6aa491eefa429",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "4c51ba39f64e3e7b56c4a65a31920d2635ee15d9c8423c262feb1563e8c9d5d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "1084b78f48fa10e96625d7bab9d650f293b06df171f7375ac0b4e7aadc5a5922",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "1d4053696b1a8c7664befd126c3c736815ee75b9214c54b705a266314824e13c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "36b9ca64ffd518bd4cd62acbb90de9dc8385f79e9ff4f1d0e3c3bc7c3e6a9256",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "81dfe94d8c96b593ef3e0624579abdd3a9a381b63817215bc7c12679dc20b83b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "ddc66bbed11ea9963fc65f78c9de3572c1a70349d6a407d4a153c77f48c88d00",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "8496728730ff8429ecdf54052f5a3297aa1b75d660c83924cc067e8ff2486825",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cg-ask-fz-009/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a015e8292288c8e143af0c54c16ba576e94369b712a4649223a9766967f50375",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "c0262a489b59ad931cb5238c60b1bcf8d276eaf75294d724477abea2dc2ffd12",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "136313310e685a9072b4e9257c34fa8f7bcb7703d0f3e21d112830a9352670ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "167f4d0d938ff6e0849e2680f9ec9fda10b4b6a2b0ab4f32f5ab9eaa9594371d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ba95900936ff80a476e526baae10c4473fd1cb70e9f6b4382413e57d9c701dc4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "604189e84622acd4c90fac64fcdd2371f85cabcd76cf7c07f0a583c56bd14807",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "b0deba67c4e5b488be5fb9e20b14b06a0b20958897f4aff485a3d41e394e4dc2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "0a390c42e52956c9181785f2380381f2628d3e3f42f3ecd9929005cd6b96ed87",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "e3e8cab722bf7126acaf6018ac6f86ec180a243b6cf91744df0ab2ec2ec27559",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "51a1e1df6fa76c8426e7e57dd18b35d6712e682ceebe82e8aaefa6977f3ce230",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "2d7e2a9dd8b44c27d357ccb09160044690e8842521a15b4b375e99694d7d6eff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3443cb08f074f5fee655f31eea37f0b2fd563b3e3bfea82db12b8af4465fd66c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "49f19572c54f2e3a0cb7af666a12d0b52ebf509afeaef3ce5f9e239c6847882c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "b0565f48fa7e3b9590283581ec67a8e129bb3935b4f2f9d9c6632f0d5697720f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "a900d2e09aef5453a050b1e2a607ac7e4948592c89468c0bf83b8c362419b454",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "cac177f9e708fc3c7e2cdeeb8f9f743cc722d09b69a72a17548605177a73d7a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "796ace8215b1d3f3ffa8ebbed272bd8d67e490d84c37466718d00072f3778ee5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "920e0e0eeade7662b658273386ae6b739c90b99b43071974c57cd391efde9d09",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "89f9b1a457793185dc2fc1337759f060a3d54aa2cdf0afc6c976c7592886089e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "05ba4b0ef97e9e6fce0a42cd231b0668a0fe7ed711bc00fe16b3281ecfeea28b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "1c35c2a3696ff116608222a4bf4cde9ec50fbc8fa569a88ce03b146dbd83017f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "52d2dd94f3a5b8743cb178cb3a7d61eeefa2c7554569455533c54e517a1cf662",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f932dcfbd659cca353a1a5913698eef4f51fa4ed454e12b2474ee583054cd7d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b6eadf11bb53ec18c3c5d2b8a0c92a703b627dc3481dda725ddaa30f367b1a96",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d97c07c39054eceff79e11b32afadb8099da4351f2f6ebda8cd01a3eac9abca3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "e941f0e1806fdee8f5b5e21ab21d5b43fb0ac55050949cb44edd357bb0824728",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "78ca37ec59e38aa8e546298a1866f7d2b0a1116a6f09921fbab64baa1db0e151",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c3549a01e35293a8bc73b9d093a82f51580341fe0f015e77e340822837d6b1e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "ba7bee089d64f161169ceb6ac5a349fa86c033b87c4ecf2e9cd2b3dc43ab28b0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "7237136382cd51a60c827c77e2b94c96f482d50c38d7fa271bc9ff034bbbcb54",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "6119709c9b281b0c4506e015886e0048bd4f48fa6e70e686ee0d21241877f3cd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "8814c3e27a00aff131b018225c304466f15e45ec43556ec7095077b49988155b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b91c8c09f4052fda81846c1992ed51f6386725fbc7c5e70a29d33b3831b88bf7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "f4965dcfe606ebd5732bb2f1187e4bb70267154e4ae12e2de9ca70140782bbb0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "9f99fc72c4aed1af7f8eafe2138f2c9cba9e9386aec7435f5ad3e9f684e58d35",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "01f6fe54344de3385bcbec1a8a1ac9ad86899f8b65a96b629679218c3a7517cd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "5cd922faae1a7645a86eb3e1ab385e401453b8c5f278742d73cc8fa533046859",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "28d55efb1c4477d2b08ffa7910f2e6e43f09740473c53aa74e0bc28d06251ba9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "85dba2bbb4d55b230e6bf6511673d6836499a508aceeb528d4e39b76cf4b8502",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "50b0a4de9a24472e20a3fa5c1dea08ba8697a748693291a135d7e0abd2604fa6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "47ff3a83da343651a111fe0fc8a90ec07cc4aebe92fbdf53efcea08cd6f1b0a9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "e49debc524d8cb61c58604c50ef240be755241291c5d7273538964621b459ed6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "7b8fed01f3ec3a73213507e9217f5e70c5c655c019c9952a73e5770725d65507",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "5edab88e0d45ad74bbc4ef68676dfb1963def8f4292872957f4f5b666ac0c231",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "65cc539204237544b1e7476bdcde4477db441780dc23bd9751fa9d2c8a0ffecc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "dcdc6222483834bcbfcb1b85c7472d4a37d5fef53547c321b20601c8259ac989",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "637dfd2d9baffb687a497b682d956e4f207c3e2e05c5cc5ecd95cfbe66c1ea84",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "bbd5f39630070cf5888959e40aafb7a3d25b47fb066ad100d52ae602c037e438",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "5913fe6cc690c0f64b3c1ceb0addca52eaae02c68ed606dd263799412b319cdf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "cde633e11f9684663a1c1cf32effe4133ee522581e10aa689a967cb48b88c91d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "56cfc97410b94dea2a7f6f0a94a250b38f4b342d29fa434a489a59d506813010",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "9ead216b95095e65d84a7fca93a691c45cef5ee36222e1fa642b8336a5693391",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "3db0021d13a1b3ea40f79b6d24e95b47d8f7fd80cfc86769d6d9922fb2bef9ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "601421235547c524988c1d6caa9115ad63bbcd62be1661a3756670dbde73c574",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e20586a8bd969602721e7f3985e59cdbec5529876901f698f55bd699a8b44035",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "e30ee02c81e72414ce62b91c6ffd63f4b3c059a485989796601af65dbe83899f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "4b569ad2e1fffcd3551ed5eee452f0a9b25290bd414fc2d65fac5b8d15815e75",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "0a8a36f555fee67c734dbff2ac48edf9264ec24d5bda927226d221cf316a87a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "e81feed00400921d16d9287ab24b56b6231b180e5d670d691c1afc606c0ee501",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "5ebdf4acb884e99cb4c2ce9697d9ffcf95b4f09c34148c8934e248d433fd2e1d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cg-ask-fz-009/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c3574e50d891ff09aac09cd78a87692bd7b4ec432a52c75ba712367f33b67793",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "6604e9334e62218bd31b74da00846d099c2fe889cc9258b2d6bed951540a6710",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "e8e0cf73c0a43a4b84af09cb36c7db08c214b6958a6a146404dbea448a68298e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b8384e32384567dcd07bb1948d5a83c7574e149ab3326e21c05bd1547dda09a8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "87ecd54f8e3d037c7b4d604ecd9aa2b0333246132d7f369e60872d1c5dc89c06",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "5c540a162a4e69835628e54ad5318ebdcec80993519bb680699d7a023d7a9f1e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "d774eb741f632e9b5e85554d8b82acd3f939302cd8149613f3f08fc5a39c4d3a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "6f4a561661572b4524c7bf6cd69b2c940b7dc567c5b591953dcd1daa4185edb6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "bebdaa331affbf37e5356574050c996abf1c23b6d2fdfa37f765153280881e72",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "6885e626a4e2fc13c49c49a293d4bc6e845e00d992afa0c779294d59d4a07f36",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "a47cd5e5519c29e7a5176e3056c3c6807cb51801f624109866a4e031f248b34e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "a2f8c0c806c3ae0ef665167b9e0deff48b7ea258a92a49786474ab56f4e5e938",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "4729571ca729b93ba8c06656e0a8edb82bf7ae47cb3e82fd3c6f20ee7bcc4448",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "6ef87736f5c73e59765a9f7e4f2be5982402cbd7f9114f368edcca6f23d02ff4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "324414d62ceadb74a14f3c1c7abca986438759e0d11b4c02173caf37561832a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "540b81b420c3565a198e892b71553113d9d7b748e7be711ba2d2a41311b6d5f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "d88d78f6407381f84d21e3437943f4476e798d8e509ec0fd3ccb173685a74a68",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "8740468f9d17c2bfceaf45be00f6887e338a18b038bdba86421e26c3b7e3b24a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "41140f30f6bcc614895ae5903c62ebe10cab463a8b82a5b01efe205666ef04c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "2b9b42a97433721e1f0030381953d71241ef46dcf56439bc68b7e9b56f7a2ab5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "37eb116a36c36e166bdf5a9457603089d431b7f47ee04f202defc20571d3194e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "6869297eddbe06c7376228d635234e340a84061d0b37bdd22177db689e3962fc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "a76b563026e7da3c63b8cfa55f394a7124aadfd4b385b26a452c8bb24f0b1acb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "e39311efe94c7b80d3eb00181c9b41592606c2035fcc3fe5647225e64a67ad9c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "14d6ff1f38bdbe421dc614cd234079cb1f093e32629a82cac4811457d8c6a0a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "a677e1a97fccc51177de1b6be64707ea9bba68f5c26fcb22301587145048b491",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "deb8626c213fe325c60cd3c3b3d892ad05c597d2c61ac9be32d56ad8ad2d6c79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "a9a47b97d7427e22e91437497e364e3e89e21fa3abb86051c558c6f80e902eec",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "99bb7e6164b1029915e8163d42b028c75b511abd706d2260dcec12b140ff49c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "66dc52439471d8e7d1b634fe08bff51fbb9a10e3c3c6ce23565107eb5ef4f98a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "4f4ece35441c5d6dec226e1aea97f71a61c6133261067239a52694af1d232919",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "c05e601f99e957eb122fcb550e4e497bcabaf752f1a6e8ff1991d3c8741c5e89",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "15bf5f908f4a79a615955a4980cf310fa0718276c68216684cbd5ef78c149c5a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "33076254cf2def2b5aeaf92693200ad67c25e4e0c09de9e70e38367a177c6b20",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e95e3f0edc5f896772e417cf8fda55f601d5cffb351dbd797e5f423f990e3e0b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "4c237809e331fb97ca96785f7554782a3a7fe7a0fd73d5bbb24141c4a51a9289",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "8115cbe109ee8aa0b91e7384bba69207b5cbf901c491f6864883f85706bbdce7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "38fafe0fe94e52553b0ce9f66409092ec376e097df94f7b851ce27bb2a4d855e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "108948725fb80e157477859b064a99c76179b05a4034bbfdf88d38bb6212eecc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "6faf84e9670749759e72bce72d9ebfc504374f9cbf065dd2409031e74362eca3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "6deee7518e23e0c58d6fb9ab1e24d0589175bc6afe57d6b383ba80fc4ec3a978",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "b32cfdd30b9abbbd4c1e0cd40b8a871896461bbb03c42c21e2302c469b7050d8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "04994472f84e5d7fff77f4c53470115ed7602738536827c181c62d7f85d6e637",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "a85d6806cd32e495572a4c22f29f3dc704fe23c4b9032dd0026fa10fa0a46a3c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "0c4c7bf642c4e1d933af6ed5742715a737ed3d47e1014e0722f9eca92ba8ec3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "46bd7be900f158ad66443b587ceddd2e93c3792d8bea0d56652c7e26a9e51265",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "56f5a01e203b641b0f54ff90027207d2b889dae533be32f787b490f7a3e86d4c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "8010d28725280de63c468f7101a4f4d23c5257c6b93e8a1198bdee10c7894ac0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "8d3551bb90e434fd297c50cdd97792e622be70280cd0ac4cad96561c100c6576",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "10edf23a7d1383e73be6d98f45fb5571bd16d2ef0837a2719049ac588c859365",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "42755c00510ec559493266e1ad622f3aca9f2f9a28ef03b132b813116f923046",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3d285a2996f321cc61eb40a52398d39902fb9eddaa08fe866e8ad2c4288b08ea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "6647ab9f32998a83b59e8bdb77261e49ae1a14746ed848b0e669f6d7c907cc0b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "44362882c77f0bd70f7ed1a883bb2d874e3be9df23e23a76c64699133a88db19",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "a919cf447233942b68662d179b7f1c72b97d0e94e500daff4093a8c98bddd49d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "81d32cc1514733da40d24341fcc217b87f491773cde8b18f303eaae44be920e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "11f51a5caf876c1f00856e8febfee39ad56700f0a1017b0d16825359e41292cf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "387e70abf5c7cd65f455f5668621f96d9406ec5043e317866b96fb28539ba316",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "66fba47b907bf1042a986b2489d7eb44ac6d0cea62fcb33e07db1dd72d15bb6e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "32e70d508cd4b5d7940189172cdb8dfa8f4a5260b10915a3e4ed6a84bbc301e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cg-ask-fz-009/gpt-5.6-luna/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "5db47e60a00c8c3e3a1c38cfc7c9ed48cb94777843fc23b63ab3736ec3d73545",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "c4565db9989c1fd080f27779206e05ec6b999a25be305aa4ccd57b08a446a057",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "90934cac4b7a0f042a2f79b0547eb2052e1b011e2d923413371ef9f1c41c425b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "d0996b80242d7d0bdd95bf05932a2c6de2e6c34dd9334d5272e78a0f62f7c825",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "f0d8651fecdad8bc62588fbc742a8e4f2c4e802f9585c732fb40e4cee181af63",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7522ba5e496cd07d9ecde1f50d1f18cdfcd15c20cfbe7a489e3b3d098af6c955",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "1ac3d45f2ba1622ad4a753957c2462c10c7aa7707ade60f420509a2ab32893f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d45fb1a860d85ca77da486d6dd1ace2199ee781b487297c6720c267de4fa7b1e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "c438cf0357fe593b98d256197a44fc0528c9fb72ebe9e953334f2c2e884fec04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "b7de595fab2c6fe3a259681c5e8dcf17f7f3895a41749a1058cc6b296ec59b1e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "121f958d799fcbf809d7813d155e74c19b833ce248ec7047eb79f5460f01cca4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "38f057fb2af7819ea8e10b13da5ef27f5bf8f1e91f115a0e993d19e8b231f740",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "f4a124e6269d00ee8e45f70afec3a05650a9df02607d831ef0779a5854feba70",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "994be18bdbeef802d1fee3fa63d1afa06f1d90ceadecd5ea401c3d7fe3f3e703",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "c6837dccaef83042844d3acaf306a89deef81d378106bf83799cf24ce933398c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "6cf6f80747b684345f5938f860d2eb976f1227031a26ed7ed4d8f4a950a1ea98",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "38e3560c29193747456f999fde1a285d5bb6f3b22c4630d6dc1dc4dce813b350",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "9b79f4c6cbb1adeec9b72cfad650540a51f84f77493a0278d6e26a67fa9db3ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "3e7772a79539ef635d7a140ef43e56aa689dafe3470f0cb2c3d9bf8ccc5f8675",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d049e981672be9ab2e485a89c6105e184ebdfd8253190a40cda1bfdd6b237233",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "5f3e60fb791edb7d38a03e2bd09c7cd39ddb19a221b9683b147b5661797205fc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "5412465e130f9bb17106c7d8b1e950239362343e8b861a71f256cec86aa46a76",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "de50205647ff533643aeb857f47426f529bd1766639a1c2baa01e0f3bddb22c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "62dcc0561b3ecff494db81b1c8b7ce7a5586c86809cf7f2563429ddf0aed701d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "68df38a22987c8174d95be87abfc5ba4640a84190d7d3c674d83050b55a668f3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "3a80d108d43ec3b6157f66acdbc3984e86fc8ddf8d098389f641af43503477ca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "db17d6661a2e6d9572cb785fcb528bc527c633eb8940706462de39d0206982f3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d867925b87a2f524ebb880a29762de6e3ae991da8afacb197de8747a5f9e9eb0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "9b4fb55ceb5351b73cc5399352725eff3506c8fc1c509bab74511a8fe34b2435",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a3eef6016f352d80e12d2e44e721ea0e96529cee3c0b55654d70d4f481ff49fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "9a58d60284f397e5fe23a3df7669476d5eec6bb3eafeb22d15f518a60b609cfc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "5c087519b0e134fc8d4158f4d8952c3b5d2d9ce7170dff05f3f2388a814e83be",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a32f64f7d19f177e8be9f7e4643ffea327e192367ffb61e276f255cb9ea49f1c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "0ec87dcd2925fad889b51438b05e61414869f1f22c5092b2c6aa89338a074a23",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "76a3d403514794f1738fc3fe89bc5e78cb423d7212bd1e47668ec32fedd024fa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "d4daadc275a246a791a422d31ff4efb5624c9ca0d4f56679bff715b049842d25",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "9139aa5be4556afa60c82976841ebef8b3df892db16911edc6180475572c747b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "fdf5b10dc8da236c6f4659c49e40bffe34381b68080fe61788adec040f8d0a28",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "b677acf318f42a2f1dbb99f00451e9ffbd67556dd5c22176fedde1e3caa3e114",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "52bb318a434eb08b4c29885404ab089a3950f62466d46c33e407815ef5f098ec",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "3b620bc7ba753284db47ad93294673cb4003ff2c26010ea3e6aebedfb2f4e837",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "39e546644806671663c1f8c7a98a8594ce2a8c6ef7b403333de1b42de32f8d37",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "ae1cb9cbbddcbaf7a1a444bb76f2037fa86267de1c6cb531ec4b1333c5b7f0c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3f251ce024a386a3434dc1f052ea3867be87af1131286f4cc0de907c9abfd21b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "a3d3b532cf4bdd5cdaebdd1cb7329e914b678a0f14ee2693671363bd8a0256e2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7f49e9d7381c43e9de77d08cd562951bc470cff1cc7827b19e29b2a956177a3a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "1868459c1a1f05a6eefbc2489fbaa032e1a07966ff3a7b64452f4b713f32d187",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d71e32f038a1b6104b21f6bcca873771958dcecca0291ec7897b5a5a911926e0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "38ac031a240923b54ca3169009411654a696c1ee694ddf6353254a07bfc2980e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "f5fd8bb2dde40316f50670ba9141a24a3e989266a9b2186cd6ca3a9735c7f37d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "e99cd281f291afc8d85d5b2ef080b6e7cb7648f31a371efbd5bc72ef2b6a8d27",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "a3d12721c7f2b2b867dc4e30f931c6a61fe46148366cf57c9ee9cbbb3e0d9274",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "3baf331622e4f5bf91a4543e96a796d54ae3a949979d70dcd03735f50256283d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "e588e4ac59e39c54492f18de395065216fee5b926e5362bd2e9db223d0277ecc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "fb37db055152dd808de018377120b2ba3c5336b941b15f05b2214cfe6f23003c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "dd314f621ad93a7ecbf01a88dc42e89fdc39bb3e54dd4ceae8e247005e572483",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "1fe0c9c848b4a6f38157dc19baededf42d721c473ace2d8e5e90d077ddfe0f70",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "f18b3d3ca4ea6cfa9c15f187c959dd908ecf8652e137c48a609f3d4b0bf6e879",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "a65f0c03a15c35c756157d90f3e473c8f3d8a25bd1a47ea2729b1505fc7d9e7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "a211770b4227b70e37f4ae5a562ead300c7a91295071a15c332e6bfcecd5110a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cg-ask-fz-009/gpt-5.6-sol/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c8d3b26d4da5cec00d0b8d305849219b1aeeffd5d1e0584c5b689d755d3e6390",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "adeac03e71b71e2f970145c473c0a587db303f2ddf9faac8fddf1a7b8761d1d2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "0fbc24423c7b26e106dd9c5838dca8e1e43af2eab1c07ad7ab0e6d5b00218353",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ce15f500e5ad39e7b90a66d23ba517ca2382d3d0a060902a3903283930b1f25a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "c174d77cee86b3ce98be13ce052566f2aab7c1120e84ff2ec984fe18543f13f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "cdd43119eb0ad23525e813df7537a6d2c62b329b1caa34616da247db3547cdb3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "83ab93bbe629c613461dbd2c3100331b1efafe2cde229af16ef8589baf9ea585",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "acab209448a5f69f15fbb5394c51d842effb369a34d4c21451786df8e9fc7003",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "5a9dc5318bd2763c4958c4c238fe66b6de6811aac963fc278568cc09c85a1e74",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "765c3c60a5dcb415f39dc7759f2fec7b2351b48de8acc3afd585c219db52c8a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "032df59fe19911b614f29f8fa0f8a59c778a22152691da73fe7006492a56b990",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "d9c23222fc69ea556936e7edb8bcf3b6242cb856b3a87b0c688f08d79c2e5c2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "2adf178ba9f457a257107a5707e54c97b206acbca79011ee7359930ce4bcac93",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "1915b9d4c469849a13c26b74862951b72de615b43c1a087e3557644273eced66",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "4275834959f4d4156f11c03e6fed8ffcf7ad5a7d89bd387d6b0dcbee8a07b90f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "aa24afb3fbdc5171286a0369ffe24bb72a66c377c1dc5b99022c1ce3eeecb69b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f7bbc52217011c6128ce306142d55bacb130f809bb64d7ea948e69d9d6b22f79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "a15ba5456c05e3897f95221df5594a8c16c3d3fd1996236224ebaea4e3d4eaf6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "a56bcfd5a447301c690d584b61ec6f75a6d9016ae384be1e2dabcbe78fa441c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "c8bb3b112d9395ac38e04918b47a68c3057cba68dc10a9969fb73e7f5fc20fe3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "87112ed5bbe5cafe7c76969b9b2a8373347ed3af3e2ced83747137a0015943f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "74d3d8919a0ad1a22fda285ac11eb9bab176220691bbed7029ed74608e4c44f0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b77a492c7d93d92f03d472da1c7b8b27e6a06255dd821ef28399b18215dc28cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3e940d888fbb33023542a587da9251ab4cbf894c7796020274bd500dd9670718",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "5b47f46bb82e0c5586394469574c06647e0163088321e74f932b5fa5d1094d9c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c78824b105326f6aa02b2eca7932d3f7408ea02d30ea0df4418147116404e340",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "a9c72c1346e57fe38f6ae0a72de1b6668665e104febd307e1357e5369494b453",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "4c4d126929b9617880947d3afb57a64e394e997018bdb5126358e99212fa971f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "6fa1cc6e535ac6536d6fffdd8f9018d1d44cc855e3d78e36db8420ce897c4692",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a6943641153dfb79db5450193cdd1a4a654830cb2e55227485cdb8c4b21d53c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "42a11843f1f37a7080c93a9651e84e583e1cabdbd4974c14eea7a95970f4bded",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3878254628f324f838735101538c0bfbb91707b0f658e8ce674e821716b472e1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b3f8982a9a583d6c413cd2a440bdc157c79e7688f8417005113bc78baf7219f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "9738c5cc04684242e242998ee43b2e46994c8eee4de93723806a16c4bb56e9b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "557882964b821943a5b866048fc2b463d2e02d3833846b9e05eba66fdaf9036c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "28adeb715b626c56543daa1a5d533647841289d8c20b4dc73dd4f921178dda06",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "3d20262ba746e128c40300e45c42314b0724e062201a9ef8b13b71105d57213a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "705b2a630f3d56f20ba20576480750abf4e30aa969e45aa8c68e5a10f8ed64ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "50702b3269c5d4e95cc496cdb21125090d8748d1243e2815958b677921e50c57",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "1802d873115329aaff04c14618e63187614d8220d23f9395cae0ea5066f30bcf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "659d95ac1ad89b89c7bf10fe9f2fdb2bcda5c20fe7d42504ae9c233870623f0f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "7746361d3e36dec71e0e6b9d0fb356650daa5d8b47a9b4b9b5308d2bda285e99",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "de97808089ade0bec5adf3f7bf11a53041705acd2f0a2bb72ba054da38ff1b34",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "57dd2554db08eb1cd2c7bedbbf5b59b50954db6a6d455c1f16d7ccb990955a2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "8b39779204e3a3b8b41f683d704c1e2cf37dd0609ca941f91682c3cf28c8a31d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c75fe4e839f12a7d3ae6c5d3ebda34ec6cfc1a41cf78fd99a5e4ca591a7bc807",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "53d8e2a4f8830bea8ddd9cc85f794efbbcc16d434a8253629d98692d34703337",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "752f8b1a2f0e8bcefdfea5af011a81d421e03f9c3e771cd6d124b70213694949",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "cc44d6abe05aa217a7f5c02a4eb189ffacf0561e50a86e7ebfd2be20465e4ff2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "2da58810f994e0bf4fb8cbff0695b40e311ed058e70dcaf5251a244b8ea02cc8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "be1a534a14badd44f2e273e7bc8c7d90531cb40a559254ecf506a5c186da8299",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "9148c5b96f7f2b9e9fe9dbd6e57627a833cbdabfc3cec1d8b8e90f0f06d3ac24",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a89948db6ec7b07334f52ec70a7d0d009cb75b1aa81358de8bdc26e30c23f9ba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d6dd78a12d0b498694d6021f544747fa73d7de3b3776a7d89f7bc6ed82eb6ab9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "8254783b8bf73fb2ce2463deae1ca961392775b4ddd38557267d34d8fe440cd5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "318715def3356ac1589e14ed3b7553e797d573fd52696bf28d986ff9c0fb2124",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f33b804bfd77e0223889e5b4c67a8eb73545ac63f13f9eea757303295634e834",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "e2bd6151cf0c55c7d052270b085e1a31f77be309f813e219fa5eb5575fa2e28b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "f3c3767b5994316240eb2c2b355e3a45e109588f65e8fd7e2c8712d673b235a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "6b1295a4a6f169578a2f998789e124a6afd2e528930568b4051fb181e7e046ea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "cg-ask-fz-009",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cg-ask-fz-009/gpt-5.6-terra/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "242940f68a4edb8c268c019e8e30c9b47512901571bd6bcb83a91f92bb708811",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cg-ask-fz-009|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "corograph",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "b03f7d05e29719a911da66d68adb7fd99c31a2b6df7f7d1a405bede593eba096",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "19fc587b6593f8eb65aebc924b6ec320c9af5e53299886b5721ef263fd80bdee",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "997610c22464955ac6318914d2455d317c80f474f5a56a97a58e7ecafef5bec6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "5e75c8412ed9c8b38d232a46c8becf2dc4c20b91cc92e31a7fb0ece3b0569d8d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "972678d3ee3bccfe49e41e5390575c882fc059365f00811c25931daced5e630b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "96d5c74ed59366982a9fe88c16441dde11a8619abb9d3d524e5aa4c86f5f931a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "1ed5e26512c031419cec2d0a65ed46dd4e55ecfb5f69381874f1545b35a2ecb8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "75ff952e71cc613d2dcca95458ebf028be14cbbed5ec3c74185f2a8c7ed16290",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "6c4d33d98c24f9c0f15d366eb785b6a981c0412a3747a3ef48ce7d9c48f95339",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "3cfa1c0158ded49b496c7cac7ccf316cbbec0c99db868f25ff6287293df4d73b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "17a6efa7af3b44eac7a90a8d54b08d5f1682853f6d514857ee9c8062e8c5cf38",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "c7228e47d7ee64a20bac6fa4948c04c7504016bd52f3c033b31a7f2c0cae4911",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "df8983d0b1cbb35a1ca7192692f94be71f700f38f10210b87cd8d02db538c305",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "0a8778d4f315a1c4cbbcd576966624f7203d8715d6960484d48c30f050890428",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "c869bc2cdb86ca9776ce2e459165bd445ba023f80c67f3fdb2526f0605329890",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "92f14443c1aac71f527ba33379f1dc82cd095639d55d6da4c0bda2fe9bddb0aa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f2700d8d7faeb8c82b1832653aef456971ddf0f1cf22c73297fed45ab26ce7ed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "efb97a91b8be2f889dd9d925d7b3444408b7b6e31c4f9dba98f69ca36ec6256d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "37261775935499e1b74de53f62de45cc9cfa3937476eccc309510011de207f3a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "6ac1c031fae16d7f385e6ef3ef4d83a405223e8452e75d7f26deb7f858f2bd57",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "60dfd226bd4ea68fb15a105f76e5016d9ced1a8f46771f2c6c05cb05236749cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ddbf270b8951252931148c27e29865baa6996da86d9e70773a00ea0fc2f0448a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d920817ab056ffe77777cdcbfd26ee6149247a8cb319c46b8b11249357f87891",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "ba07e4c8ed8fbceb58d6675feb6486e86be70289e8482985f02036c2b00b8c63",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "1e8bcc7e28081673e51fd4c52f4cbadaf960f41195106a502005bd6dd8c42846",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "1a3ad87ebf28ff95f9ea905a6311a327df3b1179b991d0e5d59e73daf2f8d1f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "76ce0c1f761355e73ce21a95ad3fd81d4c6b8cf31e5bde04a4ca5adae9cfd542",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "abffb863b637b0ddd5abbf035a1c5e0c37ce3660b1d2796a9fbc93007733cc31",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "a15e5345391bdf28b27063e4371297383cbf3f6fd62577c885b950bc1359e45e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "12d30d9d56fb48f6a8f6eaf9321fd6ec6da51681225483650aa33029bb19453a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "f94ce5adb18da289a3ed8b64be91ef6a5a30993522574fdb027fd7084c099f9b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/codegraph/claude-fable-5/ZeroShotExam/memos/rep12",
      "rep": 12,
      "run_json_sha256": "53a3c65ef66ec434a13b96474fafb6fd208c0b3e36d2ae7197ad3d4a4380fa2e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "ad44176a38cb162460a148b5b3df6f1c232f70bf28c903c2a228435ad51605d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "165618e44e7ed688fbe33eb697b949c3f64711cffe95e3fb917736b4fd7eaf4b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "6e37be2c4c86751c517d9b0c8c28751364ae60c18f19eff03b8d6f46f4697515",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "fb6c6207bfb44a2cc22a2d0a9af28d13fce4b169df4ecb7db48a1076f7042ded",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "3f3c9a78d4100b7fa1aed00a0220cfd76ff6d2861481a15f9d98a9308b200352",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "be7f30fc90d42bf67c89f5fcb6aca1540747d2ee43b62bec407508498efb53c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "ade18ea9d3ca445aef8af63fed2ac061dcc77182c4049b80855fd44afa727b31",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "b3a7505a881852728c1c4599babc588a9326727b31ab6a6146b09dbc9a0a3ddf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "d99feb851d29b6a02093f5a1aabff48fdbb0d97d4ce6676b0bc84d11d5709208",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "60e6c784626a259cd99552a7712f699bf43ebb9235c8f9ca4fa4297349772edb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "34cf3d7ede694331ca9abfeddf0c2dc7549191de4755e5e9c548ffce9328035e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "5199a274632dccbfe9ce9a41ad02d4f584e01272023b67a203bcecc831c31a48",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "eec587c1cb3e0edd0583a687cdd4675848838f8077c2587cf363e9ece72c4c8c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "5962e09d7596dd355ad7432c5da778fb536af093f90bdb2c78f8f74b4911c310",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "2e437cb470c626c0db35af45bdfa785d9d0f0095ebf08db0da978ff435272fd2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "4f5f340b466d9f8e304af0ce341df92705fc8c2e5d503f768d9e3d43aab9ba9e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "6f828cef317e5a18353497098e709d035b4007075f658744f147c0a7bc32562e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/MultiTurnExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "4bb5948cd982b7d3187123524a1fce92cefbd8c424598770bbd5fc141f18790b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "24b625b500d4b1ca2cc336d86237cf98ebbba46ad8b54e2c0b05599fb176cf60",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "da9177cfee4e7259d752f8a70f208d5dbbc7503569a662fcb6ad9e02b73bb51c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "61dea62e54cd6e55ec2190d9caa51642a850257fc6d03fe842ddb8356c87deb8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "6d5fa30345f3a12552c784e559489be87e433d98952d5af9f754b2ede5d73e87",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "2fd18221e8603283df70cde66c0f21f0a890dd79fd0fad5ae046029cbd9b9490",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e1a178890be8ee153ebbd40b48df370199263405cc2c0ed1f25d00efbf591f79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d20b955f02fb6a953f0c695c5941c87a1cfb95642247d2e8c02260310ebd2f41",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "5d078bc3090bc40a11d282aaacb7b5d6304e82e5d536537a3f67f4f354c82d86",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "24fa66e2022de54305aaf2236a49e2b5d63e49dd74bed8fc50eefe16d7972354",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "4f8abf0c4a672bef1920c38e89c5c8e32c5b54f7652e810862c1aeb3998f52f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "55c0f3903f5e04dd5d1582d2eb61e8b93951a8075c5f6b3b376a10baaeeef074",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "c767ff0207e462c22847816c0c2d45d6bab15704dbe8076bd822fa28ae742d22",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "6c2413fea91e40beb37cec72b2b6c922061c44463358f7f76355f8ee9c2051bd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "93fa121622b90ddeb05638797c8988ecdacea037a61ccab65124a067504edd9f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "3b88881173ace22248bc4927181e01f1724266fb82ebfd286d47759a5022db36",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "89803e16fb4ac35586f68a01124295ed7293b662aff6e0fa516223360118f5d8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "607f28a4d6171abdd1ccee874618c666ddefcbc10431b917cf881cec4c4c4bbb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "f84e4b75797801042f9a170aa5b4a9893e3fc5aa2f16e32f2ac102c3ffecdac0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "f286fade8873d671a89a5358ae2528f9b36c6b2ad5ef654a0bb4a6d5ab8cf40a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/codegraph/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "fbdd79519af752c42afa4928ebc2fbd2cbb6fea566321750f6b95d71267c5cc1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "d2ded601a4a3f155c60721272c3fb231f8326e6cfc330b38c26f44e0badd0515",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "72709940d221249c4fe3e4ef2a27f9bdc5d8920e12f4568410aa909f361a54e2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "15ebc47954ac414b6f7b9f2e44b68d996203d31b2b0f4e49f87600f103bd19ad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "6bb59b7f826baeb8e2503e7c9c72c97bc44be791515110ae888b39285425f596",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c33df4a8c821f4cf27d9c4e91c7fd89d025eb69d8e937008da403fe4865d06d0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ea5b34c87c39979fdb11bad26a1437ff3dac2037eefd93b69050930ee51bcaef",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "8c666ea72d35fee8d2e627765ab51327e48047ec8a6edd99c55c4908249b533f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "9af1a8f8070c9733464da53d7484dd2ab3eb8325d9931ad88f0cbaef0f4c975b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "ab075f44f117a74510aa1513e06c1ef30e3e5a3bd83d39514a4f18be28ac2e7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "7f08d6047c3277d715c0386c78d14a5ec0d2eb435909437f09edbc08ea0cc590",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "1880e366f9c30ede8316fefe4b33010ef0535d1f8d00f06faa879127916a66b0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b78cbf8d49dcdc2ac242170cbb77c6d720245c499576531b7ff5a04e5ba0e78d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "b45a8961a2b9389211a85046d790af51718cb6525398d53070f84df27ba06b96",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "5933fed5718ea163b74feed28323943295da99311eec24fe531babdbcf23e70d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "2c82864cf710b31a4ab2518d82402d2dd15ae4642e2325e752e0c85f0be53e83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "b75d9991c3c78d4d664900e1c357515c3f126c9c9ad5688db42b4602dcb422dd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "432dd82c57d0f1e301d31ba68b505aa96b1255615205599794f5c683ffa22364",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "cf7c83ac2a5b4546cbbb435499ca9b26be07a9ec5dec14d03c66dd3e941790fe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d4f3d4af4fbe4049509d6cbc9195558a468f3ecef67487654ffc280343cb8d1b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "75c34da8f3db36ebf449a37f8c23bf36ecbb368c18015cf7cee0fa69108d43ff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "20c9d06af09abcc9e0ab4358f1f6fc36fdac6b0dc1e2110b0066e49e4b37ebd2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "5658e8999d76782518b4d6b7a2ab7b956afd9cfec012289aae70457276715671",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "6d89a46b0bc653e32370cfe9b38c7327276d7a503487376bb6721c0931de7e07",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d2bfbd151219e318d1df838c85b12350193601ffe266bc0231e654835f060322",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "be594e075c39df3d52f041097a19e75bd7f91781ba5f9edbfbf0c100e671dff8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "b75707e582a8181fee0c95b47d7308ab6d6f1e39be4531d452f520c78730dfe9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "11d7f3d90e264581e63f0701232b1fce332b4a64618159e937d8d1c73edb0843",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "5a28f414506136aae2b044aa8c12d3c3dec98d0ce7372afcab34a3696624b758",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "e44509d76730be91c44d98c04b0beb7015b8ef8ae0c53654dd27df571ae191bf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "b983573a218035c6448f62f4ed0252aaacc135f1d1ff1f895c429e37200ddd5c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "6ec5fe9e016ba40ecb175e8ffefb22157c66c578dc46c6dfbe1aa18d0fe48026",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "77a07de7506958b3f207c631b86e9faa5c48f54ab88979cf4b238d9935b1dab0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "6329aa7d19858e07cb16b38bea6e7679f0524bb215fd1fc65e5c6f295f3aaf54",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "f52178fa795345a5b820964d0245f9b025bfb2f4c5b96224a533366c287a0991",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "b32c149170c2af6bd33b9e1d725af6dd8d1170949baaeb696c8f2b36920ca65a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "e4c70e50ac72b17e3f78a424a1ff2256e4b06ed5a0f76bf022cc9d93a36ca103",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "058e9b5bf99635bbe1d9e11cc962d3c40b22d3da7e858001a7d9cd4ae5fc9d7c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "b40d1a9c266b1aef6a93e5a96b1bf4f050722843dd878dfe2e3e97dda2d55c05",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "03cfdbc92f923e537d17fdc21e96d403c62389a6f75132c134359b7e469be0eb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "74a06438b09b9cc50da8443bd1185ea9d7741c940747c51d8573ca29369be83f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "cf478608cd5162cc1234cc827c4d9bfc3727bbd47d6a757026662d3679c4864a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b9f23f64efad63bf49f602b4345a101be4b9642baa4fa2ba9743c97bb30de94a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "c6f7bbe95a62a70eb6c140df9b005e707a90de882731576883bf959373e51922",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ec668333432aad147be82421234d22a5a37f1fa1997da8231466a17ee46c345e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "35a64d76032d1fd86c099c8b435be331d33975caa4f2d3e23e76c92e1477c32b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "6fa835d3f0994c52138e007307f8ec5fdc0e7f9670184ad2c86ddd3f26439c5a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "3788f6470b434f4463bc5aca8b0523264e6714a8e2b163d369db11eaa4c02088",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "f1ceb34a6791626beb6a082645bc6b94108afa034f80ab10dc80ecddf4e84130",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "e731c053426b2dcac641bad513a43a62efe698fc8b5785c5682499f8bf5b8021",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "212ab97f482f032c39284dce4ab3d06fcf3956959d9e3b83d15233c087b8a966",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "abfd7fbd4670902d73f2c6504f06e0c766aa07cee19a786a3139f7436a04d773",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "1ed7ec8a7af26ac64ef8ce68bd7a3009c5a5e4329b2c0ec6a17786781400c885",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "94a6ce7d031c6eeb28cea9d3e7adf73764e954a35a1b836d062c15024a59cf83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "786befa715044de326fbd58c7057d5c250729276a428fca2b1ef9fab3ec32f7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "49502ea03d2a3b5ca551eadd9a807cca472cdb2ab17b5b88c832a831b0fc4068",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "054b7bbc5a3f75f58d6fe50f663d71046b8396475f1d2c6e91469c86fa0cae4e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "bb1846eff1f20d3bba1dd8a9d27597be8e89692b77d21be063c312cc1899e3b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "fd9e8b7e5b7e34a4e8b981b061f2ca9adc49f42d9fbe8a9e375b00144fe21c49",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "bbeb6192e93f94419fbb27a16f907534f644c21e969a0664d1b4a101386bd7c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/codegraph/claude-opus-4-6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "898d02f7bb215fbbcb49c51558dbc385dfa3e36c29ca9369104c553971e24c40",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "23738a7e8d468775f809b0bfbc421a10c058e2577da9a28a19bceb36c3f358b4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "3d66efc5b1cc1f764ccf8ac28f54ffca04801973cb7afcc3182b5ba3d355e496",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "f6efb38793498eaf081727ad1b588811aaaa737d8d34f435502aed8289496b91",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "dd5212be479638fa0a7bd49435f7d8f3d6935462748b1e52ec4f79eddd67020c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9e6c154f23a5f63574d27c74c4491259d70fc3a259480bdf52962ee487c6a9bd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ca453c3665cea9726891f1c9ef4d24037d76613f6decfedb62e15afb9daa2765",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "22639000e1e2aecc92d6b6a85ba9730d96a1c1a2edd5dd517ef7a738ccd2b191",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "0a39c0a847c904d81ca9cde28b06e47cd753193947240a1d33a9b7dc2c18c9e7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "0eb96f590e08ebfb8a7cb417a331a68193d16387dfbaa9e0e77b0d77e81b2270",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "cbcc7e683d5e92ff8721178fb68dcb03aa86f382f3937f827d59759b2fd4d415",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "6eb181b437cfdcc95374216d82e573ef3a9eb5764c0f4816e1b5bd384cf41c59",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "67cc09fbe62e25819f457a0b9b674411708b62ed761af8c787d39c9ef72a769a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "1e85d79811e8908cb6ceec8d348d7399f10c340a0c0a2960ccba60ba4cac017a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "30e73a6a3f6ebcc8e02bf3af2efac9afe7d9e8c95fca65211ba9fe5355598bc1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "ddfc1d99a22513fa40e1eb6089183a135da76bb8b4fa5c3b145b812c6f99b6e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "01cd4c6dfe1ef040276dd75a5b6b3c54930d7a502d5dc6f2a51112bd15128b6a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "b062b0a4f9134eeec5dbc0aecd3f43d4fce9fe7ecfbd7b4121f178e24072bf9e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "16cf0f42f9d5f3e4af3b00501a76417914792bcdddde9275b4040bffbea08a99",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "34bd71ed8a8d3838b5e8139bea3e9d2c699623dc3f8f4df3072245e35d5b85c6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "5b52839efdf2534e64e66a88c232e65c71ab5e75fc8b2f02af54ac9bfcb890b7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "984c15fb5b5aa5e7c1926cabb7f28b46f44f8a0ba93f24c286ebf1c84b11873b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "cc1633ffedc0b9aeb1926daf7d404ead9907c661008830b2299005a6f84601dd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "e54cefa1583aa3812b86bf29c7729b3f7cce4560ccbd891fe7f94d425c09916d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ad0e1cee3a30fd8d45c855e8573a2413f322c2c457c6c03947032a9f3cd6620b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "192107f6dafc66eca34173f7fa6e61bb04117f3603e14fcc6d094a53548b41b9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "c1d4e8691289c7305af86e6e28cc82929e2ddef6370cb3032960d67f90551faf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "b39916662f47fa5d8995f14f3fe7c73eb04bc4fcf927a58537ab07c16b8f2ab3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "c34c837c012237f045f877f1adeb188062a2b0ec32eaf1ce9b76a8f9cbdc3d67",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "df2a436ea10fe911031668111477ec5902ca38e7577e6dbae8811d63297a4df3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "31528f424bfcafe71eb3b572f012322fcb060ad3cacfff9d31eb60a0b83c8dab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "9886f56ad37746530182200d15506525aab9cb8f45f6e1afebf951d103243715",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "f26034e05cfe06daafd53f59c5a10071309edf61b7140b4112f94139968f8230",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "2dcbf09a6be31be0570f6c648011f833cfba25ab669a2042e130a90b097356d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "be807e46bd22083b43d8defbf5a7cf0f7e4381d6f27c145e376e7cbad2583f83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep15",
      "rep": 15,
      "run_json_sha256": "917b4bd0a1c33cb46f4a4706c31067eea33218ea954cf0dbd4830e5950b333e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep16",
      "rep": 16,
      "run_json_sha256": "4850c1ca94b57acd56059309f39862be129e8fd0053238484a1b6d85575e3cec",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|16",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep17",
      "rep": 17,
      "run_json_sha256": "7386ef19c3ca03b8e8fafba55200301cf45596ec18be6dab13f937037a772382",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|17",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep18",
      "rep": 18,
      "run_json_sha256": "2782f743a3d1794e42872a843ade9ea9f99b3977c1436ba6586c2a3cd0a123b4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|18",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep19",
      "rep": 19,
      "run_json_sha256": "8e5b1368f8448c44235b18785725cd323cfd0a167263ab70cfd19c66da360d0c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|19",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos-hard/rep20",
      "rep": 20,
      "run_json_sha256": "2c6fd4b74dc9b519548235e67560f7064480544cb300f22e4a20f04a2f2191df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|20",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "26671481b5c88d88e0e61b1e3339385bf007440a74f837be489b52b0eee302a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "8670a4928bf85c78740c037304e8c4471615748a41c50e124cbf98d3bd64710c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "ed32111ebb309b35ff53b48457559401e287a7f7de96408cb145b189f5687f49",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "ae389b6b18e96c0c3967e79e9ca6802c9da7513d1d73c50239b13e40178a00d1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "5e163c697952254c08fa43ae63b8d0b084138565562d8c08b62412d76a05aefb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "89530a007e6e19d5b401383780a9c9ed3fbdc75cef0dcdb4babfa11fe866f5ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "85c468fa33234d50611d467b359e96bcb277cc35fe3854a2a4925aaebc5aad80",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "04f0c777e49939fb6103d797e5b820633658c0ab52f15f1cbf33bb9a4e2ba90a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "581ed287ad0feaf83c42124dae8bd65422e71f663b7a0740155192bacf019e42",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/codegraph/claude-opus-4-8/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "39656698e64dfb1df112bba56be8e59a5395f51a31388baccd3d534409856ee8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "6429326cd4b181d4465bacda50e1e7548c494b6e222fc3790b10897937010b2a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "50d530aaae143696d89edb7abf6ba1a1b15a184a09366625c10ebcb8218e4c10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b131cbe3553da729990ec705188cd68f3f3658f4cf8d0e097171b312414d7c30",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "64c9283b75db43ee4c2065e88d5e8d9b4e7285dce5eb0c38e947af542e2ef3ab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7630a3c7654ea41fd43f5cf8b7e4ec4d799f83bcc874c34a97a3a5800298d3d5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "4381c9ef8b975682b27238b61451a48681846e920d2b88849eb63d546614fbfe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "cef8b489a93e05de4507461b4339f1055edb5a9a52faf93929a6a8cb035e6340",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "0dd26a1090ca40e6026acc8604ae37d9e3754283e62aba89e2a26094909dfcba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "0640ca8f7ebf4abbc6e753130de21eb8ddef55af227fd042f9ca9d2e89475cbf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "14190e633c71dfaf167f170e1a7f36ed3698bf92d21f8fad682efc2dd130a30b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "b5a5894c0e3c39c993aef513f07f01f44736bcd69d29145a3dc5c6ae10a4335c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a519226a9a2acdd9150d420419225ade29f6430b7d501e25befee01688e22db8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "a35c2cbfe7f1557d4a21cdc2f206cf57d9dd96a00b8fb0574233da519d442bfe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "4d0fd2c9a5c770b26cea9b238c006608a23c7dae000134def706051b2b4f45ed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "46882d3584395e1a3c8eadc3ff2147e3c96beecbe43692a8fa07065df5cf51f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "e50547e878b03c9bc51601c761a69b4b97285fd77b745230fe0fcd56cf492152",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "3803c8c3ec13e9b91304a457f92b6435f5319f6d51f9fe5a3698375d32880dc6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "b85ef5d32f066e239037105b8d8da274eca93e009ac4632c56a58cc0069aba08",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "a24ccd513b7d23fb6997de03477470dd92a57a8f8d63fef6cf38fb9b0a465bce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "142c8fc93753afd6a2661e6304c7e8f4bb3e56714a1a34baa09a5758c421ecdb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "430c40bc270b66c814108a1eca858df57b5bb8a9ec2b1e95d957530c09beb135",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "d3c88f0f82a645ff5590eb1e8a6e1430b0b62cac54165ca535a5182739f61217",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "39ff2d9de61a8d63a517a7cece2ea681cdfc6dbc07b62b3b1accc5dcc9490b1f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "3f7b3817dd2aa6e69948043f8abe2adfdfe3c5ec399b4edd076635dda85a272e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "5ab0b80c0f22704d954564cb1131a3d63c7f126ae0d43f28892b737be76a4629",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ff88eb40c13178b79d6e1291cb7577ba2f4a455aa3f17bf26eb5d1fd758936e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "96d5fb12f003d60f0be2bbdbefd16181b983950a8bd22337418235c9e21a8c3e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "9a06f1930c77b1f778d3894f617891c1222f57d05533035b6d11284f79ea58f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "96e71ab3325c6f2ffc09de978fe31be4c7e8b35728a856de3b5cb984497a63bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "d46d3e26ec9beb0aea89906b5c92961e4c8cc638f7e7a99baa53fbf1057cfd0e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "f1b1277709c1c9b9c4a96802845977ebf54aaad54e31a5b9ad1b757450534952",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "8f508a5096f616cd5211f41f878245e7fdb7fb2be51c738388ffff93810ca4fe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "38deaa406d86dc80b5eee1d6be2b64e12a4db5c3f4e5fd08d192d11209963a2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "dad6345ebac7b7d6b51c1e16c06a939c9e072c9be1353cb43a6245c55156cdf8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "dd026c0b7ad2541e30ed4f2b952f5e6f0db9ce2f82510e01ceceee8134d83428",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "913d9fcd8a517df5a9342f7ca0ec9c62bcfd4b27d32685996e95497c7eda466e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "6f7a17d67d8f526280bfc715ff9c9a36119a722f19d92f21e7df73db5d7fdd37",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "3f213474e3cb942ecc10382a4d5f354c6e7d1757169673573ba8c0c316ecb628",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "177b0fb45856f66a44ea4bc71d7267d43fbc4b7d3f95df68a5575a09bc54d39e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "7a37e5108661a31c79503b764740ab780b16e51340b163118b6e0b6209361801",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "a475f4a34dc9b804f4b583756697c7727ebd2030429636cd258996db2558d7fe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "a6775c6b2dd3c824a2813550217de1f5d46c6c416c93ec3f5d11a3d85b265673",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "f5aeb8463fab1811214f937f63b440c969baaca6f59d07c7637889209b116390",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "c76d36d3181fa16873e62c6d8aa52439542e2b11e4823013281bc501e10e59dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9d7ebfb9dde6d8d913b64400824862e38da2cc0eadcc9f5e6176580c08b174fb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "b9108692216d9d05b220a122c527909f38e770a22c73b39c4944af5e657c17c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c27f77841b673852104566dc34035e1dfef90256eded2d78dc5f61b9d4a4de10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "22a48eb5aac00f4614d02c73cecc4caaa2c105e6026c443f3b92c36ec0aa8db8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "72c12559063da18ef4e824d8cab3da5c21505bdc3f31aa4bf5255dc2061ce764",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "f8e3c2884812b78f7d05fb7809fc1b1f581d742cd3d4a02418e5964ed3f2b162",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "152f415870f2282b710981ca90bd8d21a54f3a1d3cd44c8b6b27ce36e60823d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "dbf52cc062773abd1d86dff3d9d32a075b7afa84d8ce4b75037e63c9241502f6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "8269811cd67e44d84256075edf9be2b412f2de134a91b48cc506bb7204d3f4c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "afe4c6c00af1c1251de56c94f07b3febf1e2781ff6f3ec1ec2afd1413508eb77",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "4e3711693a70fedf732c18c3c14d0be4c8b93ea1c7e85892934acbfd8b2f1ffb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "988d1148fb35f833743afe4bd751fc5aadd8abf94607684bd2e2fb5367801571",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "5814e28cce2f816e3876cff2723dff143afe48732fa4e741c1d16f578950b2d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "2e5662753898255018eb13c06a31fe1483286774b236595de4a20e9217b9f73c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "99770f50569330f98c1dec6cd3ec532aeb55dc73af589e78e4f2f4d27454e176",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/codegraph/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "4724b1c6e66119990b1d9e1f917b74dca59f9e11d1de9e19d8737dd9aae50aa8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "01f2e048b436cb43659f3fa09791334d56bd8c6bd0395621ef1805a6b7bafcad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "e7dead32b0a2d5115ecb2c4387bd31c8020f9e1a1eb6f412f96ada68ba214769",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "a200a5f3ffdadc728d5e7d44bb84056b211dfb40992cceab555f4c3e20954168",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "7f7476926f23350443cd6de21f94d7c9ab390cb6dfeb9093fa1cd924aaf4fb38",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "3d5cbd21f8eb53bf4851292c8e3c592d11516ba10c99abb7cfc46f8d32e453ed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "196c30a82ed84671ea81cf41b2d77af8d84f3394bee5681afdaf8200b461ee52",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2604ae236fe9b97ab4aaa3875211838042a575f8218be8aae283140a495ad2c0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a810e3de56c8c8f03c4dd72b64c09ee9ea57febb9e9c9ae297676545e477bc09",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "bd28f41438a069f8568723c7bb18df5348347af9ed44f997346ace50253dac1e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "3aaab1b506c26dc4ea09fc17386b0de5458d425d8e5efd5529b74150a5fc6c93",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "89535da1b193f32aedb6c2723e1884fa9c36714fa34f7e730b3b0101ea4e1611",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a1cc8fab4a9329ba689b6582b543c6fa0235782549daf09b5cd4e1d951ba7c9f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "77b7783409a17adcc7a0027e2e01b486fab54a1fac935d2dc5aaa8c5514ac557",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "66a01e71e9f6906058dc60c6865ec2509a444607d437c892f4bbe39d0fc8fb36",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "eb66784f553e722b5939f77c02a3a93a4c9be10a333d4fa4a23a8d2a408ec910",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "612ddd48e8f5198c4a7b3cf32733ceefd3686b2b80842c5009c84c516edc0d82",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "a0937cb24f40faa1e8629e4de68da9c118e48b6bf43534ff7e219f05725d6194",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "294efd8a3c532caddd9101d7de3deff4ef5b68fb391d7c3ed6fb313b4353b548",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "84a72b15fc04dd4041b82495cfb03eb01945e4bbec98f7457242d83448d64e4a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "328a1634b7610beef67ff2b9891a633c4c62b5557c78a964eafe032f33d75d37",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "fa833f90f28e5755e5a99d62b766d703e0262b38ac2ff0016f3b2f8efdee581f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "8feca829d0272cab4d0b80c332133a4993d158bb1eac7398cb6cf52318d74ac6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ef02fdd5c5b5309642d426aa70bbcc1eec1e551b16389898de66bc72e57757f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "9f8af3a3e19dbd6e5bdb8d3d96a5002814bc4a2c52f9681fe9264a2062a52e8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c259749c1ca0a29ececa4b2533e421355d43e41bd09d26e7b9cdb66d56e96c12",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "d20ba3f6f018c2ec82ca6353f1da263a4d2e9ed8cae8f36b520dfd6c63ea2901",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "b50e480c536a21bde8e06513ee64528721bb1e70e2fede0df15c5b1e02745a7e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "6d10f08330681853cb9d14c7f303924696fec8923f1c036ca87e908198894969",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "1f55e221e3e0744b9f102de70c885823eb86058970609f68663ad07a137f54ea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "daf338d7e7796d949f280d0576e7837808191ba75e292a3bbf765b7c2b290abf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "c2a6d5c67cc97d0f9854be3db602b8163b6cc89e59efa0b50f3fc58f563a55e4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "3e4de1b86887b48f58e94dfffb794f41130ab8e3e26407b77cb8a4084a64c79a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "39216b422a4ab6041cdffd66f8d380942359ac2546b27eaf4ba6f9eb6e24db6d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "a6b4a464e4568ef94089d53b8d078539e3edcec146abb2159edeae70fdf5544a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "f548f3de688fcce314d238b7402da16eef69c654d26ddb65d0d7aeea7d8a61b3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "2174f490af028777d803d321dacf79fc8f148a044a61321d41cc1f02dcf61f11",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "bb9dbb8f75a745255895b8fb0a810fe6f96f522dbd48f67d490e351639288ecc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "078d51ac59a6913731d461658d2565f3364700bc41f5d32434373223b2c4d8c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "c60e31e3631f57f650f943c5378fb38308b5ddb08f020f56555d70f12765d78e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codegraph",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/codegraph/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "d7c4fdc9b2d4a02275402c387873283b22c495aa3f942dd0cad9d51ba3023ce0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codegraph|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "3075fae9f63eee508916ec8f5c2603fffcb90ebca33c615c9cbe08a17b826755",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "7dfbe3ba7a324e0fbd68c1c3757700d3ff9ba9f705ce7c2638dec96b747885a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3e4e65ff2305887400c2c4c77bf0b59c44d3f491e5498f4cbc32aa2595940ad9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "a4908a266bf30266234247b3508f39bd36b33b77dff5c084d2b22a46abd74d1d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c942aae15b9e45077a608a563a8251966a5c7a26d17245a6c296fe76e4bb3ebd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "4ad2c74fc46de738571dc34c79f8db9be9c3d00143173566fdb192bf24811a99",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "b2e71a4b66129627937026e7ac39a85a79aac32065778a727d25b58ace9df9e0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "e8cf7478a701a905af9b7d30a1d88f0f8ee83bd91f2cedb437d09ff3c0b5b953",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "65bbad6454be9cc1e399e2eb1df8b84fc713f754969b35ade2044b8cebcc24e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "496c1e720c3b547abe182ffef391972e41c6714d86b4f21f31a83133f103ca52",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "f411e870a27b1b2504a5331e720429bb55247bc39cae342884625442076f3ddc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "690e3cd9a750ba77690e8ace877deba302fd1d6d5017a8658fbdc89186f7bf8c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "c3976ae9120918e191a509a1b0644daee8828aa4990c68007359f167ea7e22f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "0cfa851b865e3915e06db7f110cbe09c31622c89f5dcec56c13e351fe376bfdd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "afaa0afd339deb3eabd8dae5b8fb2373c47b9e42c9880ad3e024978f52ba4c8c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "c0dd5d06e68fd334b952d7d0d082553c5a3d8daebf9f9747f7c3f13c09bea0a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "c583114248ca07df18beb6a97756413f3231739408ac652be88dd4d275968829",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "3f0a19a2d877549b0bfa16f7f0d9a91d304643c9da816c2d83991f105fb5c24c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "6bb6f6566d0c5cb98f92c23daf46ad2c5a3b6c201d84c3edb5daaccca814bad7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a318802a9fb8a78fa017c9683839c073a5362324c7945fbe3ac5092abd79fc27",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "a3b53bffb090395b8a47eb9ab6dc76f73c34099e3b74eb59625f2619fb3b0391",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "61417fbdfe922d216f4aa608b053431f0c30a8caec9c1b3f359fa8a3e4858880",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3a210a8c73ce3cf89a9637a3baf6d4629688522665e6f36402ab86d33538decd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "e67c2691bbd39763040dc262a50970f1f78b32b9547e5ceb98bdeb8c90e476da",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "d0f868bdcedc33f6c77edb597011d7c5ad4a2c980f0865618a59337588781898",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "06f481854631183a4dd22fedd1048299187b4d998c48b20e13af5a448ad786a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "15e8f90faaadcefd90c55e618bbd21ad2abba16be669ef5d2286d8fa975e3de4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "d3cbb40ddd26812ebf9420786136d902e33987d5dbb565dd21b74a173f83f357",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "ff7d0de2657c43d3a797b5f0cafd424ed1a6d39053656b7669d9c287dbdc7024",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "aba9273d57f34c4b7bfd77501cbbad6f4d9402c39332766992c37c45cfb35dd2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "20e8193f1e39e60834f1b6982cea331f93f935f7e98c946aacaab94dfb1d39f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "825706ccd079dc523d58341f839d5db507b929cd2eb759bb0d342b056f14d9f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "c058de09f33417ac67d092316ebcf09ae8ba1062c430861f7160b9e5ec14bc13",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "9fbf58dd3d0275d163b2a264147917b3beff5e3ec55e89d21893de0c5a4c8130",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "19ca24cce98ceeb9b0166d259bfbbec1745fb86a793152630a2435c866289c65",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "aa253ddd8a0ab347177955b15f651d5f656d9535d364d7933ebe3f1dbdb98364",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "2ca29560565ad314d92d32c967145e9dcbcd6d66e6eeb26342d5aaa8f3247772",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "472265d6466992a90be9fec39811c56bef2c70eebb8586cc91e8efbf417737a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "94ee1da40efb60683b4abaad7284196309d0e305f8b03724234ff19b09ae42e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "20d57459a9a59f4c29f41d5928896effeaff52d47be97369d2fc996b3f24cf43",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "c94dff7ce81149fc5dc2b27e25305f5baa4194058ce1cd721c08af1d004bf0f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "dd258b52c7fca952218fbd228d4ed15a816cacc11425250877521b2efbb7d778",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "2f1e04671ed94121e708df3778fbfd913c7a41d1f596a2d732121a7e2057d823",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "a9c931c2c271006709c85c54e43fe4a0ff2a240ef1b4130c1dbdf2fe2bc15647",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "6ddfe87e86b1239451ba96139620f6b9eca96155a35912b92a0d5cb44af8cfe0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "f34e939c8da664ca7e9625ccbd756ad3cda278bf457c81149d3e26ce3e7bad40",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "cd348bd7a42bcf350ce7cb2d293aeab9bf2a5b0c753c91761f6a9676631fc47b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "62762f276e7ec7b5c7c89b24f7d108e5f8dca2c041432f002465ef7ee8263d9b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "f19c0f6edc99c800f4e4124be12da2501cb086f6a9f919ae947238c920cf9857",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "19b0789a8937aa21ee320927e3d335dd606249aabe8d3a02bd1092f93af2fd80",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "0a1331acb94eb444fd666b0a0e476e12a61f7ff6ec55adccde0141b8daacafa5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "07dde35ac6c5b90bd5bb4d16c2dbe64cee4d37c601f8e1640fc1370ee9135eab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "cb4bf21b45f8f3ceddfaf062d566ede0903528200d1a09671c5950c9ccc79d25",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "50b4f32ce5502137543755533da11e2cb0dbbdffd25840f1e4d937283bf37596",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "6e1b2d0be472fe64cedfb94e7fffa561f3fc097084a8b3299d4297f365f03135",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "e0eebd96c26e1ad1a6a0c8208fd7bffc094a9815d96e3e2d0a865e446e792cca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "7d6d2af861cfb949c5d17d10918b294ffc7368f5f3fd1a13eecb8ff509a36a8d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "62b390b8412ffaecbd950fff1ce00f1c588e483fd75ba70fc4e2d70899b13ec3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d0e44c1c9490b89a325beedc84c366f6e8d1af17a7386db0056a691670d8b969",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/codex/gpt-5.6-luna/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "6971f98e187f3c1c810fe8beb2a79af1e039b06e991403be39dbcd266474741c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "27ab097ab91fe458f1eb52a934e04eab345f8b02ffd215446340b668140b721f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "784e8872cc529fdf9ef304699c30debdb65152a6069b48638cd2e427754d9169",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "5fe575cca30f8c81884438a70808774ab113e02c2b52fc48d22bcb80273bb94a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "e5a8e1b1377b84b8ecb8e076ca302f978214ad9670820420bfaa453250a4152a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "6df26487b14cdb43db94bdd93606fe931bf63eb472f6282f3e251187e78f410a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "471b29a0467077040823c82701c7c30389be1a752f26a7d4c4259f07d065a826",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c045b1f113232bbea62eebdde728c5f9702dd5413a2e657e97fd1957271ca544",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "4da4c2302fd76c2ff31904ebaef1b5e26c6ac917de80a40f728309f6020eb577",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "598f6cba6b8a57739f37aa715025479ec35360ddd0ac61273998bc3b6c65aad6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "5a4e534cd43c7f79b003de64205745a50ab40cc5b9b3c347e73f52f3ba646669",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "7f4a56fd9a4eb58c65d4f0eb4708c2e323ce7616bd9e5879c012636adb554008",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ff41799dcf84055063e83b185a3f39788df7012ac0ef56f4446612562d1f7e8c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "1bf50bb8026f4e3df56473866f18c1d56c209b06f349f94c2cac288720aeebe2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "a1f67ef5fefb5bc152fec73e040fd56d3e2fd18b58a4b772c414605bdf4ece14",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "cc9e8873b8ea50acf265a59621f3ea5be2d5545c855e94e58c5d94f982647854",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "beea2d3cdd9fedd6ceaa14faee88fd307641b94665b87767ab334e08b19aee16",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "8534ce0ba13e49a78e710f08a2b791ab95b22cf77ba7f8f2e67ecc0dd2b91bf4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "cafb142ea5234ca6e68418c944613ee7d7ec2fc16605e9ac10f845eddcb32aaa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "369c215dc87434f40995c3d25b45d3d117e009a2336e30968b5db99c0db0a086",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "abb7005a16e7f687b4c9288355d48316a358b67f5e8711784a7b527b16a1446a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "5a3093c3f7ac902514e4fb4433e70b97daa27a660f3ce28a1921dcc05857489a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "9beeacebeb2ca8d2e7b8b4a26e7e5e541c29c1ead79d1fd8163189af021a0472",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "4a7a7134696cceea202205d160ab6464eda211d06f8f5ef60885b87ec483c49e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "fc7887e150ce38b12f30caa23877f9cdef0bb5b18067a8c09f036fe4748d0241",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "0aa90098724a0110470168efed51b917325e5788cee30032be924af79d3091c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "58259012573e0fbf87bf9a5dcdd71d0bad529afa5f2839bc150416803ebfb028",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "b398578c33be3df187c809c1efd94be221eacd3e83bbb1e38f96a104ea22bf1f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "69bfaaa33a6d6fdca7e234393f4320a8583f778b2072e69bce819bfb42755206",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "c00776ea0e6bff695bb9127d04c037dedcf1e605af2ff4261b705176cce84e0f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "751ab1ba95c20025d13b07fc0865eb88f3d43ff612af6029584dbc29c5ec69cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "9b903b8cce1343762246f64baed4d63d8afd67e9a2764b09dbe64207598f72dd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "26e9b7cc0e76f1ea20e092474a5e8a299df6f28b88be7f1a4dbec564e58ca5af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "3bffd3bd1840768de908a79bb872f5527455cfb39d31a2e9bebe6bed82fe3994",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "edffcdd2aa4084ecb95a36f2cfc0ceb58dea91c2e7493397f04e9a455637882d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "f52c1d4bb342f9cc2b19d7710077beb624dd1aa617bf4182abc9b24dbb32dc2d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "2af560ddba2100071b2c5aad685c42a7250633fd5806bac3fc10b28f424b9463",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "833bcf51030adb9560a3adfb7d4642a096799fc13ff6b28244249553fcd9b172",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "97fd2796703bb2d7676d6af502a18bfd70df57af9b66b789d02638d08d5e139b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "f45016ab22c29f3ef63dfd5682095b4da2949b30efb3f335a7809a25349bcc73",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "357c7c3c009161e0f7261d9b2ebae05fa02ad84c9f8d8a7ecc07e208eede81ea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "e8267f2329a87f7bbcf04a6cd209a9206584001bea72b17d9f2b774873b94fa0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "3a743334dc6256bd11cfc691cdc0355056f6832c577b564481858cb9953752a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "03a1434e1b8bca667beea76e9f59090c08afbf775659b9df8a74cb8bcfc27c17",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d052979338f2e9955f8dd8b26e800f1dd387bc415a3c699ea8628986a6f243e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "ce9172740c8f58e0c1bb0255e98ffec85f01a5f642390721e96aee3a6c0ab3f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e7c6155d70c3d57586d98e321dfab41fef600a803714a7f7d19ba0f7a27973c8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2e45f22dddf3b79d01dea6c93fbf6262821304984d8a9880a8d53c86dc8591b3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "0f6f9450d08552e4da26f3509977f06252b1cb5774a6789df98a823411bc4997",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "406258aca83c93c97dbd72e019d772fabd9b195053b0e01d1090ed54c92da124",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "eb0015e38b9bedfe3977da4cf17380f85433e306ce02e47c6891fa9650a131ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "39b0c4d9f9cb05d1468ab1e52e545bc3534e9ce04dcd9dd8a79e150c7a6bd8ab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "05722226eb019ca89943bc3a98e9f8ee27095de88b02fcf678bedf3076d5efcf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "64bdbfb9e5084d35fa781edca704ceaa35d3ad8c7f1188e3e076d25cb8cb289c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "b196f56cc4fc214377948edfab5634af485d8254d1f6ef7cb4593de411e1eb46",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "b8fa4dc9f2b2a38da69846453ab3a53b4c110a809e2dca686350d871342e55f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "91620c452f7f059c321c02d38b90cae2e2ba830df4dabd16c6235bff47c8b316",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "7ba77db2476a0fe186201c765c667dbb85f5b06ad3ff203eccb574e3cff2c5f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "d8ef67513f67fbbd8b6c2953708dbdcec9409adf8a2113621dd49531bb752d44",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "090c25e6dbbce7fa3b1408685af9fc245ff5463e51427e49e04a7e9950efc0d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/codex/gpt-5.6-sol/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "98091357cb9565fca6dd377269666216f5b5411925f9545a90fc89c584934511",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "02ff934e61ea178bf59e7cef7a9fa8c76f628104eb0316164538865fd1aa950e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "3609ffe6a46fdaff9b347a0d7673b2e813fb3cda4348273cfdeb057bf1096d32",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "74ca2dae1b39768593c74527feceb0598978159b2129f584b819b5fbf0acff9b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "a7756ebd0b5768c9a6f3a76331b15c383480574d245247697c3b735279e592e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9b33a50eea8af4e4c8fb6addf18961a06a3686a9e0b46d0145112a9dc57c45b5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "03efd2e4894b0d3e3161b5fdb8bf899fd3aa1ba6bd2635e88cabd05d461793b4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d96999185d83ea740f9509eeda93f0a4e5be5c7fcdf73f1702d2b34374cf74b1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "2fff1b6c77dca8f7f706d556576868cc94145783d948d5e1897c6c0f2a0d4df4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "313a4c617c69fb3b2aa12285f70cc584feffc55ef850c5b4e2b5194cf40d0d5b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "845f959ce604398ccaf542c9b393992d40a4b76f4145682119b359e45af019f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "78f59d091ccca310cf7f1ed0b215f3199b23a7b67abc40f4cde78bc8d0dd6ed5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "4317b331530f613c5e2949b43d465d485b0684d8a3a57ef39d44b5eb9bb3205f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "b61db15afcf90c879f423936c4a11b65ad13b26ecae51c1bf0ee36f22ffdacf7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "21bfaca2a08965b8951efaf801dd4ad6426ff1b8ad9bfb6d26cdc335fcd44cd9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "b3cb07464af68f3750027cc6bcf2a0e9a4c4528bb49f761fde6c5087a879f7ca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f751616a529689689bf7434ddb8fae0196a5af388d2bf0a785b6082557aee37b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "ba006a7ea9ff7cf3045f72135526a8dfe938c3c0cdc7751cbdb05c937059dee9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "ee6d02dcfd732badce66d6c869017bb78b0a50f34f46a7b49d0b11e5db9be807",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "4f0fa73404876e58e6c4e53d894a40363e8731f8708ee86433784b5928b34ec9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "79136a11299fbc771ee59f9d75340322e43a30bb100fe04ca7da4c3d4482c70c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "3212ce80555f1c42ffd84829dd351a1b7ceda58f2eb344e468f2b58fd0f37b10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "8a0958ba114f5ed9ba78cd866c0c23b742d82672dd3530185067e66201b52d49",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "d5a993b7477b93cefcddafe8fd1ef8975311f3a5d946bde3f4a7f15c9e6913f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "c86f4d358b285e5b23224587fd5db92611ae332bbe9d231661cab0f13471a44a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7cce5513d4e774bfc9c79e0f9fa1ef2e17b504e9cb1cc38f095b08acac9b0615",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "75b0969ae8835d8b9cebd553031fa93ec0e5b606709cb5bae6ee8acb2461b8a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "60aa037436155787877428feac396e0991ec5fbf78c51b96653ff2a67c7edd2b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a3b5d8e1d62dde14fe59c1bdd7fe6e0c9352965f08e379433cd94c23d3547472",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "22e0a69292cabf277f89aa6f428d6da73115bce0c9f61fb628ccd5ffc1f1edf3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "6d47b3418a47c3295ab8d923a9ce7e5313711c8daafab66c2be88fe0f5ee0dfa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "38b43a8ddb6ee6e1a32b9e62af5568fe43a0c25b60bd3ccc630d487641e6e0d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "30031e6246aeeb14d33a1cce64e540e21db3dc42d728a4d27beb358628a1ba8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "df8fa2947892879e1edbbd15db34f8cde1ab729179b1c467f1a0b60fa6562214",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "d33a795528799bdf12dacab96a69b0a716e3a6f0f9e37b3f72dec6c2e00238bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "8f123b868408ce8409891f38d6781ba8cee9fc5c214ff02366f170f5e4bcc8fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "1df0a46d433f684a35a5e36963fe0a44daa50126c23e4bf291a151b98021821d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "f432b9046cb2142a7b5008551e31d0b804f0fb5c8e355c34cb4f94e6c9de9265",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "363269817c8a3ce8eb7289f5dacef3ab0ee96378aa3b57d7dede703c6efb569d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "0b3902d4c4993a6b9c4d1b8820d94e409aeb4d9fb6ac00cbfb82c7e712c466ad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "bdd84d182ade45af2c62a8cce30e884eb78d6d56d5a2e8262cd3e781da3a48eb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "2d3d3d8b34ba1ae610c3ca9674b5ac11bed32bf787a832420368f090d53434e1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "e77cdf21d6f3092cb9fbf2de7988ca1fd5b9ed9307202d88f568fde7fce3568c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "160865b3fdcb30b79e5ac4679675e2a5579e66a77b1ebd3662bf535850f1e42c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "5dd4394cfe5e73a7d1dc75ccdbbec22cba39f20b771593b4ea5fe1c5bace6060",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c1692db04ae6c178fc67bed06c5776144b5751e82bdc214c90a9a0387bc78b52",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "5c0da5d19a924676d4362cb3eaa67e274270cd6b957ea8945a7c117ae721fa8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c48b03294acfd55d19d62029ede000617a5f9870e3e4e3fbc4a25f71011e5c18",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "014eba4b9a5e4b044982cc7dcc067f37deb651afa6f8b3d79145bd197708117a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "3b495929ce3410758961b4849c56f4a3b991996c6f9a3cebe46abc5f97cdbd9d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "09bece37bebb76b98433f3c55b790be0a45f65e015713f71c9eaf46904a02ded",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "efc04ac79d204a5961c78110ca6a5aae15ce8c98f5113d770a583231c82ab21f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "3bdac2a1fe2e7e590289f6a83070e409abd0d98ad2e95ab8fc17c4e7f0b205a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "db0d4aef91323d7f7d2de403160314374528d827b1a82f9b16d53bbb6449551c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "36ddd3b62e103c43eede53b1d781a79bd911d87a240a0c298925c196b7aad044",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "b327bb0a603356b019426cc1e58b1147e4bb619b94a98d5e9817e2bf573e7903",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "b44db19605c6d9486c0ebd9130da7e45ad71999669e80cb4f87561dc8dc4a52d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "3e551e1c75cb2219eaaf38b708efe21d4a200c79ecb6c916ebf9968c109368c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "4c3528834c5efd41404c04a1ee3210dcfc08400f0ab12e6e5d15f0facb819504",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "dffb184ee5aed28b5116c2003f5b2a0f5beb95b185a0a903fd4f89da6639f3c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "codex",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/codex/gpt-5.6-terra/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "f65a6f492103179ba957f0c506ba46e146ea75262d7a5978f6b51742e827d0a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|codex|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "534135bf4128c2535062372e50e488021a629e181738ca2b7cccfa839420661f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "d6f88b4130f8d40471d872fe0786cdb15a69bbfe3bb5275c4fefb3d1b0178dde",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "f08c29fc2b65da55bb3cd11f480f4f6bd762e7354a990f1cc378916616b8cfb2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "978aa6486cb6de8c7988ab5565bbdfb235773900b0cc6f3096cc04e7f1f7a835",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "572db97f0bb8add44cee95af4d3e516bc90a54f8b270d62abae11fff448a7df0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "6df7cf5e3d89495c0a55a2e76fd457f1c30ca56010a6cb28adc9b46540b53727",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "35113dae001ea6d51aee8125623943a620661a3e0227727cd75c06ded29efde1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "7204ed27f2d417c4e1446c7360fbc8c5a60b34635585b16a1517e17f0732c392",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "8ed181fed5f642e72cb0a7c8cccf47dfebe4e540d8f05b4c2b6f0b9f81ffb2d5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "e40a6b408e74f65d8a671fef0f92874da2c1b1d1d032cb52b153cfd4d77793c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "44c2ed59099c5bcd6c4d9f455051b806d366c9e67a9c5ad41bdd38a388850b4b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "11b46aed6848811ec63b5fcbeae3224db8c7e02667811609dcb7a5587d90fd13",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "017ded16015c3af917516eb99edd79315589354767b7720d624d0d4a72e01abc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "bb5dfe48f28895cdd8ed2a8c32dc6ecc0e35c46109604624080f80dbe9d46a98",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "51a2f6a451f4c2e23431635cfbeb9ef890be7cf656dee7a861825f0c70cf51e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "b25e17625544d1732c081e7f643366220b66ad018a2745c4eed72c5b6f375168",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "dacc60cfe86c90a94a5e60a7c50b066832bcd1b1989ece547bf81f5d07207878",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "8e27c36555fabce973f1cd69474cb67b56624cf95719fa585946c207fd085b61",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d6c3c3d0bb29bf76b7a2ccae5cd51c4c0bdbfb238199a638a606cd75e7b01a12",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "589c1f3c2f771e7e49ededfaad56fb2ecff9a3cf66947cf4dc25a47cc7d7be9d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "63148c1d2d9ffaa71250b6f6a32ec3c50b552cff12f882be78d7d4d9493368a5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "8d0ea953ef10d6151e56e65230b845363a99d616c71652e641f9e37711ea51a9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "0338b3ae0d9c889879dc36d28458477605d1d54b432c8c9b905092f97af76fa2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "64904e53625ac55f3bc5f56431216792836d0b53f02fb33d16e314eea595634c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "968e7001f364e20b093c36aac6d7a8abebd76a41390763ab66bdd21cf9c1d4c1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "c547296e2bd35211bffd3f926a8d295b5611a01d23c25457639af6ae4ff5e26a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "24222fb51d7b13d4988eda3df3ab8e5219f5e59ba0a7847f464dbad723e1eb6a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "0acbde2464818f73779459f4afaced80fb868c19a8dd6182cc670972778d858e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "9e6afb4b6c1b321e1c1bbd6ad3c552b510e912681701dc484dbdd4915036d477",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/cold/claude-fable-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "5ed00590672cff0e95161a388426889de4958d805ca53e90ecbf617b1b990742",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "56c340f0c46c6d73bd894a274846470eda672c8645e7a82d19b7d13b76b9978a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "c8f0613969a524ca5ed9ba4ef4fae42759c66a828db130c3852c39972c18c2a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "e6c50a19cc5bfeb930a9382e3d2167c52bba29b9bc7a0cc5ac69d1e243fe03f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "d1420bd880383c82bd74f4975c5cb838b3b798e9efa2252788e68170b64e01c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "e331fffc2825331eba52bb8d40ebacab75ddf950d896da77d09c5035698a559e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "bca72881986162c1a75a8ca3e127ba9750c20a13654a493dc5509e1e830b5adc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "fb9a3479fc7c4d2acec6a996f3df96d4065809f4dbf7339e1f7259cf6573ff75",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "2d4a5d9e62a9f6affbb62a0903541a70ccb471df3d0dc9af5bac264605fce401",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "309dba90ed9864e30e3594d04d637b154716e97bc45cf4dff42f63401b0cfa5e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e267aff9aa92b9fb592dae3167f0ab47293e760b777b134bdf0d8d248f678c37",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "ae3accd62271f328bcccf4eed1ddc82ce71585743caea7ad6f4bf9a50149d498",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "b7dc352ff53aed734315719635825769540d169fb859c6decdd392a438bdc1d6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "37db7bcf3841196d6a5ae866ecf6ccdfdc0a92130d1300b4dc98fc70fb71e072",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "abd99fa7e9ddf8bda48034311c83b734268f7a56b449c449e7410025552f61c1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "cfbd3610ec347ea003b55078e3a7362eb458838a7095dd169d315886b8fdbe3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "dc474199a5e4993914d00376e0bf684da71211008330a13bdf1a3f9f5890a435",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "60f971c8786565c4ed3a32be5677a3e35c0ba1c9d7d1875d3c64a526a211198f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "4b224854595697eaf0c29eb63a92e3c5cb07b7013e1b745b91e1cd6780e0cc0c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "17dc0f6f504006f0da7117fa3cc26f44ef79e6a67c8b7dab35fcb764b316e606",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "1ae9dfca306532ddc2da093ff7762b315cd25d6028e310e2151f29bb1b3563c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "f73b79e94c51f3b128443bff4ffb1b2e42351d746ee942b76ecbc13ee64329e1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "439379094f0f7020338a417103d04d98421bb885fbe2cbc4a73430a1e5d37923",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "9d47bb9d468721bf0bf98c80f346e7a65a3dd58d09437ad2168a8665c4baee37",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a0b8942cc8836be873dc4fdad590701e97a38a0b87ca526bf55bf9d3e73991ca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "328bbae4bec9495ba922e39fed15115b132c97b49fc4f0e41b747da182ebe205",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "cb34bcc69a36e424e5fb939eb85995b80d267b0060d031eda068195f17681ac4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "878cbdb0aefdcf1995f89ab3b90f24a9462584ab49e0be126182f36e17049fa6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "063140534d15b3b6bd7e58e9e04c629da927afc5e9490f4666acbf9dd1389155",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "71704dba120a8e027da6e58cce8a2165ccd2bf0ba78c0baaa6acedb90fa272df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "b94b6b8b7e996e8307786a95f6e0b1b21383fc13d646efb8557321050f693c03",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "6924d3447c1f449d22686a4c548ff74b69e2211f03e22ce3e58455291ecce869",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f515c6872755c854b55b3a96cb7cd404482c8e4b9f512f3625b7cd1921ab6e27",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "29fff52057ac23f0bb91caf42bba3dd725caa0d6ecbf6f49d0ce28a49c11531b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "8edbc09dad5abbfb1c8d18dafe23b877a50ef8dedc714edde613ab1b82ede48a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "f1af7f098361b1511ca0fa048003d86a77ca04419e8203cb5ae73d17dd1691e3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cold/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "75b47418f39a264f08dea9b9edfac06aebf8c5bf77aeeed1fe54a8aea19916df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "a77e5e6bb26a6bfdacd843987e907a16041d4b8bf9d7eba34479f49229b81844",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "5343c8beb7667d8a5ca32b8846a4b04181eab079b33c684a2a10f8ff74f8e48b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b23ed86f34941be1d5dae754eaab8e63556190b515a5fa99d16d4949d3204bb3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d081e2ae744a86c85d04119e19535869bb42ab5bbf0c3a2b14266361024c5c3e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "47be3d76e8053bfbf2aefe48f095dcb553e85654eb1ae8d10b419ef7bc4575d7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "579467a63731d85c2c1a497f46097cafcbdf091f294418b3a53dc4c3da8303d8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "25eeaa6dcaa1864ab6a57aa4624178a87382da9709a96b956a896f1922e129a8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "33d1ba33a23ba07dfd3cd816b090f4bd0ac275bc805234fa9b67fb06bcd3296e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "ea7dda274766026148816743904ba7312911233d400047e7e26fa9d9ff60deec",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "bbcc7d41722c3bddc7a0c0b4c0fcd14e424305d1987af43cd07c88b7ad29b088",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "d9c62ed659c9d88cd70ef22b50e60e7e0a767a4d4679edb4166f755eb26f9dfa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "4659562e809f8f17fd0a4b41f62aeff46fa012df45e41ef55298f60d6c5a3421",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "700094f4c38df46538c35352311aa67655d9e71db99eb14b2a477cb1b1a87a10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "55cf49088d56fae4e4bcd91062b37fc2a6658a434c984585767417f89b6f1ad7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "7d0c185f080d68647184054f75fa55b55e939bdef99e4dda3a11559389a5a7bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "d6ae2da287f81056ca45c83859cbc42030a1531f10e26f748d019af82dea0408",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "7875ca12d8c6c33049755948da7020ad0fb5fe19ec5deacd6e7c61c49429348e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "993965e652b911dd66004d3156fd64018e5441957bca79ec93277cd4572ed6d8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "163f84b2f193acd25c5c6fe1af69c4017fb7dd153c6c38fe92cc6ce024abc85d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c15f412710fc194ca76f5a38d286193cf8559e7e67cf165bc5bb79cad7c4ee59",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "98f565e244ee869ee970345c6463e59b7e28710f8ab2dea3feec8a1302824a04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b93f6d63030614a1e10932639b32083b7b3432a093f447d7c65a668604164987",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "7ace66d4b511ef03f731bfce9b407330c4807acf9c56c7ac207506bd606cd8c8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "25870ec808664070737a3e70a5ae41693857bab9271d297891397bf20141ff37",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "934890ba43977fa03b7774f2e179d3e0ad3336ee7c91dc1eb690b0de5afe26c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "96ecce5ccaf6f4726da41057bb375c2aacfa2bc33ece28ba9c0709a9669ceb76",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c55b167c9b970a847ede24a99e86bb2fc311f054c9c4a62000f0eb437549f906",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "845de4eae24702ac1a2217a5ec8aeb85b92d648d8147d6614a78344fce9a017e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "d8232e34518ae8a869184cd6494ac6df9fbbd8d2af896dd66ef8027f611d5aca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "2c52f944798937464e527e1948ea2876a2023495bc4fc105598b0e8f9ecd51d2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "c291f290ccc414aa78b0fcc0eef7a37aba99599e4818dc79103cbfa28d2d5a0e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "f8fdfd434463f851c232108fffede9ea56c56d7300bb6e53c8f823b7e4c7f238",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "f81260109def11577b13e7bfe6fc0d5dee03361dee424939a2594e33b0274ae8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "728e92ed05df1565a68792cc80313af32d7b308b5cc3747adf2165d18a25bd83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "a3976cbdf88c2e9b36b710412c04a0da09da14dab8a2d17633abb3dfe12c8ebc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "6358d9bf3fac9fbe64bef1f745ffe3ece82f90fb6da7a8c0ae6fa2a8d772d3f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "24a1c3fa9696b4b55c5766bdf9f2259e73c9e8be1de74b3718503009f56806ff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "05a5a59e493a49fba8d055b9168eda48a1ee80232df950aef3c40b25aac3c555",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "ae2cfc4712fd26d97f6f9f9554c53b9800437b99c338ece2f10c4edefc2aa4a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "478f2a6e7b9c27363d1f88484520cb74371754dd6e311e76c2a3d285b396a9c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "625cb97847faf076939cf0a282b9eb95063d5b8d4bc4f15734c1a7e2372944e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "437182a3adaf929bc628ce71d5144e3c68470055ed3984b11583527475d91712",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ff8764d091c004c352ea3c5916a31a373287d2867af67d9cd9eb18f98e5c34f6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "39c1a9cee39f416f31517d9f5142dacc98b16f097a8ab9051f1beddae8087792",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "6d993f7ab9fbbabcb9c43534ab460473f1ed9a6370d562163556064bedf5a18d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "aab7325c6a643a7b1a9163d4b6d909c79f6443c13c6457a91e2f14457bdd7f05",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "613c0f4214bd31aa6b5a0f359f69761332d6cf0c200ac5f87b623249f37266cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "1009a6dc920d57d9658bd5ff44eaa611ecf8fd375ef1407696c6b7e8e74f77f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "e7d54295616f60c21e9bc4f8da17fffddb6c905b8e4a467fe74cc4bb6426bcca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "65f10f598f82743aa73ee42f8d060971aa941af38fdacecd379f4649c8ab0182",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3ea76258eb94dbe6680eec3eedf711d496d71786fe2fcbc0ce0585667c408b11",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a12d6c9ccc5465a1aa70013735dc515d7af313841212b1912f70b0af20521e3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "25e6caafb0dea58173c13df14221074b022d902115b8066757ad977240781bce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "46ac77cac7bdd35423a29e91dfe0c6a7bad8a176f9732529eeea2353f6e87dc6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "8e21ac106dbc7cf91f5c1d47d075be56669879c4888b482ebe62d37064ebf06c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "5ef2ee7e4ca448ecbc2774e4f40016aad4af373d02d93a9fcb7a5e8fb292bf39",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "b9623e79c37e67d75d356131cfe0b96add0c5c903b0909b4fb0ee2b7830ae4df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "847f7bc89145031179f997f23b1e7c2ad32010a18315277e547e698211673ab0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "03a31d5a372d0459f02de5a20dd914a3601720a3e0b49b2cb0d9839a09ee1174",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/cold/claude-opus-4-6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "11307a147722ca479a341324f009c2e6a598d1cb273e9de781e9dd05a4326164",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "522af1100b4c1ea02aa594db83215b2bc8a536503aa3d5756ee0b3bd72524899",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "101a081abfb71471a10e95a9b7c7a12d09737052a64ac0f0a7e0f6752d93e0e4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "11f76fbecb667fa7af7fad31696793cc952839076436fc069e381b23e6745e21",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "9a66789d38076515e747737ea566cc22a58151e0e8f89ca7d12e981d684033e3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "4d325bdf5033ed392079b7f7d7ca9bf2eb6bf9a16fce3f2d1d48330f78a196eb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "f462a95b4081af9884538675733b496eb62a588ec6326988f70bb24d63aadfb8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2400496db24eeeb4e5caf807c4eb8d48f9a73c6175db925b03a54ba55081d947",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "f563cd81df75db216a7d6ad33857bf65b6d362635b9e0feadc6488fadc2c0f31",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "e1277b251417955e25f52fb40384fc7f8ee82db1a2f72fef8dca88bd13146624",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "a6ced1915c644217ccc08cdeff5b63de5fb9f44c459c533e5878b60707266ab6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "02a041daf7ce20e27553952fb129ae0213a8da3c6415adc1e2d062107addfc5a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "bca2c61a5986edaed75be33fee1e9266a01c6965f9d1e55a92b60dc74f6fe8dd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "23ff2c5eb153bafa969dd2fe8199619ee5eb259fcce199e81d6dbf000a818a70",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "9117abf2e9d4d61be2c6f1dcdb0d4d9f5fd390af1cc06ba7f60aea02bb0a37bb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "a8ae53170da984cff5562eaacd45f3784e17e803c3d6f6e50dcc8d842d988149",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "d0a38e75335367f0fe79e8332434c7c897e1aa9902a37a421b81d58f12bf8c47",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "b37f1d1324c8272ee9c7a33bb8753bd4e46358f2082a76c6c579316ebb1be100",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "4b73780463ce568ec81f64b0c3aa8522f0dd936155c46af61c527c7af07e2011",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "dd6f0b4588029c407793da26dcc87207e46bfe7176a25bb873995060e188d313",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "52bbfe007811e633751cf4ed2890c91c015323ac5f6ee2e8da4f410ba9254ffd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "aa9fc4785a57fa2837649ed6488d67701f30d604f5e72ac2b81c67f0f9c36447",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "6550365e4e33dd627c46d76593d5b1943e8d612101773b34aa82a0a7c9b1cbc5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "89091d72bd7b9a42e5894ffe817d40c32fa1992f8973ac3cd1db6173ad4e560a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "2f24d787801f632cd87a90e5d6637b85cc5cd374228209c38c37a4580e304095",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "d8d24f9d2356978db2767837479cd9b23419b2ae6152bef4c0c90f4fa9a723de",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "bd45e5f5bfd3c74ee40a6e5d74e76697843633c181a6abaa871d17fbbfc3d7b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "072030091487eda27ef8cb426eb9db3df33fdec3bb82a631f62914ebe3fc0eff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "e78f2dfea99a7536d7ac25740a65578101bcccaddb0474602607249faf91a8a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "73ceff718080c29dc475e9f731c075e7e543f09f44d4a2392bd729a364140e79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "ec73b9b4043ce7f4719dd89362eb171da01a05e9dc5d45be94f3693ca0257bcd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "be22126de6e144c49160c5f1559c0679dea60a93b06197b9dd425db24b27eadb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "46bb27c6d818b249e8a0c788a27385d59365ef668946d73c64a33868ee0688dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "75869f1537d120c3bf3512e02a8b21d0ffb6a18f327d6cc8d6ec38e2f2b7c011",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "eccabcac4dec2a26a6ca0f24d91a12cddb8dcf304287b95fa4283a228919a731",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep15",
      "rep": 15,
      "run_json_sha256": "186af502cbf63d18e4314f30411814120fcb6ebb0094543c3946d4a73a710123",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep16",
      "rep": 16,
      "run_json_sha256": "824f5594d337dca1e24a644927a7226b179fdbfe55568aec7a92c7cd6dffafe8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|16",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep17",
      "rep": 17,
      "run_json_sha256": "4f9cbc68ea2f5d3edb40be6f0764ef12a6a755911f4bdcfa64d9bfe8c664556b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|17",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep18",
      "rep": 18,
      "run_json_sha256": "e08a3138818863e2cbc40e3bfb22e973ec796a8262d60aeee1ee0369aef957e2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|18",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep19",
      "rep": 19,
      "run_json_sha256": "fbe055ed4fabec4ab35ca31a3eeb8823c653060a45e5dd1afdf88bc661e65bcc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|19",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos-hard/rep20",
      "rep": 20,
      "run_json_sha256": "2c48e64b4b256427faba48b31545b3ed31db628ca2b652e48f146e566e6b0b06",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|20",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "ab6a48ecaebaa37390de00228384d6adf50bc0847f5124482e9e31687b2bcf47",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "603eab51219f9db90b1770d981361a753d300b98c2cb8db7a17dbbbb61268544",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "ed029de8aeb6856678b0b49e2f47cea2d399f8678c1fa44cff9b07c32757f77f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "0845caa0d517c3c2d5fe7c594de6659c285b528fc0ebfb5045ff85e9fcd32db3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "d10ed2fae47f51a41af76e55043d08badb1a83eec14c06d2f47d8234688cc27d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "d33fc834ac3ac1edebdf457a78975732e5dfec3fedaeb4b4ca90c38b2148c86b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "3c04845ce4e7d17543a69f96c797db80b0f792ff7d8d32e3596f3b75dfeb8919",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "612ca0d0bbba9928e4555ab93ac8a5c5186c8ec707e8ec5ec9d8f9001e70d620",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "c8aefbbcfefce3fd1570ce7ec2bdfde9c6f4b4d2969403ee912e7c6e5881fea9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/cold/claude-opus-4-8/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "cbee423d80a8a98ffaa3297df35435499eefc3434570f27d00b8cbc8a84b987d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "e4c56416fed252e61ed668a2a24e329e3be7a43beebbe0c579cd136714afbc92",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "3c4848aed3f9ff11d3a4b94bd104d97f89ea1aba5adc8b8c1e689137d08a3104",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "fc733756a044a4a53fa67ce7d3ce4cf0403ab73efdcc68952916ed7a4fbc48f9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "bfa9644344e1788a293f83fe724a589585e507df8b47f27e7f9f0e58dbde7aa2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "b097329a38c2a8f6c00e3fb97a181f36d4413cb6c367d32d6eda5f9a9989437d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "08e19b0c7027c85f0dcdbb15d3d8a8ef166e3750cb6585e33981533b2b721442",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "dddd89a5399adabb713d1a8b7c62276fa515c7737ab56fc4a7b07b7d2fa5688e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "bd1d9c497fb7f2e277f0e79c8cc1ac29ec593f5228e462dbea9df3e056af6147",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a72cf8bae67ea7722e94d0814025558cd3de826193753e4909c0c7dbbcdf27c1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "f4f77e2c3518ee13922f660d5abe2c96ab1fa58077b596dfaeba89396f2940d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "960d6ce8b1588ae9420aaadfc33b4e9e04747087873b5723397521fc2bf54178",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "8bfbe502f09651686313a4c51c0a3d9745c49c60561240acef228ce4f5827b7c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "fbc03740451cc610ccf64572ea2326ecd14b492c72c2a607a0fdd025b0def34d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "d7330435e5ba005c48d5ec5b82fc95c1ef92cf60b82abab2a7638e150c2ae824",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "d514b32fc2cfe343fc3cf48c5f1ab381835682340766e5b018e385269bf9146f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "4b537be1a753502ccc1b424e00fefa70f12b05552d3bd9a2180ccd5289b06488",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "81899f9d7abe019ee7a0243029374ba2f738e13054020ed2a5bee575035e6421",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "78eac92038f033508c8d56b5ce0aff175f88ded8c447cefcb44adacdedf5e53c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "4a8492a3fd62b1eff9c0c03979be5c24a4767b061a172f99a84ff72c8a52d027",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "429296b2031be2300c2b39e0a6a039f2e99bb6a2e7cf6f07e85bf2070be11879",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "ce1fb9552625549515dc46fc5f44f2ffc2d9bbeed64ed1cec84b70d349da1c04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "81096c2ae0173470c064549216906a1d81fe8c130e0a0e0cacb4d6f7c1476c77",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "6d78d6797a4c3c710ded07e8f7862bb6eba73671049172661a6b40ce988816f3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "81f8c8b81f20aaade7eed6fbd86aca4147c60b3dbb270b51a2646636c17e1e5c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "ff58f033cfd3b9b2e8d7509e97520193d32015c0134e0465f723c998cae7cc25",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e2db06bdbb0d19d7340f161315939ee88f294389877521cb1a6244e96d514a63",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "f00dff6b918ae6bb796cbfd34c52fa2b699aacd67c1ac406cc34ab99dc4ad0c4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "bf502b91dc4ccf7e33b45b4200b80e4b61cd2a42cae3d4f6f0b395488344426a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "90d04dab1cba20f0faec90cda0aadf7bc7d993d00ecc449743fcc7639b49fa10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "3da8caf900e7c2ae6f755184393e4ff9a3ff13af584e477d00f43f813f9842ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3101b17b16045ba2c0f2e1cb8c9a954cd8c09575309147adb3c67dc2f63e00d3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "11bc45de24960a8d51129b306d4b127d4a7e55ff0b5aa411984fa8dd22a02001",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "2e580100c39ebc5cd47b947c07fcd1367d232a750dced198fcf25ce876093138",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "be1251c8554e5dfe51d78d928da7b29f2e96422b95d41b64fdcc355497ac3572",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "ed7fb6d0a4b789f33031fff3dbe906950c64193ff6b251566fcee35176fe7c79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "11922613020db4a000395f04a61a48c963c8a689695fdfa176748d445fa2b961",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "1c0e5b9873de1b23a54445d4f5ad5b65e37d2998414c104e1bf00eaad3e6ee12",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "67208d8e16c7ad599d3a25d990e45e5c395af09baff2ec5d62156e6ee312d925",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "46797db5181285a8d86640fd357e3dae1c83a4406c5bf6cd9a91afab344ccfa1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a4f75a97db72fea19fb483d8b3f1ce4ed3fb28334401a4bb430240ef7c708a3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "ca037b04509a0e15d6f542f64193896ad53b1b3d552bd7dabcb95bb16269382c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "54cc3019d06cd19f6d9ced51e4b24b4e2b807ee31a14e6566863ca776db3964e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "c5d531873cace859210f8ae0ff3223561d5c640b0527141c256a4b601d80b987",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "37bb9943108c12ffadc7ba37d445a5eae10337a1281db34f4c84c737bc59386e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "f6d69ac063cf3718a5887028984599553d3d9fb27458bc2e0fcc8e72f9dfa9dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ba20d155a83ebce88251eee99e53d690f44d645558d873d5c068ee59067df4c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "8bb174ad9fa8678586bd183914226bf3cffceb0231173e11a73b8d3d8a161e78",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "2230a6e125c6b255f6a818294025090e722e03ee517657fd9ba32c8113b04d2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "06c00f9a9e3425f734ab019fd958dbde3136b61b7d35473d4048ffd1129716ca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "48dd11b653b119e568413666e0ec632aae16025d12c687bf77838f8e3ff49db4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "f078ad2806d651b6740cf920b8ddd665d95ef707e666974866d26ee59b85818c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "d48f9ede26a133b500755262e4c5ae4848095ae255c15724d238456dd09d7950",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "3689320af234bbd29691cf45e80ddff8ad5dea6430f8fc82d50b71da81613f5f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "46b18a9b0c375b3fc4dd3bf96644d51a5de2a2863c08c624c806b05c47ef18b1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "3f2f9f731db778b8eeb383f50799bd9ab2d91e9679a79e5986c463602010ab4f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "5e271dcdcb9b22afae1a1ca8e6fb3cbc880f3fbbabb9b9f60693bd1145957edd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "b0a0289c6ac1b49c0956ac0f9e0daa3e15cf17040fd89772e0b8dc45f469ce4f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "e420dda1fb1f34694a78b6ca800034b4ec62aebdb12c0dd47a0c368f0c3c8b30",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "5468e072007a8fe7bcdb31d8172d7194d445d7928794d9091558e6f6870b57da",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cold/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c03d04fd659172f7c7c22f4856eed413f1f7318f7a226654b154910bdf25e076",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "eb874ac269804076febfcb309dc92e8414ae7c19190edaaf52b5307767e17c44",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b5c66f2e0601a568b0afc418b4700390f8aa1ec68ac9897e158c2f877de88d85",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "79261fec3fcc4affce08447c899eb92b939d005008dc24f998b670f0510acd51",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "1d84d2940831a5b2592780f9d086677746967253c5576b307aa625e0904808ba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "afdc86f48accdadeafd3210ef8559c2c2835f3b678a0e973ea8f5bc083d6c997",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "2eee1a132592b821ed3b343b14c6d7488f80a4a670b1d6b62b0247b482bc465f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d9572e25203ddfcde3f51bfd0275686732cb2a74ffddca02613a86b90e0b1c8b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a2fbab3120cc3019496bb954e290205959a3522577799e59230c58379c8d27c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "49fc57a42ea6525c524fd64d1e1488dbd0cebf077aa1f22da84c078442a1bb75",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "311c86728e14333c669641d9a44e3e3c76bb578477d280d2bea7c5b53f47c79f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "9471c9d72f0375f6595b021415eb02ff6491d9afe2370a5a28d4cb3d09ae1d8e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "812c01f48648c2dd14d26ec1cd5b5a7fc719fca2fb3e7d9206f2c0cdc5623913",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "968997cbe0b4cbdab66aa999fec804916f485b15552e7be865beabd95968d477",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "46f9afac137ad17505b2d0419fe4c9ac05e0dcea3f48bf2e1ee7f8c3bdfa29e2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "27b07065c1228c370a18888e0ac8192247e04bf1a14297e57e6c5d85a621f8d6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "a9b7a3b174bbd35caafd5b33d3f6a7a9c71e0dc02094b720826dc3172791e43d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "7470c01a2c311560fd07e08c0d78eda62bcfbe5152fe844938941010d5fb2301",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "11a5426ab6fb5188fa88012bb97f98fbcb869f23bf81157792092d7781b6c174",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d198cd1b79647c6f356fb9c4c5c290685ee523a6c1b0692f9667973583e0cfae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "ca83037496aa715407068a24f2d8d6a87db85c5e64495581b106b0b32af73b3e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "cfed47f7060069af933844d6e61e10a7ca5c1e73b138fb80f140372688040b48",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "365902a6946d23c291fd1a685f6181a1c6c4824229b130bf5fbb8d959e3af42d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "9b42ec9a8901142df329b0604ed6c81077f201008808886a7a85010294cfad18",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "c6a1ec1cc42cfb291bd62193dac518297104d78d4ec98ad19be959409a98ec86",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "8c70e56589d9b543bf6c55fddd42984a155d5388759b8bd9fb1075eff1040b65",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "0bf0862f7168fe207b0f35798af34311722bb60bd0b5df39c6d577f5c54f33e4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "eb3717d84d2e364ce1f157ab5b369598a9b96888aa04fe5b969504a0b243d164",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "7e61db2207a98286768be8193e6a8c7be1070b73b37275a5a2e1c64cd32198df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "883e182b19ea01b74c61cc464a5f56f5fbd0031ea0e7d011d3d630f3d335b3a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "c54a239637820e0701256a3ed3f0167c6dfa6bcc3e32f516d75d80d0910c4fad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "e79f8e4712becf96b4867cf6147342aae7008aac286c12d5e900c8467dae3468",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "1f77ad3ffb1e23be7e90ddd860d4f3eae68dd7201c695662eb36f66434745f46",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "84cefd7cfc36647af3f93c76afabfa07ec9a46952a47b935c2631b863c13f86b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "d14e7d0c84efcc1e92a5308f28639479b1bdf3e4cb550717256c5226ca9facbf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "be8311ac1e2cebbf409241e6d69eb01678c3a113b4070ac53dcb329ad79cf081",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "a7f1548746bd52772522ab6d118a954bebd7e5041a4d264c3dfeabc78fbcef33",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "fc87ac4c70e480004237bfd943e6fbecd09349a59f0d98589106a29e8e98b19d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "deb07dcda4899bff0b42404e1514d806a88b92663d03bcf9daf648eacc2a4f29",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "510aea6a4e56e23174e8831321dcf3077252f9f2861d5c95c4e28f36fb610673",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cold",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cold/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "03f42a6ba5052a1693bf883543723b97886e22aa90e68bf7f5d6004cefc308ca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cold|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "c49a900607b4c237a98ad6d3dadcecb3763dc2e6fad9691f504c4658600f1b06",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "b92c40bf937008a58678e6a86fe0aecfc450f4e5e3f98ab4f5edb871275744c4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "8df0ce94b31ef30fbb797b746657a668223eca059badee751563da00c33ddf7c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "830fa8fe4751e360e1399b22bbe8656bee94b6120284c8798ca6b90f38c2b06f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "55b7dd9ceba5a9aa631cdd5e5bfeed00e23306b8459c2fdaea087663a31a7d2d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "bc9fd4b846ca8182da1ed5253b5f4faab7ad006ef0517c204e269a301f743a55",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "7ec62a6afed54849191cdda4ff77f7f08f6e82ab8c1705d3a6807ccedacb950c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a8705fc52be57e92ef6a3397316a81c878bc6d380b6ad55c6767ccab7ce9a374",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "abbe5d42e67d26f7f3f7ec6a2648066f4cea25d1c8a6ef2db84bc252e212188e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "e2fe8c36d7bbb3bfe904ac8cb2221f3bdf7ff4fa69187c62ab0bb8a7d21fc448",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "18ba599eedd8967e0c86eac59efa744b4066a12c83b4dc9550fd99bb6f8e0470",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "3071e428a2031776006a468d24e20800a32852f6a17ceaeacfe343c316da73a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "82edb70d2cc10f53bc19fe1aa2a6d3af94c0ba1f6c6b3accea77a83e4ffe24c1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "137d27fed69e1e8edba3e4407acafd2073dbce1fcd8c6a6b19fa00914db8a97d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "9b2a95c83b1b02677b87ccbe9ddda6263c815ce7a1faa9867d2b1d9e9e23a67c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "ad60269b2862cfe7156adbbd4a2175e600267c27a9ef909ad5f5fd5a8c7ca0f0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "e893049b33b326212e64ff6132fdbaab39fa5e4ed27531d40f67b430f1cfc277",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "60f5f9b921eefb76da8ef15263b3e9a63ffd86ac59d7be11f59789b140f0ccde",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "6fb0486808e83ec229a528846c184ffc1fab6f9b8c4814b53d503d689d1cb016",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "2924bd7053742c655786cd90f3f44a9bfe75587bf46afe1f87e54b436dca81af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "b6df29db1b93d2a7362b1a7ad4cd5fe41abcd5ebf8976f5c18d952b6b806b47a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "637dcf0128794cf9f1565caf1b1bec1be9c8d8e362aaab94dadefde6fd0095b2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b9a58eea152ffe583d1f9b22676ef84df489675817f787ac3938e978b44b5c1a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "977ad5dfeb15e8b53d46bed04da9eef8e7ce470980aa962a53dd0ddf7f0ee69b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "b3339ae383bdee5aa0242b15b0dbbc50e8e5142d43e12c91d17838993aa86e5e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "fa39095e6518877ef09d6f739bf601bab592257532a8e65bc1a43cec43bd3339",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "fdbd8772081d1f93b360b212f7d99a89f6088a44e8dd70c2ea020fd785eb7abd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "cf835e338625bb0e63b5db030915986e0e1d5da653cdb22d59c03eb8279dfd01",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "46ca3f1c425070b0ea956ff26b8af4d182f3edca351374d64684fa92cf5061e9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "5b00bba5ef9b4bb71d91b468532af69f896294bcc50915cbb152145ab8450c0d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "c7d4be1bdd8632f82ca3511ceeaff226c5296f9781ccb777c221c7c08ba16a67",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "65adca20d872c2d0d2dd502a0eff02003df1bf407d868e0a81d463ac2b8bef48",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "126d44047f44a188be3418f1e2e4ba173a8984ba2b378a192efd9f4da2897bf5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "d95617dcfa57ce042a726c3fb14c73a7ea9d63d596ee1d3d9ffb03863f9b73da",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "604641dfb8903593067ac8453a432ebae7ecf0f9bee6615e20f64a1a57fdfaaa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "6e5d57f12a5ebdae1dc148856386ed966e7c5d19803458b1ce739a5b1ca9347d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "180f45ea00034ae7b984666febb4a71ee50625a306a3fb95c43c2b379b35246e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "24a81bbce8d5a4e93246c9101d9d30f3e2b565674d70dd755c0b7d217cb622ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "b8c849e28e7d79b91c45d38fd13ac5e344df091dc2f7a528018991447270ef77",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/cursor/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "920938fb5c5e499a22bf4b41b7361c154dd9fa41926bfd63dae3be8bb445ae2b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "6b06b84573e1d09d3a920f67909e09e930ab8f0c9e51dfa93fc85a4d51af70b7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "7e56dfba856c8437d04244b99ca49ca4710550af14f551f36d0d56fa60eb918d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "4631007845ca9330a8a77f6318430fc6cb9fa2d0ea9ab4d1a808016fa07b169d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ee561b59afda9540d89fe30a0b58ba5542d4573f74954bcf029e0ff7bf60cd72",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "78eefc96d6cdcdb9b4957958f0bcbb19132faf02aabe7e6f0716a01d0795a0b8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "b77ef68c7278708f16c0cea7a2e1c75a283c72d5714f531974174da422491c0e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2a4ee7e3d811eb98db2c6a9a2fb7f1207e38e63c68a91426893bee8d66b9ae6c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "ce6a0e2a4468015249d71a6c5fbde9fe8bb1a696ca1f55084828199ef713461c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "7b7a8e9e42fd72ea4ec3d3e84040679bfd2e6cac34e5b77d12d97acb8207fb55",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "9309bc63692f3b398520ae89ee5e62240512b284d9658e3cc1f27e10fcb59473",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3fed79d63cb965a31515ae1e0b6ed63120800d3303e74f5d6b087da50e79d5e7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "0bef7f0c2f0975fe33329ec9ae1da9d564eb324556c295968bc493729df89232",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "de1cf3f6ed333e806678240092515bee76a3b270d8afa16ca47aef96e63375b7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "441c424d3ae866a6245186ef4942ef512f77cec6d4c3642ad99ff2c0adf1f7a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "affe4c2d9bed8d58ef306534250b6051f2723420b64cec8f643f2702857f569f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "38489c5f4bd55b3b15a9113fd8bfbc866cdb024c945c45a1bd68f85c4f83df67",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "f4a068c8686f0298c53be9d36f196445d37c4ba2aae1b9449ace902105098111",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "0dacbed3ddb8c810c4acbc4e10aed4787008c41ed7942a2c519f81e5d392e554",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d1fccaba322a02a403637516c74c295af5baa447084011448d264678073aba18",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "6322e03c6a260927fd5a01a173fbddab7fcc58cd43f61fe4216d25130c0537f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "e5b1225aea2f1538f333c30462112bebdda05fe15c65732824059a2f77f70470",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "9c4e197340fce8803a7f3095318b542ef1f92091ca868ff2d79adb294f0055a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "572d2d30299fc745da418f1f4ef2e6387b59399031e085ab68d87d84f2a42fe5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "9b3df1d75d1b21e3312a1a4cc581c47345364f5ec7dea9a4cdad175f138bbfe9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "a51ac4b6346ca051f94005b7be44cea077cdcdb75e4e647ec29ec0edc45ed377",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "958f7bf0f041aa74dcd275d11c7dc1969d0bdd9f55429c49fd0c84cb10b1c4f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "1dddd59700eec10cb5ddb88376275b129f2a3e0617e9c0fcf258db514e8a5653",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a2e77ed7f94bb673e9c2e89dcbb273790041c2985b4971932746d3851787d5c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "4ebf009332b5da28c8e66c8fab6b33b75620a22277a4ca4d6362ca57e3e9a152",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "6d18a040a93917935b6eedc5a1be5f8c991b744b66bb45e26bcf449f439dd30a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "4a4a20d94f0c00505ab368615e1beb0bf532b39b7453e425d19b7be0c1950c7d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "27bd5c8030dc76465e6c35cb31547a77060d16e3203b5706e591f20f52dc6ebc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "f85c44d391df668ef74e08f22feb6c213910d60fe6e92a1741c1bad8a4b7835a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "9acdb2be42b1344ede40458f312df2a7abe868cbedd72a519b71d5dfb6840d9b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "63effebcd7948cdf21d448252ed61c17d7aebdac746665175098be99a09bea88",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "bcc67ac095b0591185bc2f978fdff6624d4b8ef62fad7502999795ea26f3627a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "48f1bff4fbfb948a6d8bfaa9f45d92850868d3f780b8b2044ffa772d0f4e02ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "596666c8a8df7d2c1f93c317b67a948272185922211d4132000660b35d2a6ca4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "6cee7fb5b5b04bda5b0e6980d180779e121e5b635b05c59ffb5323efec3cc942",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/cursor/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "ad94c00b1b06a821282988b556e9ec936daf3ab6b5c6687413be0d7cbb87a10b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "40aabff08996521b359ff2159e4206571f764e199f48b66bf6e80f55cabecaac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "5a43ff3d129537c8b53aa771a890e4f139d36c443a517d1a1f13e3cf4964a723",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "eb1424c531d9fb31c89edc577eda980386604a236de617077b6072d889ab5bde",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d8c3bd402cbd0e329c14bf574335affcc73a2d2118831fab96165a5a11aaa60c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "ab47ca87d4f15a3d35bca882b615ad11736b4c14000495c9e07c3acdd7b5c61a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e4e3ec121bb79bf528292f92f0116edfd9d8378f6bbb8a21710ea660082ee29a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "9125d07fee18a2c173861a2e845e2cfce017c7e7d37489c7379601c96a865e09",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "40a50d6d80a79099b0c9e8f7e9b859d09ab8966312f6be76706a7852091b78db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "d0a85ffb0cbdd0dd462e2ed333459f9941f52cd9d5cb2de8086d9357d22990b5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "c195892366f2ba657a639f9eedf17f147c40be559a5ce8c2fac1a3741c208a34",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "31f3817ac8ebfd5bbf72337cd95fdb5937588d610bd86c7786227cfc7734c859",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a901c2adab1e4cd89f81a8afa3e334450c43ca606a25bf14b739225a3d369967",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "32f3b1eff6a3c7ec6f1ffe3cbb504a8eaabbfa15e02f55803afa14ec3592365b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "f6470a87cb8d53d4530b7db15399f24aa3e229faf2472fca15628b9eecc43d39",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "38be0081217a82cea3ed79805f2e9d83d54e43522bf9335df0042ba4d501bb45",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "c39ade4ee4542624e9a3c04d08a3fed45c1254c7a336f62fdeae997f947b08cf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "46f34ebc66ca942c0d1a727b0457fd51c84eaeb39bf25c1bc33186350ad1befc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "eac9c9faa9f07bca88b0e6163eab41771ac4e204d41d7c8d1970d5aea18b68ff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "62277b61b997783aaf4f9c41b22fccf734901f766c717ac86bc4a17812fd2ec0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "59b5b398841def5f070d18c346f9a55c912a78ea9ca39700daf022c404bee021",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "854b68568ee5a7f8c58f20166ae04c0e2a6a64911954a9a248a72057dade9891",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "e0383f1a0cc4a62db207cf31537f1624c19d52bf5698caa73cdeee5b313feead",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "cc67427a3f2df7b548597ffb07c60b2ae27f20d9410e7548c430487ed4532b4f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "29c4071e2541707f58146f60e457c27820edb99819aa0ad69e39e5448c9e1032",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "b9f86d0ca82b6383b9d05d07e57b7db55bedbe9d29fb6cc14bafedcc45d3c4e2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "2d8a031a9bcb27d5f944863a2cc3823b5383d269c8aa479af75a2392220a99f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "090cecd183d5a858d56d06191e990c496e695784e1d979c43eefc7b6129f9c9b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "5b8150deec23b1c32e04c1675d5035b443faad758d90e8070df243c0562cf590",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "14e10f1d17b7f6d57b64d01cdd9d5a5ed97084d1441d839028c9e24dd4ea6fe1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "7effc2c4a81d92faecac00788bc8d0b45c54473c2f0ea3f5a24b1a8be1c2c32f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "1cb67a4c8e4b9861de6e9f407228e24fb328681266777071a6158fe877fc31c3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "d401e8927bff8178058aac4fb7c256eddb95add453516e63b5f99278d279871e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "f96ed75d2408395ee51ab661fe856d31c270dc31f9b54fe2cece606ae0f94b88",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "b1d51a607d65e15166c5829ecc9e2dfe2596f32c89637102fc928701882821f4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "35e7506f6aff8a754a0aef584efc5bb746a4c642feefe3596a832970d5cbd1bd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "582b36b00334c3af8f3bd6c1e905829d1bb8efed5a9e54c6744c1f6702b48edd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "2cc6932123352c47c19829e1aa68fbe3cdcfb6d03c7430de9949f92dadb5eae4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "69e94354392e7d919d41a4b4f9d0d0690f3c8108ce14f0c550c6843774bb6c77",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "ca2b8f70ddc4b17bbb037b76901b7bab4c2c06ca22a5b81afe5eed52bda41c6a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/cursor/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a9ccebc5e0a48ef702978aca8add982602b64acc13523664104f01f594f9c471",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "db0e3c95144eb14ee3e15a3f13427e29604411eb1515075346bd1b9321f42286",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "95058a4abb90c13d0744e5a7757bb03a58ca527512583a3e4ad2425d8a52845f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ab56038b1a0c52d4af5795e84cc9d10f6c0c3ff66b4d3b60367913def38cc781",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "f702550c03289780ae5c9b65d1063e89372cd08347cb3e98f80390e3a735cdf3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "ec33e79ec221c73919aba77786c6bd8cfd353716248e0ae3026a4d1c03771e96",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "d0c6482a4c01baadf4d1ebaf8294f45010a6295a0460d59cf01a25bc10d338b9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "b94fc1aea1803aaee407ceb5c855bda5ceab49b27186b2364cb04c0136dbddab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "63982504b73c7d1d7612f77587841258b3551265aba4999fb46e8fe274ccc4cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "76873c575eb5767c989666f957e899f1c6b7114fa67f42c4c0e98c2b92511863",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "b4a060e034fb4d93370b7205c75fe777c344a5696b42e262675fc77699087c10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "23daa66274a754056fbfa5ad4f98d99221a41083929d3855b76726e2a3345127",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "929c8b45fcb149ad3be002ad3aedc4c0cb7e8861e7115beaa406b1c8129489a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "1eb86459dd138761eed5ada06e52da82905a226c0dff6acd58134614c47bc947",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "0adbc1bfe29dda49a805fc79f7eff5beeb70453eab5b3c741ef55da64f13075e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "0775c6ec7746919289373458414f1f82cf07f807e3c95b5ec89c7e5ae497cc7d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "89365ddeeaac33dda95d00e7ee528a86f85a8dff4e7555267627912639174b91",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "83a71aa0a938696e824344ced7f6fe58c0247b93b4f313e7d13fa4c5a6037f7a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "d76bf7e801bb0d321242ae1600d8f595e920229df6e7282d7f76da4b4f8ef7c8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "1905759119cc8e2b28286830c2d33b72541ff514926054b612d4af10f3cd84ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "2a5a58b20269087f2023614555fe1de8839f615502b3dd3e489eff184ab22861",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "d1a004fcadfd19e322629cd7fa722e7e78a8a567e63fc24cef49768ce057e127",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "eafad2a93410de7a1e33bd7310b3a4735a8addfb708592cdc578bb99cb6607db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "667c26c9ed2db83be2f258f6f5396867b5838cdc93c2ba593dd00665e2a33ca7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "20ecbbcf59249c09f0f1cee227a33eb123fdd554d2c47bfb6012b211467cd814",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "d023d6ad372246777599f749d715c75ce6d23c435aa9a94cd3984463c59a90a5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "4f33d5968a3663bb960dee13f27426158accc089846789e92ef41692646b5ddc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "88d0ce3b35dba0750277637668b9a3e7721e5427efd0077b12a0ba4985de754e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "6aa85ee0af3e2e7e242460788b10f24886cebde0e2d7f3b9eaf2b7a1e802007b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "16954f2ad235c23e6fcb8756ad4fcf58799102b4a0dd1951febca3799f0231cf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "af24b75b41e198b812701335273a3de06347ae20a261f1780f71fae5bfba958b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "13f7a719e8f06ba04788bfd62a5eb427a95ab4eb3fc63e96a469c89669d33ff1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "575cd26f53957ae61f9865fd58deb1e6a9d408dd7d93ca1e1b2123e0cbac891f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "1dfd5187157cb889a68dc6a1382a33b7218d737538e3b9be02763e0bce05ace8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "9b6fbef7c3a8886d82a174a07222b124cf9c37c43e2e564e228f54a86814c071",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "ee87c7b68351877663c63f3a19d323a869f78490b6e46d3f32bd6b8ccdb95b4f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "23ddf41510d08f3d7a61548a63cc95a1edd7188d5234e9184a3b65cf9d680c09",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "84afd80564d426283e15c8154548ce2b3f6d0f90e5bb20063c016f2ff7fd9bd9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "b57698314b4f8811692ab716c83cae5f8cd699927e7426775ca400126e242668",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "43f66be288d48d848ff56eea10359345e59c39545cfa461502f431610516aa3d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "composer-2.5",
      "path": "runs/cursor/composer-2.5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "3724fc9f6bd299b4ca65e39bebe9beaecd4fb329627a1395b873566932d73807",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|composer-2.5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "bdcb048d7e30db7fe2fb44319610c13955ef6323e83352bb4d627ec7a2b7a62d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "46b9b0d2fa2c6dfd213b32f6a7e36655e0b65484d03ca261c6aac0c7f39072e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "e5887c8cd07ce9240a3c4049afa2cd6125bd0afae76d44290409b38f1199cbf0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "db20eb28e6e85e82c51da68328bcb2b5ac34ce3225f139c34c769c9c51668f7a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "de747dbb41573d6ef01956743158e2c01d2fa134ff59b75baf09fee338819cff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "66e17dcf6876bcc6434c192eb8455d51dde0908e91dbd5a3be2f63c80b3875ad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2287501aac4fbf3bcd8d16f9dfdc9f637a9755037a54c61db60dbf09b57ce6d6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "21b9a0a98cec3a611e5aa0aedaa83a30ffbbbf6f8e7f6ae2ae6a16c895a8f012",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "01969ea92a78e64c07010d6f4e2fd5c5a0bb5ec9c093c13a8c78d36c9280a7db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "969751e29859364448a12887ad03736ad74f42041ccacb926db0851011a7fe8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3da2c6b291f93af9b8859f73a475e261d415fc711ed5c6a986fd6b48a53f3001",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "e7e0ebba5129b2e16c30ec4b334a168b6f00a2c7b8d8b831167f818aa9c01394",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d1f9a3f38a667d7f0ad55cd8a559c9c0fd5bfc3d486f63a378a49ef136c6c9b4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "9089ece45ebcff2df1a4ec58de4d1e962799da7d4785f8a96fa06207d052c414",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "eb99508e4b6432698e92d149c7c1e6ab9bb319c048d0fdb3e46ff147ac7ea72c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "89bc21bc6f08bfc2b287d933e034d571e961c2e9f432befc27e8b3b1d60cf805",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "428cd847913232ea7a3954054e7fa6206ecc9fc8e97e818b087b87f00213218a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "70b1bfc65c84c60d55db9d5e9f9b89b68bd71a46d27961e522db5b2f1bbe133a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "04c3c8bf32104319e28a1ab1f0631ab64fe0a438358be385db2c0e5f96ed559c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "ed3b7a3560b674c0a1c88b525aca4a45b2e9dbe6b704a1252eea4eba13a0edd6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "09fc940fde22793520ab5a40f3c795a3952f0e60efb8b52848fa1a2290ed9f93",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "6f2c73fef68cfff6d734d3fe1ed54a7fdd230731422a9d81cc706f32a8bdd8ad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b7535f78eca88ae917ab7ddbe027c362fcae7ef94250ab23873d5e53e16fb624",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "5287de027a1d3cdc0b0ba88c8a2570a3146010fca5f3647555bfaf41aeda3cea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "dfb00037f94340675ae56fee7f1e09c055aff250f9b09d1d2ff0e820c4e568b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "2e5614bc3848f5ce8dc0afa26260b0b41629f6a89f6a5074ea892dd3a2f5dce4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c21e8017801851ae21fd5a1172a2efd482d9632ee38d29adcd73652d4035a025",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "e300a240fff9e199026c4969e2e30e09e3d052015a0bd74e192400f542d7f624",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "ecbaeb21a9dc79028c30c3c639e019b1dbec83a485c14a4b8db2d6e794d483d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "81e4d94e555b7300100cabb700a5b3fa951e36f464eabc4c048d2dc42fb04061",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "e6ccff502b8329150c592de5d49f0feb27b743758c4d124076cbee51058c3199",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "e28228865bb09c1a92b77409820214ae3ebe45a25f4f6a1e305ca40325aa9299",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "543e2328e786c9aeb9a21e3b9a88b621c738210def5d0cbdaf855c9b9288db3d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "f5cf8ec5e3fe1a5455fcd8cd0df14d3ff374af8cc1f45a3d70cd098bafe364ab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "c4d855173850cfc01b4ecdf0c8abdc033e01ec94f7d91b9788dcd510af9bee01",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "5a494340b401a1d0a39e1d4b9d0621a222f7bdeca648c9313dba5e52ef8c8594",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "fc9d8449cee4dacdef80685ceb3e9bf99df242924e0ad8bec97d66afb70d23b3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "84f827ff59ada41b308c0b162c04ad26f88b6516f11f3457d7cbffdda8da8184",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "0e16c3a429d5437569f47a22ac97abbf5af5338c4af8d23e278befcca1abd891",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-luna",
      "path": "runs/cursor/gpt-5.6-luna/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "aa546d371f9579049299d81aeddc33194d15fd2bcec670ca66c5a31e4d59a069",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-luna|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "5d889a3bf50f92a49e2fd2065dfb4e0ef9817c0779ff675ae04bc2ef33e18cc2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "950ba51a0377da47a2c8db37f48475786c52068c0132d6de2c02c85d8da51378",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "338331f458aecc61e6d6fde8a70dbf559e2deb8f6eb640c604a664f464af0023",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "f892e5db02122d9011aed9ed0c02113cefb79a61f9ae12a5af375a04fd0a54fb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "552b0e6dd3af18d9adbd7bf023e86cc64b1ffd35d4716315553157442b275833",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "99478adf2966b49d22c3c660c790301ce53bd02d8f954ee6453dd6aa63a8d4ad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "3367110493b49ca1bf651cac5372a210f4ec1a548e82e173b3001ff5c53be702",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "185dd37b28e67d4bbff54021e47c6298f28873d189ee5cc456cb65b4ad34b728",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "4a0ea605678e7ab9daa453e62886fe7a93b347bcb82b487db6664392ef8da6f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "13daf443c65419876f7d959e8024d849867034ce35fd5239b7a3467cb569391d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "df0aff3f25b05f06c50ced1ebf6f0e86ccd3e69537c7450fd2cd00a710bd9024",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "0073ef16be0626917bc0c1560c2ae111c3192a9f1fe3ff138f2990a857b28369",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "e53c26059a084863d1050646cc555cca72016f5671bf983ccd176129390c13ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "728e78a30933a47ea1c36773df35b69819096dd1dfe5b15c9d0bfbe4aac38012",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "4235681afbe458b28039a3d2baba8a426a0f2372fff0e5ee4a7742e179fc55a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "36777137688417e6ea820d061f3c11c7926c5444037b9abd23be3fedf20c92ab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "1946178608edea8ad89835df8bfdf8903b920986a2555ea02d9fb73aba37fee6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "16cd863de295794fad4a379dca88c3ad5ccdadea477cc15fc41c1010f5dedf1d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "efd0db8c0388d5e619e3b2f166868e26a7c6c7218a0b9e88d495c1dcbcc4a05e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "dee9569ce4d4f9591784142507c7e3944288f7efc4f76d445e038af37a21cb20",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "37676446e43f5c1c5589aa118358639fa2a88c558d29c3c6839fddcd5f6ddba0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "e39cf68fe210136916270e21de2d898f08c100231b5969a5bed46acac1c9cd89",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "86c293cf47203b3811c6aeaa42d1f612cbd446be254ae97f24b28df9784a5b64",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d8c2d1620f171189144a06780453af27d07e8c8d086908a57e8570586885abf8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9179d7dfd52d789ec0f2bef3fd0fcb024d8087d81f761b6873982ea8d1b29ea5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "3efcfff21aef767f6f9f83f8597f9dd693ddc2b7a5c94548f5aa90234ec0a79a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "747309afa6bf842dfafa8cb2726ad1c3dedb5a52ef6c64a7e98a4ed2b9fd3042",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "f2d389eaffe219b2b1c28c99f915a591bee3e58afe2bb5bb2d68124a7221fc3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "51ba3d2794ab7324afc2d0f341f4c7ff1fad1b0fb07106088f29571ff3e85bc6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "9abbdeb00eb8a8235972d09a93ffd3e1ef9b029bd045e479abcad08693ecc2e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "6afae227a5d33f12d9f698dccfbd1b866acf39db5289e6f5502583bf72d91a5f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "f545921ab9994620d852deb8c46daa652f8eef59fdd6d983737819efe7207a1a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "4401ff2922fb72d165cd471ce644e5f4d77a2bdb63da703a3db22cb9a169f6bb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "57374e789e15c2f7338dd3d75c536c8761da6f00705214b4dce65b40fa5e9d18",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "28933f2606584e7e4d0e667cc361273752786ae6b59aa18fe9827007ee460b6f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "503281ad18ad818728b5f469a9a61c9d53cf9ece47ee7d4010e44f6e78747b7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "7ff1ce4f22ac086de1bd75027dc45c0e11ca342a77e4bbe0cccab15e5709cc9d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "85ac1eeb988af0f51e48d4175d2d1fe713f07c380fed7de22199dbdcfe8296cd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d865a7ca8410f583dd788cc5ad0ffe3f8f1fed0e38860c2f73c58669a30d5b83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-sol",
      "path": "runs/cursor/gpt-5.6-sol/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "6192423662cfb9fef5f1edb2586e2d643dd40e74da25751fef0d4a20c3d48aab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-sol|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "3189bae0ff4d970c50b2a49408a4e6ad7992eef0449d74aad51a72bb053ff5aa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "1103ee1bd82a57f75e49eccb14b52840628db07a05164963d3ff0b1add7188ab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "7884cfad751a3e6b5158482cc198420b1ce7c4c31074e79f84b4bb3afd862f66",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "fca0b7667e89b53d593f81f779a48e8982960dd405c52a557402eed02961bfad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "89b21ac2b5d3293af6201db924701c3b96ad9b9a7e8937bb11a8c92b99cf285a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "27ac83c7f742cd1e81d2c436f1a15ccaf6623abdee532ab6dd622f40cf69d408",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "576cd12e716fb8cda3ca5d7c53321f9ca19c54beafaf95f68b42c62699fae5ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3bbf42b15d6c4d928afe7c56ddbc51b7342eb318e92e7c9fdf459d0d6ca17fe0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "b560bcf6892d6c20fde0625b736c60f176251cc66f72c94311a58c735a410a6e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "a6a1b8addc89b221d4c3daee0dc7f748b02951402a8e83fa271d03064f985e13",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "15743d92229b350faace38811498504c7db43d49affb7c59b16b0c3a56b1e108",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a5234291ca92e4233c38cc83052a2046609a457c021e2ec63d7a6bb9b26e1391",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "a6171a55c0fdca48a8bc062711c33635b7d7d03a30a4a015ba7329a02d6fe253",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "8b21a0a862be6c3ff7ba0fec774bec738c731451d8d04785a6c1ed0797e3ff8d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "41a1bf030b9de3d4b0d2e1b1019ff4cdcd5c85e90fca87c541d83db27e6f1537",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "6075dd98d7d973c47b525a1c667d28c8666edc85f1a1e7e58e8c09603b433020",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "7336f034fd00bdecc9f046fb8ed4b7da273c323606927a00f394f948b785cae5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "8fbfee0b50600e20b774fde274d6de519cc89568f5d65d628ea19e018f27170e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "a7520c069179b46f75dd0bed1d179af1237cbf8355cd9b2b56680040aec2a684",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "338fb7ce2c10c6faa39eb3cb7aeed079244489adade64c282c08e74ecce7d69d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "7288789ce60da2605d55fb37c533e7ab875a2612f26c5596f2742c56fe14a9f4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "74b36aebd1c391a899b2c16ae9f7415917aed1f4bc66e89bb31469215869e910",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "fb1000c3207c06ba1432abfbd4334174bec29eadaaa787f3ed30a66e48490f1b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "60e4dc92fb4075945a395388ecd1d3ca9fc7e3269b9015fc844507f3d807a13c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "d111fe637c727b0e0650b90016b24beebd9fbe9f71511cbad8332fe1f0bea6cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "c8da08aefc05629a9239c4dab78d505266bcbf51970f4c0dcc9192f3bcb29e40",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "e615189dd6554340aff0360ec479974ad95f44f5dda5e2fbba3c35ffddcc3afb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "d2d2cbbe35a880ad29ea3afe2f422a1a5ac4ead1c670ba79d4581bcf10be50de",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "4d672356c52e08c6128f659b3e5bf01fe92cf257317d801887f46b607220921c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "e1b0e9b6a5aa7823ba88fe6c22286971af18573f3993021bc9330816718761ff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "dae5daaaaa4bf839f81466f9c73ea8e31ba1eb5f6928c3f7af4c87dcb67b84ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "5a1194d224db7017fc6dd49e61ad9384032ceb19ee78c6a62f96537683e8a4f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "01c11c064b2d9fd2d262bcf9fefbfea20eb21f935fe77e32303855cf7eb55266",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "218d4af93af3f3866d30d0e13082590a67787b208cee8d8979e0402a895bf04d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "7389faaf3eaf46ba5a802aac755ec0aa45e442e8760f5b7c59ab755d265318af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "41f818bab698f75216a4366980697cdcfd798c82a4139e82b9fd818bc25fa3cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "120c51b958165701746833ee9c7af1bc296e47d119731d50a53b8c78983126c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "f9d7249d2a148dcca91c14d2c31c293d63743dd4fe9cb729b18dbac7887d5450",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "988967a89bc2f4b56c8947c6a1ac0dcf0db640e9ec84dc5a9c85bea7937659f9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "gpt-5.6-terra",
      "path": "runs/cursor/gpt-5.6-terra/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "f826a31e60a28d38078341cdc8dfbdc6aa1a3dcad23ec971e24dc7ce0d0b3231",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|gpt-5.6-terra|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "29309167cb37065612d60c1625c9bfb8e3e2ecd1090b971e836f729e07b4d561",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "a31c3d41fbf053d13ca95882d9b1edd1c30ca3cb28a09a3fa3ba3817571ac3a5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "336c49e09af313f14bf041f0dc07f7e21a21f08bf833872c84f54a2ebae9f31d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "241c7bce228266f37bab947841eea93c828b6c4dacac211a71f4d64499073afb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "0eea6294cc698c384e936a5afa69c21fa7dbb20ef7bd308bbb2e2a8c03f09acb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "7ad2342d35ca1d6920ae6f3e7cf396087698fde88f2fd113ea293468a4256aae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "ecf7fc6dc434d775ebdffcba43cd75e6bcadd870dafa545bca56950049becb3a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "121ac86a5c186460e2c380ef8b5b1339d9331a3949b7558a31a8624454e8d828",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "baa0c580a9929c99a8a18ca017e68d8b1c5422575dcb38ed56c7c9cb97a0db9b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "079e2a60683157041e6fc7bf7ebb2de3ae937b9a0b13b9ac1cd9182b80611073",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "8816f575488c144af5aabaec072dc505811c818294e3a75fcd30ab61ad80b20e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ba98d14132941c2cb6023131d6bf91a876942f99417d3a939dadc56ccc8640ad",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "84870abf0eb6c3c6d337862a544123416e842464d04f548397280b52c4cff884",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "5928ef44aa176abdcd4e5acebf790f5fdd97e51f3e7f679aafb69387d9c1aa01",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "a573e084c1ffea5e7de2c618f6eadc71a374ccd88f52459c931cd07595e75806",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "0fde574ede969787135a0fe43cb0a2ba4906317938d98f74afbdedafdc379adb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "d103487a594ab931133f8ee89902a99d80b474cc230647b6468dd2fb45d49973",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "cb39de847cfb622341f2f01ff5aee439f45e94d891e7f468d5a4e8cd4f69135e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "5dfb65f35c5d26b74ef6aadc08bdf9694b4f1c90e70afd6fb4049c1bb1d45fba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c3f56f0ecc3e213d43902b6ef58538c84f13fb85bc88b293f12ba2829947a0c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "94f2677977d338607b35d2969c7314d6dfd72d3e5ecfa9dd2010a548e75e7fcc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "9b1dd7ad0003d5d1a5b0873fdceb1fba20eefe2ed4cbd71a614b2f7734b30466",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "9d09cfdef7c7207b9e44418e3acc551ab2191694b3d3422e715f41527096ca66",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "69fa60086cd9149b16a1141508e382627c1e4cbe62728b36d3ba3759e4656c20",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "60ff3aa46b55ce17da64fa321ba45f47e07fff9969b303cde2a2ca40323d6035",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "aaf80afe49a34710b298615c809a702959ebd1f9c0654377617a9da7b4858db2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "032622d8afb335f470c020458bcdf8be17e130d0a756fb545ec2a19a37578549",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3dbb8b5ccfd2764744a3444265aeb8f973cddb685aab93579158f269633f6fda",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "5161d1d0a10646598dfcc8ecba32f42e8a96c910217166df184a595d96aa6baf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "e1b3d05e857bd1602f1028c1428121f8fb56c939d42dcb64a25307ebd07a627f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "fa3bd763f8e2f37c9624e098038b14bef761cc4e025bbffcf2efd3941898a0fe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "370b4f74a19aa17041c1a8371b5b13cdbdc9765379f2296b80bb3b83306ccfe8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "cb88d76060e322b32976279bc61d619406cc7884a71c229eb845b3d36857cbe3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "61119b605e6589a7cb2769db9a0f921fb5df5eb2cef3373c7a2a96be3a8dc885",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "06e9ccf6bcb0900893f44c567ad0f855a2ae0a67b593b19eb5648df2eca3a46e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "2b4cb3e38b2c57cb01831a86a0c3c2aa6e47eb4f1532ccf781bef473ee43c666",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "1298957458a1f7eb6feb049b1dca64d7a87ae3f3506557427c9e53f8e678fad2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "1066d9868ae24dec6e51457ae282e5b727c6a7b45176ee859ea66bb8ce85f9db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "6e2235c229190dcffb40f66de95fabaace38fba5a9b3cdb19ece8cdac863a4d7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "cursor",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "grok-4.6",
      "path": "runs/cursor/grok-4.6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "8a081f90a33508adb3b554babe65db6dbb61f55c2b5feb782046d1f6c494f909",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|cursor|grok-4.6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "1ecd5d04d69ea38335232448cdfe8cb0e3612632500a24dd84636099aef6e843",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "96790901e4d9dd1d98bf382768843c566074b9b8b88845de585b531aa0f065b5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3633a34759d24f771953674a9332a4bf2b0ace922ecc20a9ebf9a192893289e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "82d13dd2fee438c378843268fd2cdb55cb85707e47fefff4d858df3aaa177201",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "e5b1d80746fa91f4c5644d41ffc579ca7e35deb75fd325b58e20c51fbd2757f4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "0f2bd5b1e9411c1f1a3f057af00229767198a8876155c736ea29af3418cbe07b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f77afba79549b8572cba34eb2d0ac7747c057c5eafbf3325ed39ff2f76e860a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "aed663e1a357eec224f3f6dd8381d900853f2fd23f8fb93a4832ce1248419f48",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "a8f8abbefe5aeee830b0f5489e2109698989abfd5a18823395fd3b5a972c4a77",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "2107bf1e261180e936313ee49ca71a06de7f2968971314a2e121c4f5dad2a896",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "93f5de68c0b1f8b3c1d92b9b3e9aac39419d7ea17e9c5ad6723165c3b491ad87",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ca76d53d36a675d07a42c201111fb187ca998084da0e79d625e52311b43ecd69",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "f1be6177cf1f7d3f0e528c93a30eb2c747435fb197c82b1b99f358b103363bd9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "de9ddbd13fbfdc51b705039542fcb813e62dd00f8778aee4d73fce622fe8174f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "eb8dd49faa151f966ab7c2146b7168f7585ff8533177b85bc3ba1db2e831043f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "a4fb6e852bfd04608b27897d758e7a9b56b5e4ae1e4aff94213ae6e12a196ea0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "bed4c6bbcb4bccd2e03d7e4add4ff4935993cab509f3669ea4d65d21675e58fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ac85f29dec70d36aeea947784612aca2a4a94c7160d44ccf2095b8aec4973b75",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "44d848e050243883300406ae983bdf830e67c8594d927d2db41e6ef15439d0f0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "aa655dc85c55b2a241789f25d9f9a102504ba363ff3f06c68fb107254ef00c38",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "6fa2d7fa3031207d5ecfc941397318f8fa4a977758e63a396b07d81508655ac7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "5d9d2e490fbfb8b6bad27c4d2f6ce717c00cdbb601e039148beaf9b1e483e633",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "7730f51db31dee7444f9c9a13a9c50c9c3cd61c9225e7808e9e589610d116e86",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "a2e78bd07c0edc118eb66a2ecb4bd0af1542ae32e460639e83ada01bf4ed0bf2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "64be87d5c679a3487e1dcc0485683046064a12d680eaf434acfcdec398c260a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f4e2a21ed2c19caf70b84cd55041ea85c41136a8f8a902ee53ae1feb687e0aec",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "df5f9d3b9835dd213b8a3fc5820adc6ba004c2c9c47608ca51498c7e8f581efb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "3400144cc088c1d8ed44629546b47bdc4aa132560484d64259b5053bb18f1a92",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "8fc9a4b62e7de59670ceda439a02a0d50894275671bc1463c6daca4dc7cbd267",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "1735dd2795ade534ad0a7a6b61e7f0a1185f4fe9f3ea269281209efede43dc3f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "40d2fd2a37f0afb8811ced1b7076843b2571d3c1cdb7a0effccc3a6c62c8430c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep12",
      "rep": 12,
      "run_json_sha256": "049681717d471e140b578cd4377e77b9376b848e1c6d04ee7e982131d3d9548a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep13",
      "rep": 13,
      "run_json_sha256": "199f971b7d686f0d882084c6cb1b109ce4241096f28ae35eaea85c4664a0e8d6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep14",
      "rep": 14,
      "run_json_sha256": "692345c366ceab198c720d88b7ee9eaffe2b6e95eaff8d0549f074de6114ce46",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/gitnexus/claude-fable-5/ZeroShotExam/memos/rep15",
      "rep": 15,
      "run_json_sha256": "5f8a001a55186aaffccd45c8b893f4f87b995153eb4ca618700330ef8fdb9ef2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "4ddb76734f38cebb0de108fd1dca19a4c3eaf7dc507f2298113e4403047b5221",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "af2b4e6f922cb1573cbee771f8b748a487ccbcc42b63fb671cd62d7287f3a0e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "922856f8a288bbe093adaa022f1b599278482df4efca7f751e4a24364ee949d6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "8dc21f2432480e0d690065a154490bac62628ee32830f038731d1be45ef3a015",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "ea123220ceba3ec22df0b333482df8fc2173e5bc90b0606c59f6f50edc747ef6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "eb6915156aa78ad4edd95ecb954776eae0079fa439ec4c7fef3513136062d888",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "11998a21ab68115fd02db7373d224bde586b66c5445db2e5c8488bd1e745120d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "5b0821eab3cca7bfda82fd23bc562577997aa01a5de62dcc398d9b6b75898243",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "8c8e5604abedfebc2fde3786b2887586667e8cac2f16ecf7479e6e55160ada8d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "0863701128a9df10b08659dd5f5e968afdf776b603a90c92b20adedcfe5650ef",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "f25e375485f0ad3d1a6d0b0fc5555c55c031c3866e1bccc820f99b8b40bfef1b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "00faa6fad14eb20aeada9fd583336c50ed63f638fb0f439a6db8755a5c4c6921",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "259d763389e72f8326eedad4ee2ff602d6b55e1cb2728fcf5da3ccb808cdd0ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "1cbdb982f95aee2e7e569160e3816cbee3791bf470afe8284ff8b94e094ef4cf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "44a8a280be1a1a57f8f221e59cc2b4a4293bf8a09cb7fd462b899867deb4d855",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "e49f9cd4630fe6e04c7daeeec7df51bc3874daf9a85ad3d5caec7e6f3852785b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "f762f066d0480cdc1717b236335725aba0eb175fce82366b4209a074f2c74922",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f55e4b9440b2b2fc5a4756bf0fcc2d2dffca897d63c1069715e5e8e07b93f452",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "bb483b3b74acdac126a3982b15120a07b6c7316009af58623fc1e3ba0b8e5a55",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "4d25b18c6f9e486f86f2208e1acb5fb368f72243824dfc3476ca73b8da6e405c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "db9c99cdfcfd76d452ab0b79b3e527c639048a03a151c6c9ec89d85ddf1cda1b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "9e4cee4f716494ed18ebdf8cdde401eb71831ba80df2fb41c3a72a064974ae11",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "bf08a5d8aa7884d4ec1e0eb527c9932da310d67fb33f7183159d7c2b27fce84e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "8d79f4aadc6f8256bbefa0e49ea17c79bb084d000b0206d411aae792082536de",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "47bcb97ee4b5107f940d6b4f979e107f66e19b01e163972bf4018a5e0da96631",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "7928aeb013bf0abafd290ea0cb3f864f9baa772c425344ecd7f17207828ace63",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "8e49bd11f35a8bcd859a846895e867fda0ecbe9115a0868556ac02a99185273f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "bd2c6d4c9734d89ff9584bb8347a701020882e5503c600ef24788657c2f03b02",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "e3326fdd7afab39b8bee1cf36e7bd78d4a18a5fc8828e1dcd979b7b8afcd496d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "774b939a8f0bbbca074d483a23e7d9a17419f51e2e7c38291fd85268b8e7f6aa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "28c8761b8070268adab1a374fe033c83711e2bc782c4b8907d294f1263fbf5c8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "41d96e183f641e58c7255b8cb8415a71efddca470e76b709b43da41de80e1a51",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "5f441af08d6b86dbf1e89968859476d87a7afcd4d7fe20008b189aa3c3510093",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "8dc784b211be5afcebf4648b0d8c9728dc31dd6cc619bb886c132b650d34618e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "bc3e1f19bd3b10289d14b12e56500858f5c63e41529b562b7299baddaaa630d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/gitnexus/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "cef474a6b4e979a91209116a238cf7a0cd3e24ade3c5d23fc506849223936166",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "15ee50d0d7abdaa869f692005152bb7cf5e9e293b9739714820978dbfb9871d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "46ecd5597c54835ef9511f39346f5ffe469b866ebb9c5cb45631e80470705054",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3439b06bcc313b3b5b54078b5c9547e87bfce4f08661a5e24f0b089fa630f082",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "56d156a8ea8ebfaceb9ba8cb16154069c86b7dd1d474b5b15a0b0fae4b58856f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "94e81e77db40349f4801c35aef262a18985cc138e110f975b2298261ee979f89",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ab93725af0bf356ef240e49aa51e747fb9f5d19cf89e3698e41cf8f74bd1a212",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "cfccab92288985e5ae55b19a456e1c3e5d192304ed51f871514b1a7cdd8b51d1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "5702d80c4f852c6bbed3e2ab5b9b19eb11e620f3ec623e842b73f8615a99439c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a6511888eb67b2bac3a85db4790bd6632d328afd39bea68f66ec80ffc4d28e3e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "5ba8f09865818213f458885917302eb1b067fb7edba17fec668a8dad112e0533",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "5815a0eb6d730d33bdf202ce18b4f4d4b29be2a6c4c09645baaf2f2ae3c582be",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "170340e9550ec8807002a401d9db674c3f59b128b3e1d6cb877ffee8004459ba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "90f91607f70d7ec13f0822d5f6da931e5e2362f29d88d86d36acf04bd5d58e0a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e774aa71081f985908b01b7a02c939f9e8279f02612dd148e48ddf4b451104e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "773d3784b2acb10550f2c4061947a9d63f9e1080eae966a1b1c5215bed770f1c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "6009d11c667bf1e7c2f4b16f312c60bf920f5cd3b712c00725fd566916e3d177",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "2723ddba8835fe357b934a10466229d9a38535702cffa70aa89b0b7684a3f3bd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "2c6933aef483bd49845ea4c7c89fc01a1ddb9ced7ba84414760e08acc29e5a60",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "76bee452ff81e75f3c8d540dc710e0361b37bc4364c373bdc38e221cb0381af7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "6d456a6f24ecabf96cd51a2f29e3eea0479271b6f475b390e88441c542938b20",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "22757558b69d5563fa422060ee36b3b85a3f1fe42e7267a76dc0bf5844bb324c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "98c5b23814f92b0f84d93c56e5167c7f317e52a2123f55439d74fdf1c4c2b530",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b766255a094394b52c6ccffcf8f9db8bd6fa479b5a9dc7cbb02f4b7b59bdedf3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "0523fb08368e44fe8d5ed1962af1cd3040ecef56f9fd6c619644910769c07cd1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "472e6d9226cde403532f3e09204de16557c1278b1a6703626652108a28f21349",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "5e8a61328b96440fa0d136f1bcd8d1bf9b9881cad713baf144cd7e23135faa04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "9bfaba54603f575d9b0635cb3b7ebbfef2f1821d8db4bed533c48e5846dc1be9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "ec491f1ed9d9d631cb71194bbbff96a317e61004d01bc2ab3afb4ce4eb15cc31",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "590509f457a58e7a69c859c7e1e53edd2cc6749b7b11efca487c57601c10a58e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "12a76a7bf4470e89a573a9d633f76baba4438448ddbbff67825fe29b0601d383",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "951a5fa24e9ad5185708cd38efb4c53ba1229535b978174894b6682972ba0899",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a5bb7bf7618a2260066af8c293950930d15632e9d5677d1941a7403e05890877",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "b704510f8e17186d35a67323726d3a021187886104da4054919ab579350a1048",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "eefc12a0cd86d87403b5197bfe78909ad9932c8b44b15502072942775fb4f106",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "186fec86403ca10e16765b67af2afdd7a98b0ee553a88686605e91e43350e11c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f093d4e424b580d19c3f6ac5a06caadae1e90633360bf2b098edea4727db07ca",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "eebe7fbb9ca053921eea2c855f02ffbbcd0a2533a5791156bbdd92ceb0ac32a0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "55b7a2435eefbb3a07b24d991959f937c173e8eb44171f76a78d27c14cc2d3ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "eb069f3af724fbd9491535b27809cad782b69222a7bd41b6145dfcc97b298ded",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "efe40afebd962347dcca6b5d1aad2ca0ad63e440f33c90c95d4950925340d4e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "22b96d75a1358ba348345176ccbf51b27f2548e6338ef63aaedaecb768ff582a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f842552428b85761082f78579019540f576055b7ff3f01fa8744a3e500c6ca0a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "0449d8c0b910852c7b5ff531835ee977f6f6ba125b13ae4701cdfb9a74ef01ea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "28d8fa5f09551428dd34920b297a57ed42850f58059024aa5487eb09bdebbdb8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "2d1a2e227c8cc1a1db78cbad664de120a15eb44440d07c71ab57ed2680ea995c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "038026811bd420adb565ee99c701e58c3c73fc6fdb1e676879f94cff42adfb26",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "def4deeb9757798c4dcd7fa66135fc48123c4a17a7f9ab53813d0561a201a172",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "14ced9a086b27beade5dcce9b8e18be050a9ab6a93ecef633f37a28e6a3e849b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "7eae229bff7bf374d5126057dfb0725ab851458f0b60f7d1ed8fb508f175395b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "5e65173a8916ca8c10bea5dff17c35cd8427dd541539c0a9e22e15fab0e10526",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "61c6a21346663e3da339ea401d6e076eca58cbf8bc6774163f59c24225a181af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "6a652683e70a4c2096c203c8977e4a1ef54f3c9248fbf0a7726dbffc2cda9355",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "ec549240b0597911822581cfe2242a833458841c65d5be95e1f2273ad01ee3ae",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "9c095a53d7bd1a2d7885574ad109c33a5353be6a5335da86b57ec02c61d85c6f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "534d5f93ee4f6748f8b541754fd0cd87d7f6473731b70b8ae94fd041b97d0f2b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "68574660d87b5ed1b5b21217d1fc3ac5005649a52e2413d4ff8a943a0313c800",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "72027695037fcf19c4366ebfffe6fcd2e61959cdce01171dc005dc7ba143c0b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "a19e3b95211992dfcb9eb8729813ed06e87528d04adb518eb928561f2b076ac9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "33ef87a8716667b09881d17e34ee61c405d11a5120c973b8cc6ed8968c58484b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/gitnexus/claude-opus-4-6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c47ce6354539d14b30abc38e2e4bb2798a1d4fdb36315eb3b902db42c57b670d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "3a125c5eb37ef29988084a1ed0ae0b63163740e21915b95070485cdc52343a52",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "67bb247bc8852696d758a5e32abb5b5df682498c896c1f61f3e4b1a51f38a126",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "e113e14a46592fa9fb2a2bccdc76c596e92fb5614e0c9062fc82dd6dd93927cd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "8f71d8c92323aa2ede67ac03e108ee32ace05f579522fc70e808491bd863af74",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7f120579ce426d1ae064720937b463d98613745874c79c00332dd4f7743fedb4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "0d4876e391fc2af90e2c42afbe224d92b16a94e98c687761e3b348da89387eeb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "08ff61a9064b845856f3944414d4b9b363960da19dd67eb8ff4c2462a9c05529",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "1d27ceffe5938fbd3827819bf2f9fb81b7e77b1512ffabd46a4086a46c067836",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a1a136d50f75c442b2c599928a808422fdb8b121b2274900b2c8745f4ab711e7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "7e4971d3c93647aacefb2870340d32feb5b494b68cb5e1203e4cb03fe41818db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "12b897073238693b0db085369dbd48cfe896c06b7b87b4771c776a160c8752b0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "955ef31099e78eefd22101ebe3845fcb65d01d4ae6435b8b0abc48131fab6fa7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "796ac1cae07fb6d3be50281e3f9fa173ac19c9b37f5e4b49d8ee13197cdd7f4d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "5990219ae0a302a7e4a3d2ef068dbf3b3b6af3ae996aee19fbbaacfda4438b83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "e04d9fad2c7175fec0d53ed34d238a18926b1d4daaeffdd992cf002295e3f4bb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "034f16ec1b1596e5ba6e7b8a162b34c70a4ccc1e2fe7c5a5cc78c2da7dbac181",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "76e80e5dfc4b95d99ae75d375e9e26a9c314d7d86f1580d056e4ca4b6d06aa55",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "fe0c76b59350fb6df89566f138d3c8f38314b60fba7fb4b4655eae7dd4ddee99",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "8433df3b671b3e65e590b73851fbad4019d8b5e48e1df23c194272a0783c9093",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "83e50b4d99fb7b20dde25e34327b35ffce4e95e0f4999e4750eb2a737038da53",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "0a0888b44f63cecaca739770dff461a55bb22d4ac389f0c79afb75eacdc4e337",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "d85804a4d267e635c7be858b875dea7abf8794faa008e0f1b9b9728845d12d86",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "e1a5f184eb3ae4156e0161c62b207fbd37c77d1dcdd631b97fc1f5deaa165fa1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "6e23c6feb4fb4cd51d89c0ec77ab337823bf94e7a2bf131cc96d5b015f795462",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "468569b7bc8f362178473fa90bff9013a1ec62a45b1ab4e36132e93189ea1357",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "a48c89f649bd1365081bc3be95bdec2fe26f46465e8bc5a61aff8b936d7b9482",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2417f4770706fe451afbc53952ab12bef729adebb44d09415c67b46d1ce59bb6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "fbaa10696ddfb22f59460be44c98398bc54f32f3c30b030cdf1d75b32beb561b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "cd3a04d7a9aff2f92f4ab4bf45d9e82e5ffe019260920ed1c77ec6a8a5fffd88",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "30c190116b21002d554afdaef69c5161287f3bc6e63f36d5775f4db5604bd8ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "66f4a18158280e9e53fd3e0c05bbe8730ffa9bea5293f1800031d6eb672f6ba1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "700deee3c8ff011039747f42ff9c2b6fbb1509475548ba13bfa4f1bb5445a778",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "5b1e85f543d2bcace2eca29bf0f242e67c560a9dd142a5df06e764296f6a041c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "372cc89a313e922451595fe6f524addd3aa37f4680e1876d55cab9b03fa78602",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep15",
      "rep": 15,
      "run_json_sha256": "f10d74ec4c1764997b52f0e33dd084192018a5a75f396647484b6b39f39d55c0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep16",
      "rep": 16,
      "run_json_sha256": "32e45e359e89342b44cf8f55fae57b489d63705faaa406c80627924faa10dfac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|16",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep17",
      "rep": 17,
      "run_json_sha256": "6d8513eac681aa0ad7aad097646b2b043ebac116d60dc6a47d7068cc53435427",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|17",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep18",
      "rep": 18,
      "run_json_sha256": "afd455d52273b6b29e97074b991d6939c94a7274145ba90bccf2bf4550a7c2fb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|18",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep19",
      "rep": 19,
      "run_json_sha256": "d05b758ec4758274d9de1f8232f7b0491697f4cc37964c8399735f84bb250581",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|19",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos-hard/rep20",
      "rep": 20,
      "run_json_sha256": "0e56ec2c7a492bd2277c1156f19f259bb4a4f5f301b91f959a4bb08682e0b87d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|20",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "c9e8ad2ad26f71a78e9b5a539baca79b5cbde7ce4d42085b288ed0f6fbc3a76f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "82669220f049a60dbd74e7dfc97f45ceebbafa8a5b61fea9ed648164d53c7ea7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "0920146992827965f393ccca308eff2f3206f442498e18266cc45c0234b33ab5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "b29eaadd3249f362b1e0a743787fc66c67deb59ed11afb96de7a514c9ade4ce1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "9395972afb50b9d92aea3b6e7fdace375e0100be585166c858961e3642211ad0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "888cad8fe37ae35ae3c0ca32c8c43030e8ed84aa5bf83d38e2e7c2794fef5d29",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "99994de9396674b98d87d22e079bf546fa77b3fdd4208c255905e9b94ff0c0d3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "677cf6e9d892d2aff50c256f6a2e6d6fa3d9ac6b67e8547687b92ace3443aaa8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "f5f85db65cdd3dfc585b02983f0b4fb015d90d4374b94cf66e9d5d68cd4ba151",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/gitnexus/claude-opus-4-8/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "1dec182ac14e6e8150b02aee107e99f13bdd83a06582733139c3d74f00117331",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "8ce6ca73c7d3f95e2842bcff01cda101337e7c764fd211c5b455d4d6952e04f0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "ef293c467a1b46d0ecfa233fdf16ae465ebd976c482a16e09f812a152eaf215f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "c2d3fb5fbe7abc2c579a749b3bdc6701f58bfa5ce7aae61823746d7d1ae2ffd6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "0190ab6fd29c62c9cbb7c7a46bf0dd45bd888cdc62ecd94d802b6498751973d0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "d18576266970e074beeb9be67a9553a45f027d94df86949817b17beb22e65ec0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "1befc0d3a0d7e9cdcf3eee47987abfea2aebbbf174fafaaa8346dca9c1a27f4e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "71305bd45ee3f22fe281a097452235b80f41fb4f947f951bdf90ebeed83cbc38",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "463e1bda822e28491d4dcb691180341cc3a198acff796f48c9bae16587da6136",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "57956add692ce0a01db3d89bc5e9c5852718b5ba325962c64ae68fabb1a5d7c6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "ab0346b7dad7cc21a011179ab49409c51f60f8371b80284e9cf5680a0d4dff26",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "a3d84d6a4223112e211c5dfb4453eaa7d6fef34cbdb16cbf20ec45a6f46d7260",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "343bba25fb4cc11fd9885162860dc86b80dd41d72dc32ca38126790c423db7b9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "184d17f5581a6fa98c73f8647ab786c708b47ab303b67eaf166d90a3c739c586",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "2110fd5f5aaffe0efa0acbf0cf82ea88546a741d5efae90fbbbd9a0fe774fdb4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "449535f7cd8499b362a869f5ee0b43ba6eab53501187a775cfba20b8771fed3c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "e4b47df33eb39ba41de8933872a0475d9fc39375b29a85610224f3a8142904bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "2cc4d865db29c7e848d7bc7e8dd200c7b91298c59e16d93454a5a0ade6d9c39f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "356c961763a1c47b668ec1789382b3f63d1d0fd180fdb64f0a96647fc4233da3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "f27c8191da4d9474f3b72a7b74c4bbd4333bca9267b7b603726cee60f108b081",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "3bab2fc3791b5bcdf9fa6c3ebe0f7d49f0d1f642ca0a6232c5f7b5ae5aa18a4c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "0f35c051f99e6e77a0dee8325fb09d9e987a69b971617c8e78b933ee58b253a8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "13b25a7e87c53a6ec485b07dffcea5e699f95171b9ecff012b647aeab06f9fcb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "4afa9b8917f2f63cf88b79d667b860c820462278eb4fd38d5a69c401a57c785b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d90327b18f627486f06d9471541139ca53508de9c8b1160f87f043ebfc90d831",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "cbe478945be6144b870e77efa789a535710882f5d70f3acbebd7c4ecc24df4df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "aa815bf3f64203839726dd7dd160361f0e30202e7b4aee3935d4c1a4c65b1fc9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "f7de6324eff4e14e731e1dada7103a63eadecba1518ee0d5643c1d132e85d2b3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "0beba327c652d0065ec0daa020383c7547b9936bc061231a1a4cb64b45dbba77",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "5c9c6843c1093101b94dc6b0047ea46b228d7fa8951c82f9704cfaf0911a6fbc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "50906e921f4ee29aafde52a5139a768aea73e5a0f98e04236f1ffb2939e1dcb0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "bde54fabae67f4eeb40229b6bf9d1fb30b469aec546c268537e7c0cf5bbef50e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "7e78b674624da598dce7334cbd27bc11c23d683713dff0c9bc3684499381dc53",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "9cee24de9030b331791324973e22281e27ebab4316151c3790ce1a79f27e49ab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "31929e2ced8eaa59e94ee35532e3e9fd5059ca12bfbdc4f5276a4bd7ea6b8eb0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "4c90d35c3801b7337a69a8eb5392348cf75f0101f9a05771a4c8f4b1bf1765fe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "844adde41045d9afb1be195c68db2a135c9b143a64189b54552a2a5c916c86ea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "1532a34d685a038bdb2ca9da9ce586073e3584b3417d1098ef63fca4aafbb204",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "89be7c0cf9b743478025b1dedcb17e1f4e4642f276d8cafea4c2abd92f0b7b76",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "6568f3586763069350d5f5402a1286460a427be16b68da46b4ba025127210820",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "0454c3cf6cecd29a679cff1b17b3251889c400b54421db61b072dee04c07355c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "0722c8c0f162f3ce0c000f0932520922cf6394e2f8953cde1e3ff4d66fc3b2ed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "710507748901a6f7a0eb28a329cdb5893f5e9e5b52e3d902c67327b248b95ffb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ce6c33768d566eb8a3de6f5c36622b5467af2f12f07a2c91ec57348132d01ec4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "728dbfb1dca8377fa003bd8bdf2601514507a26fbfd56672f44aa22586dbc57e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "f2d285e3845cef5363c007648d1da911d2543ce076af541ad3a261866889c2f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "68150f1bbcdf2b30adf639df849bc0920b29933d1aa1069543f03e937120d44a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "9d909cfab2dca06a53cd6d90f12e1e2ea1d2e487f0e700b672bd7892a22bf061",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "b50c699bdb423283bce2c0f696b053e04083b2cb3b386ae539064d4cf26c03be",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "3fd2615823a890f26f7c7ba9622e6b30bc00f8b1806ff200033513036981a9d8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "a30cbccc022db69c022dbb45fc123080fabe430ebeaee991a67523355e36c836",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "0ed74982587edb7a8f607ef999a2c617fae25c8f66a2acc5312b47d62be87f09",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "30454e83d5d148a1f4c09fe9bd94711f5da3c4b14d12c52774cdcc9538eeaa97",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "eff9b6c1760917953612a265cb4ee17839dffbb71428e0adc93b9b6619073504",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "b7e8840e33d9f0d85cf0629c9db925d80b4529b7152d6b591dc7914b6271ecd5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "67ea09d837848b6987dda8c212f211be5cb2b262c2985e296b4f91ef8655d8fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "60f4669da3d6aaf34044bb8f166206440413347a80780c82d32488ca70d97b70",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "5b8329809294faa958e95e5174d14f4ab5299ea908bc69d83e2c9fc443528c1b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "15b3fff8fb0544edd3f972d2f40fb879d1187cb2c111d0991eeb8f762c6cfcce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "741e12ec96234c0ff35034b807bd8b0c2b664bd22f2deae89815a4079e94269f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/gitnexus/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "52e76a525ab5c06d5f0b52889bbb6f752866b6bb534523ed3510146c8dcf3a4a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "102924af5b2fae1566b7f5e05643f25a5566cf3e0df3382d2c16e558cb3cf1be",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "71bb9b0fe5150c85d8aecacd565508a7b0a48d63c54184c784eea60fc160c914",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "c7eb529d977a86a049326bdab1a6ad8a1e9bd67c9caa137a210e578a1e136879",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "c13a47798333876707043343fe987d47eb5d527b8fcd8f128fef053b2d3615dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "6b0563720acea8b8b54dfe32f85b2c55e7169e735c26db02177bf14ea90dca08",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "2022c5ddee25131df993e6d3495afb51872a7e4dc2f1522a46bd4692d6ce6ac1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "eb8a3707ac81421e80a7d75456c350484f6a3581174b558f8daf2e49c44d92e4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a6e802b737163ee95145b707236d9de9530607525d8781cacbb904b7b7cad60e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "6369ffcf95710750ca8f3e129c70fb2705a2086f52b946da210fbce8a2c521d9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "d81a5fc6331dd983495df13f53b17b8ba69049d2e4322afb5e22d5be611c03e7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "7a20827775302dfc5a4b0ba97e850be7d1a282023c17d7b2693acd515fab3b8b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "07f7802102a996a28e174073ac398d5b59b2cd25b653b34589171928db5aa360",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "1174e00d93ddcf1f3e59935f589f0684f5c1ce72d6a18d706466bdb4167f0516",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "4ee6c802d88860cb44fdfb4053e0f9f74ca3b5de439174d3daf5b5ce049431a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "4424fb864fd7647953509618216e63aca50dd752beb7241dd2182e951a97e023",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "8e73d54b20a460fa7bf21e6e1f2e6ec656da409ab2362aab061674c091eb7cee",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "b29bf0000c9e27abcd84140d8e1f2401f2c078a0e8d71f62470699345094e81b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "93787ff8c214275fde9e47b05c4c658ed03e1149513244aa0ff152c1bbbab57d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "526f40c2e6aa327cace6293e5feb168157042cbfe5cc05ff68563f30157cb46a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a4dd028d91f83b03cb9469698d3b87e41b793580bf05ec77cfd822149e1b7c13",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "0da080a9c9d477df4e2de891e31e0c5194dccb202f11d0e66a446857cc4a587d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "4aded5ba94c3f5c4c91e08bb7fb197afcab3beb4bba3c330b99bca61fa9a904a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "4e1753a4a516969354a0dcc03e2071356a5f662fdc9952640da0b4a5b3b5a04e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "5c5c8f055eb57bbc8b4b11c67d846f60286fdd9ff7781eddf964c36890ebb48e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "dba3340ff680a3d4ff1df1750e672c23fefc9eb98abb52dc149e53f0bc42a5ce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "333c05713b3908b29f145c217663f7a838456c84daba84e6df10fcecdc1767f9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c82d0aa119b29386094384247f1a3cd430bef7645f31f4d0c5020e26b6466fa3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "4dc4fcb9fd540e63797d444594e5e1ba92021917507c6ce5867055a25454468e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a82929e57448fc26efc4a532af5b0dc75ed878ad56392bf486d1f18a92969386",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "bea46fa1a7706a39da7c234abc596ff9921453d63dbce6368d61f39db6da7c35",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "94924725e1519073739d26c73be6445f68842bdaee2f8041c6a17de1ee28c972",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "1beedb9d2fead7af2d17cca3df1cd7d1e1976cd60a4448c3fbaeb06dbea3535b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "4a99979256474fc56a623fb800dfc12945aab4cc253f8823fa62a3c9fa481d8c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "cc078e8391e008b4211ef39eba1507f58e3a43d08e8fe8652155e49e2cb0f059",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "12d94914a60c994294f595c0ab5a8b04a401336a596a06818d03b1fe4b736754",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "16f3e5c539db3d6f3cb69c6a0cd1e5f1297363c0104637245bfb67e3a3c274e2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "ac07df63bf2f41ee20caf032b7478c845ac12ad3c3b5c33017850df2f7216de1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "835303b99e4e8b5c9fbf5252b9a32eb9dfd6e7a1f16b7dfd5094d501f66b581c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "95b87c0bce3d55d23048d7c8777959611513ac492b56962fda725a6ff9a2b708",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "gitnexus",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/gitnexus/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "e5bab522681c8fab4c57bc080ec33e84a9f6864e51a4ed0a7ba32f771dbaa0f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|gitnexus|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "5884498747aa94d266bc33937c5a987743d2536c2dca157f970ff40f3c33cf31",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "450fd8ba17582e19c4a62530b95e9fb6f7258b649fa84d2ed3955fa9ea8dc5e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "2a752347fd376c9a14d20ac10b925f73516da692cd3d93eb367e2cfcd6b34bf4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "e53ccd6068ff209ea19cc65205016e6d5fc9e509f1f1728f35a7271bad6304f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "8e236442410b4d57427b01a81917c55c026a72f91c82b84aa7bd919d69a554ac",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "0ec07badbe4b1cc3e766f6c0c1a80a7c82d8c59c4fc2110e18e283ae84989a5d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "1c396059c00c33aa9dc4fb63b17628f6279e248fc83a2bd4e256073e4bc3b30e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "dffef3e98b7568a9a8afb2806aacf474bdeae3f069d16a831964a5971b9d4f8b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "98d411aad9171a9193b866d63a47caa13f0479eae6eea2652f0c204f18ddff00",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "3c16507f2b3667f70fcc6249d4e4669235827938ddfd1432689204f575c4590a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "ab15d243842f875e2aae55fe8f193fc498ef3f168f29feccf0307d33f5cb4d16",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "1a832e9c243b94cec60fb78969270a33dfbda6227b80a0e9da231b7ae25a689d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "b0cc581269a25f5916fc753b0df734c7105bcea96bdc3c932249fbf00f33773d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "907f3af360e2a51a19d2436f55e61e9c67c2b93a6b3551337d1c1ff20b49df26",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "339baf89d188cdf2a6b380adea46fd8ba96fca53ff3fd9077769b292b2bf6cd3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "1f921e2b95f784a2470dadde5d3d1981135c75e8918a6648484883004b48ba53",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f701753c1e1d790737d86b7405a52ef09b359e1b131a75bec3b2774438dd868d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "0d221dc6d1b502b3e57f484332d0e0ed7b1a80d4a56126138f43414ef56f3409",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "0d7217929bc9696f8bfe7d13193cd86a8509dd9b6cac8466030e0db5c0384ab1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "aac315c64bcad56849bf1b571796d8c22b5660cfbd67ea8c21c710594bed42c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "b060574bceba6db6a6609b7df87a5fd92d26534af5553d4beb347bef92911454",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b1449cc9dd37386f6fbe0aac9c2fa3c84fbb20ca4ff3791e5e1244fc517e8750",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "46ee2669c3aacec168fb1716d17cf51e788faeb3220169425e8cbea317b6101b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "ab09aea0fbb3215ec4d39e87b40d23bf9f8bd02c1631952a8d36d524976c2137",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "7aabc06714488e28101d072b789f18ad470dffb89e14c3595a4821fdc4d0cbed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "6425ab1688896197d57ac4a67dd9ffa94fffb1e166f9e2703c9cefa4e9fdf8ef",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "8561b105b9e79e264c8b5d8b2f89d5435eed3db2c4ed3f139c2bf0e1a6cccc50",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "888c9b48362cdbc49c1f463d3cf6b4cbcc4ae8e3f7e08601501366acb6ebe286",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "113a83bee9b28785af36478f836ab0cea50037974c637553a18fb603d2bbe374",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "0a45799dfe190bd93dcb017917a623d0a74399fd2bf7b3925a557a128e8853c8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "7ec1e93e9d7d8a4887b5b211f39acd63c112504484bb2a7382a01fb558f65d04",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep12",
      "rep": 12,
      "run_json_sha256": "2248ec213010749381f1082d74ced48df08b8b0e2b5fd98a57ec33d68950d1f9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep13",
      "rep": 13,
      "run_json_sha256": "c16bbe32ac35a3bd71fc93c7285cd4195b762a88ac242409c320a1eb6daa2162",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep14",
      "rep": 14,
      "run_json_sha256": "34932ba61ca60a944ebe67d3a2732c2d249099971d21f7785ba23116f0379826",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/graphify/claude-fable-5/ZeroShotExam/memos/rep15",
      "rep": 15,
      "run_json_sha256": "8cb86ba9e55010957ec7d55c0325041b169cbbe33eacc1a3772f650248a061a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "2bc639aa5a934b81e749d5ae268fa8e4e48a4c30b8b88511837c3393f009cf79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "05d2318bd750a919e14ccbece71e955ef43b3bf4e367c6fb109bee701188c4ed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9ef8463ef58642de8ee2746069d3093e266ea6913bc9fcbf1f2832391d943356",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "01aa9797de97b94785cb3a3ca65883343e1003b0a09d040d338a67fd4380ea21",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "b356e6092cbb87835be5bdecaa47d7147dad77f13407ea6e9a662854178b5aaa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "c6bc2d53bdfcc0fff74a0371e2e9a5d4f8f01767f8637a3e61701955fd8cb86d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "16202939ce04e729bdf58c699997347044075fb51a71ad44a883ba969586cd11",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "dbc02dd092cd7b6e825828eb867538fe41fb20526b80060d05ef57ef32594354",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "32170a5955ac778aff775e870682d3761b20549baf70d4ad72e4d625fa9de080",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e7e67581c9096f995716facd05c292ad90344c7384bda4abced6e3c0685cd304",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "f8e2212b171f96d32f2f5f2aafbe374e1bf631a6ed82d86b061cbcc66a88f429",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "fdde20b719d405a7a555bd32faf03921288ae675ffcfe8b7980d0c09c95f834a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "eb1a64929f66ba59d88775a259d2c3de33e1814c2846e334b69e665ad6254998",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "e8f03ef038e09d73300bf3c2f08b5bf08bf1d642b9854192e6928ae9b9bc9589",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "13478624352a8bc8b29ce7633041f7beeccdde05629238104427028e6c4e7ccc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "1930f8675b25fbaa24f237d042044ed2e5eb8c149a988f5ad194453dd5439ca0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "af10eaa32a1003805616b9ff578190c4bae3968fc640bdf1b92edf867f319d08",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "ba8b2947d96dea4c576563998960cc6d24546e9840ae0ee9eae4f83a2022e6a2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "e6de0a90b6da5b12e77cf2c31744cc9860e84cf280e2ec1028e93a81d0827b4d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d5cddea5dbfb2601d558eb6316c6fa9ca8317d1bec56cfc88f07092c7b3df254",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7ddb566854d28110e3ee9deefea651fb51fc9c5bd95db8c739be90c37fd2b2c3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "fe5eac7d7bf997387af544db0fb20640e2583ff163d6e845242efc8d28fc5c35",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "189fd7ab7b309ee23f53f9d172ecc29f07ba90237d21ffbe09bab9f108b59c29",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "9987945e12ad2a62488b9bdfceb42e05ce3633812519d57060f40ef5ec7b756c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "20625a5bbfaeb3f8257de058b717f5d38e1fbce0456b3e779c0a45830ddc1146",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "b0810cbff7631d12a6b463874b5a2d3d405e5e1f4cb1a090e89ac46c20c6979a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "5201eb28e8e698b7f990c434f40c24cbb0bc503cb96acad4e33d8fd196df7085",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "72936781380a118c8075a2c1b95d64450e366e1aa9ec9e5a1ac9923f158f91da",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "4f2c0a36f3f12b41f70d20d783e8dc3ed7dc56d1fb71bd419bdf7863f2753193",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "88d2b3f6cdf436efdc9b387bf3307a0b215dd6f74f6ba3086a3e4c25856c765d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "ea588f670a2c6a8c5e96bf059b827b027b24c852daebbe7a72853e95b6e4f497",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f8bf07d4c4864eba75c33f81b737a8fb2902560e685b617d00e9fe3100353bd0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "0f80bb519fc88666a9724ccc796d83732c309f981fc176a55c5110f5a84346cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "9c9d0043b2dfc209dc31a89f66f869f58a022a75c48b9ef8138c4d143a52e4c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "b38e125e34a91e1a56a4f2fd5f105372832ee1036f6cb841f69e67eb491b69fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/graphify/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c8cd3fed4899ccdeb4f919af1d79338b0027a275d5fc33c9ebedd67d98fbd6e7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "82668eb7b45f837557cd505c9f66aae8e1505e304618375dedc2d0e7baab9697",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "938936964ef9a5b3f0a082dd3674f63941311bb0abc21adc012ca1716951feed",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "7c2d940748b9b807739f7f8fda45480a7c097e0bf43c4bad737617d0e2a068f1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "1fdf6d41851fe251d8c8c4967a1819a2eac7d099a795815dd13f4cfe29340218",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "96548df121f31572ebfe91797b0e168dadab566155337c4e999d492b19b98ea1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "dbd4ec9d8324ea72bde2e12437d0a9544afaed3cb6cf6c4ad759cb8654bc821a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "a4bb2f0f29d4b63ddad7fa44d9b9fea4bec086bc6e5d71af99e920a260696724",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "1a815db4a880edc600216de4fa738153533e5202f82c905848cd36d9f7b5fd5e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "3add5bf902dd2a0034a96dab741b8d109e28d5cb5e6b384586603b5dd0a71f28",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "7d2eaf165225cab728dcf626ce24add0bc8e792e565b5de56e8de6662ecf0d43",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "5477f32e1554374dfcd57bd09b26186ab50a719025b05239d679723a64352346",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "fbd94787b0dcfefac60f035d78a3333e079cab125d4adafa49f26d6aa40fa713",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "6d76ebb8727257d58de75279fd3747ebae008c9bdc556b8d02739e648d2290d5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "45a4da0b6d1331c1b23535f6525964b69be8774c3e76e89f8a74f9528719d71b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "d3b916c0e54fe6fe22920c8143fc74dad8e95f98bb04b7571d3da6af29d693f6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f86fc63d1f3ac6a49356234f0ca6a804b7a197718681d505c74fa8b8cd84302b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "5140968964b17193eb5d27663e2ae4a9a168042b99794d0d97f6d9038bf0b928",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "2c9307e19c442d9f35ac92ba55069f7f363dbae0624351e61456771003c2e0e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "70b57ec5f45af9a54f961ebb1d4748f6304807ab98165146ce3be6cd1012703d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "e7f0155438c2a5b8b4515a0712ab80d8430aa844a09a3eaa57ceda85db472906",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "fb5bafbaad384bce3b5eb134bc00ca253331e9d2f78bfc58fe5d5e1c4c9a241d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "ec7ff56b4f662fca6a968ec36e9756c442b3f8cacfa63503aecbb3a170c858ba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "25f266541490eed8790871a156579d8539984cfa2efdafdae277a1c618f50ccc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "4c816753cfb6ac12de675e1d7cff1e548ae3b3bdc5f7f7cc16a5908a99d3aae5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "e2544529fd8c4343d7e9a8a722eb413305b0207f6ce2429666197c7d13d83405",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "334374fe2c1bbb67988e22ff671be7999849a386ee887048595225ac85d81d57",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2ad24ea2fc47babf355006c18be311749f70eaa8e3d5ce9fb01ee44283afb7fe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "60145d0a3f5be0073ce1d7870bdd221eeeda25f64ca50170f8cfa39b222d7d9f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "e9be82dccecae4ad0d6891c8dc2e11deb3284ebfc00ccd6d231be0bbc8909d12",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "a9d1a9eb57f46858782480d9259c9b6219ed3b0cf00ee05a49a275445ed8c4e7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "ca3075aeb6f23e2776faba9cb270d9212b5aee1d8b0e2750b988cdb02772eb57",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "c36ca45d69dc618705d1b8a985f7fccfcb9934830ef552f308ba12a7f6f89178",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "26f817e9622708b677e68b16e8200db6cc8399f8718d57210a2459ca34ecd62a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "561d3b1df602f5bc64af3cb5f0e7f11fc9f273fe82a6a3a582264fad99eb1dbc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "9eba721b54f2c381e832ed00e39fe915788b27b24277d6a96cbf44862090dcb3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "867652e3e0383716ba82becd0aa852a28c5752c495e2f8aa2bedb0aa6782de69",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "81a8b1e5fec320dbc606b2932a3ad19c91ad818cac421c869af1e74255b9bba0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "d3e8417bd1c3a8415fa917923ef82d43c0605ec8b3c9d5e8001486f4a81ae715",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "3fffb91b6a692168591c6c064fc270cc72438c7a3ee6fa9c4c65e3167ee7def9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "fd41c8224ab13c0ddf32a8fdc3ea4a0e38ef355254afcf236aa814b2623b7782",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "80dd6d075b016dc6af823647106d01d0cb53cf802e643174819d1d8c09e81e6a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "3e859792ca6e941d252e7bcad06b022116547f23b14d833845fd0786cf4d714b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "431bb0866223d7ba1ec4d94d3517ddffd87157ed5e3dfede49ed5781f093aec6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "f9f060c8a7d68056379fcb518fb8a897accd85bba534b22ed87914a61c27f663",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9889a568235cdd64e24c32dc6a474a6dc482226e19d8bba967a0b30ede158124",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "335660b67c875eb6ee2c7c9fe9b940c801952693901dde82515b407fe0418073",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "b579bf0302db19572b2e963b221e7c9a49f9079b97e5f2dade6be46b69942993",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "d0adc373b11f3d73cb16f62a03a5c669242ed2012d706ad42cbbdf7bc3e44b46",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "f21b37e0b31674f7b64a927077acdf25ede50e3422f7ed949306e5d822c79b83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "fc716339cc28c949e9929ed09b2cbb7d7f80fa86154cdcdb68cdde277dd5b68d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "ec7a4bb97db3ccf9ffda5daac78ccb9324b3219dae760631f448c47cdc53331f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b115e92edc68f3b526d46df47dad8de45bc35c3ead2369171ff20da09dc90b45",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d31c3b5351d1c89f2207ce518b19a35e18da709669f9848bea691779c752088b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "908c7b6fd4314cee620c2606cd8a25d820f25ba8ca2463a715044c715b7929d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "2cec502d3b02077a1cff6ec31a2c6591ddac3d9d5ce4489d5ed2b229504af7d3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "9ed1a6bc15e84439a8891dcf3266d99108fd449e8cb05e282d97e1ddfed16ed3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "46037c256ed513aeeda3eec00b6ff726208d6624e6da281002b114664bc3a39b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "818ad5ce0e6a5548a28daaf6e30deecc8bb3f18b806225cfe32dcf1c2054d6e7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "551bf40483c8b9cd8703f3941ec23b046e22853dc0c829625af241b9482a7a7c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/graphify/claude-opus-4-6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "7b1f1c33a0344d2d3f12b7e505dab9799d154f50421d4b38978e33457cb4001c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "5e87e8ecb332c71ff337c4d1e24f39139262d4dca11c57551f22eb22826c2893",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "2c3827d13320c0a93a9512e2bd9aa238b5b087cc97722c2a3b990c404a6fa198",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "9c39ef26cc573552e0a27c6123eaab55b65f31b388965d18ca98631ea2778d13",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "815894aa77a7461b95f3365b5ed5be3d828d9083c0705c31a934277392964656",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c4a5a0663a7e33692e5116f04edd6c82e21a8bbe008460b1ec70c824b747ec92",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "4875f4b6794124f28de5ef322aaaa75bb519eb6c34f44330aba6b2ed5fa73fd2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "03967a4c290fad1649c414ed0148b2371fe8959c67470e324355d7c81d2a5607",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a18ff23f6f201e9a1383e6678c405a8976acf42691c56bd25e285b93d9af6d28",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "3af46721e3bf50612c5897b65a8d6317c4f8aa94cc215923a38ae92979ebe075",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "e0ef4b8347c2be09ddbe4231db0bf44d7fe4c2377af483cfac5a93b79a2ff698",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "a6f141220c5b981bfeeec26e0accc6864ade2f2c0397835629b7b18c6876a6bb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "2a229e7f6f004740eaea6a5487d8ac10010fc2e995c2a0a43eb1246fa08d725c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "aae7c2e03cc6e1395d069d5b108ccfcd14e864e6670cf1be1948d46ce18f59fa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "cc35d1250a698d15b0026e2a8d7bb5c9f924a203d9b93f02b9811ee51a611c10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "42d3c47ad014b27e74f718ce38bc09e998bd924852d7aa50401b46ed79b6ed0b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "f0f08822cad95abdb0d54dab796d94cb71d601578cc6b24a3b33235fb1ffb8bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "a06b0acaadaa62b1237072a16c6988aaf5761bce9e57939f1b75d842ce77f621",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "286d26fd11ea2a4b2a49a4290e1bf2348cb84af790649b0e22ac554dc31d570e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "38f6af2d866e0645565e2616888eb862587467d53fd22c51d279d6f61d0c0788",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "f5a9e98ceec8f12c95a031af4892ae756cde710b915d01f7c3e2ccb8390971c2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "b828bfd3755cb2fe0be06be1b8381e32d9edeeb5286d519b081ddb7d4e21c8c6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "d095ca3b4e70605a13c438b48bc2dc7b7224ab336f15651303f87c3432f123a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "ffe0de24b64d06dccefbaa79133407cb7d96fb72392dacd6c739e9054a61c67c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "b00522b3d1bda94d4bc4b6d91ed30dbaf538051a0effc147d9804eb688197d9e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "3d9aacb4897e606afffa1fcb8964d378518532613d79e8d7a99ea6b59a43a6d3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "44bb4499a2b22a3aecfcfcb03299b2671dabec6d21f1b75893443bd15125460b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "ba0589d24bdde8e5d313b6f772656668f23f96e5d0919d9cb1616ca8480c6ba1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "6f7f49230a331665e63e5ec8808b615008316265f5e080ce124052ab3ba6edeb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "51efb503a6b7f1ac50946572de9db59fe35f7d8f201638893e52938b12a3e48f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "095e31a91bfa545edc83921994b7890db5a62cd30e472719aa8cd1c03923c39b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "6f5bf85b3131b5960bb0c2c203cf41f135625e954cf18110eac3dbe6e3e365f3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "6e7380f691de3e5f66795b7caa9e64a7b2256711bc1f34578aa99d034e8766b7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "ee6c4c573a5ed8ae26bb123b964267f713ea4191860915462c5c6f8e1258e1af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "06a61fab976c8ab75ae173ba9633f0ea6993c6a60dc2a2555deb95332ba8b0b0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep15",
      "rep": 15,
      "run_json_sha256": "934b882fef412ad248e96f945c4612d5707a36519e2a9513cd5d06bcd0cb30b5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep16",
      "rep": 16,
      "run_json_sha256": "f5e961975beb38382c7a55bbb753ef52296349f9c104b412e3b99f3b720b34cd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|16",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep17",
      "rep": 17,
      "run_json_sha256": "4b73e95220a7dbf789906243c2f083ab3d6bf775eaefef72f61b3d893c5c21c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|17",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep18",
      "rep": 18,
      "run_json_sha256": "d17d6cfe19f376c00747ef63d8a121f04bd4bbf755b004203466d91ff0facb43",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|18",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep19",
      "rep": 19,
      "run_json_sha256": "1e63c65ed3304c4c8bfb3946f3464bd64e448fe8b097c4b51e7d8cdf2d05f54d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|19",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos-hard/rep20",
      "rep": 20,
      "run_json_sha256": "418f29d3339057cb97a898c1bdd631452d06b3814d4aa674a1c1a99d0761f91b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|20",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "23d9dc0fc4ff4f5c98a68c26efd4c09d0174e666ab8f74d7faf4230fdb27cd68",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "54d07fd05b450a215ccbf2cf1d68541c5c1e72a732ef1d70d0176374502c69d2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "783438011a430c9b584cee964f725948656254efb46bf4dcedea4e7798e3da84",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "92f17150acc685599210bd7a15859550fdf44e3d4d48dca40861d8083b3150cb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "2b42dac997ce39bbcab12a75381289be517387c28202268272fd3c1a889392a8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "c71cbca896586469f56439ce8fdd9dc3af200c286db53d4b89d1a2de7acef4bf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "fcce356110b8940764ff46f1a3e68ded8ad1dd5c1fc491a6301efb8eedf09da7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "0692183b2eb9d01de4cf0c6fbe81b373362197aa42948ba566e26772c0bc8d86",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "7273e8de754beb8fb7048a9c7b8d6062fc40857a03a8362d2e0c887e2c9067c9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/graphify/claude-opus-4-8/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c4bb1301cc9e3186bd92db0a592ba8b8206ae33ce62fe448d2efe8ccd65bbe3d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "0557390d9896aea312136c29cd1ea3bf0d1fa206bd9be57695df2483afa02f51",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "28bb686ee7923d7390458eec04eee0c5ccdc3f37f7e92f5f7324d86a575626d8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "15bec416bea99af03f53f2722c65937df251b69add849e695b34cf19b6d8afe6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ae944236927c726a13bbf17e4fbf759c6786589e0007d8fb0dd5b08de91c6e55",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "509802789748bac2380fc1acf8389153dd0829f66db8c9d1ee072ac6ba0aeeab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e003692a3e61643699f74bcf7d6f9cf56105745bfb0ac8d88103f0e0c317c2b3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "319d90ad9851c9f5365488c32a29d83690cc6f36f43fa3df839565ea92e31178",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "eb483fe47170844729687f848c08aa156191f920781fb73faa1bdd82d5ac1f67",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "d9f3ee5db437ffd7da646f87d034b7ad85d44ed90bfe3135733136b8a1066aa3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "9e7c49739c5b0f3ebeb39875fbf0a1b10a47f213222306edcaeb27e765b6d99b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "3955d8535ee00d119ed775fcc40330ca9fa3becd56314518e81bc49878d132fb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "151c1d4b1cfcf68dbe62ed4719fd562e94608d22573cad209d8f6a7579e3a0ee",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "848d011b53a7faa3fcbb5fa1f4c1332a866ce30be74e2b1715ba5f26174868e4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "38aa657db22912e1f55ee6868da955432545c897d03cc18c6eb3bdda4bb8b338",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "b738ca9d474af5c8961d355cc77591282517fed328bdb962b2043270e83e96b4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "cfad7a75b91e1338abb268a0b035e92f548cf8017396f2f0a39cba44b3c76fe0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "fb29f14df3d60880efb6f20c1902672ee2f1a906c8b0c7ef454b73a83a13b05c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "52564931349e4f15fa81e680b2d25e8398ff0d63b4512c2cdeb27998c56dbec3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "e292935b4648c62b016b920069b6f19af8cd2b8e3c14a2b24d7b8112961c98a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "8c8ac685d05226fa8e57ad108dd923c22ad31c8c7b86341724036cdaec9d3169",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "ef3439543189fc3627a72a0b3549d4c309c87addd7da6ecda3550bcfdcf37717",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "56c9b690c03638bc16d71dfa09710f8ea161337f1195cce7533f8a90e3c42ddb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "6c1a26e9176c9fb6715793de2ef8314b12f95b9756830d9e8be7aa0a69059f4c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "f748cb09a7fdd2bd12f4511a6e82dd68a75c037845bfbc71213bf90aed6f8686",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "19ee4c42edfb14503bafcadb00d07ac4d58a13aef250b85e1eef61909abb80d2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "7862ff151a752ae21975e80fe88712fce63c62dfca5bd7012d6f416ea6a8a27a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "3e0476e65b4c686d3036a8f3d6efd5925050f10e6af4fa955a71c9b19ff16943",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3c1cbe3329d9efe468af9a20d9f80be911e936aa52df98a0ff65b05e574703c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "dbe698df8e32d5c168f2b054e916eb342bbd1a8a1e385e9f022eee9801a6d587",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "1438c792eaa957480a8e7976149de300b7b5d9617f6d48b86390e5326b1fdb51",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "734f1a1015883606aaa227dbc8b85a3328da74766c8d71000e3b4d3dcce10d79",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "709c2fb146e323d54a8571206b8e24b09578ca179962cf6ad0bd17d87424e79d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "beecd82f3f754d41584a236260a1fb89078fe0575249348562f81a0440da36bf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "d8bfa9aad40d296e873f81295a2baa2cb36a2ee44f0324237b03eb819536e576",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "ac871ee2012a4bb3084c42294e1def6ea1266e8d2b5ae5602627e1ec305a5e07",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "28f9431d84073f3e733211aaf81d34d3aa4a235f211c6bc7d914a94274f3839f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "76872c13a3c1b21829e158f1f024b522b3d9b4862b5497809b425a8fbafb969d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "d15f86dd9e043f83a54f8a8a357a3461210923dd09ea49483660d9f51e14dbba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "e35dfc7881df197e2ec3d789606dc1d12f522e4e7bd9ee9efa7487afecf3945c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "2553ffbb07331beb86710637f7217faeedd319a8cc8850055c767791d0bcfe62",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "6a8beeb18d527f9f117dffaff3b2736c96c970d5754e2af5299cf02e365dd1e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "c02a49abe19e04951101ea24039a03d85312007656be03ad98e0cc8121fea21c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "7b6fcea84b00b2ead33eb01ed3916706dbcc68d759ab1d49eb91e67c5166365c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "56b94fcd90f4c9c724f4c5c30ee6467cd7b7505ec94989724425daf165b96168",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "b561b8fb9b2ef0d94605758e8e630ed7cbff752455215a11e8d5da08b31405d6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "6af178eb501cea194c5ebea5bcc11d877e20056fe28dee49c7d3793f68de625d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "0123b14233e99b9521518ea73452368d133e6df9c91eef2824ea6606d535a168",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "b2ad81ab30bf536c3cf1fe306ab29be5321b24e2929910538719193a7dfc75f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "512579e245c63f3aefe03c1b68a559b7e8abb5284a26d5b417dd78df3108b9a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "945a2bca9e6d7fd3a3954a756d08408ed4832aa4caff6d07a073884f05359f29",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "41a3710b7db0e226bb95d34dba7f2ebe078a897786a035617a9c8f8adfe6e00e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "9b5d6795162dba2df39efe7f46a9768ebbf3ad176affe34044f38b7ca4505190",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d018e27576b351be7532d5f7ec6ad0850948e232c786a4e000a67e887300c016",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "7165a1404c580d971461538c8426739d5a4bfa550a2c82cbf51172aa0ffd2ec9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "9a78a7255d50c0fc49c82c523b631e96151a374c3512d0e8e4c472e4c7428c49",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "0c93a13b715965426058c02f8c070f9d5b16dc0738bb6196ed320e1132b236db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "f84d53ee144c9d3d4a5a37a5b0f5d221a23c19e93e3ff9d3efd983e12741d22d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "822e8be6b5c3af94bcfbd30f30811577b9f8dc71cfa1336027606d1b1832a046",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "ef90920a00c9835ef95c39a7fbddd3db199d87a1ba83986959114da0b5d1407d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/graphify/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "806f6db7909b6069e9d92622850f316ff8779cfc37e886fac05b7028a4fc9308",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "c0f7e78baa867ce32a1c3e931709b83a4e73183959e8d762e9107014cabbe5cd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "2cde1101f08b5077d789dee52866f424a1a35aa7599975a17f19b9c7ae7c078a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "3bfd2a5ecd37a375c4bfd9870614642421cb13066666391452c41bf631043db0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "166c74f34f862812739fd469bd0a618b9746ddcb2da761aade1634909f3f7d2d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "796076c901ea15b4121a9187c8797ba62ef896913102512e3003a2d0015b2203",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "97f823b67754b201bb0af51ff90fb6169c6f79228820a0d9f9bdee10fd341222",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "68e087af9aa50cc9123554c5c262d7253fdedb9a90437b4fdb25f94b0abb8508",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "7f6850e091ab9e1c8ec5fbc45e87a9568ac33dd17d153c0bd201d07c585ad30a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "210ac3e0ab281c87c2d7fa6c0183e955f0d2811f2944933515458d931fa4053b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "17a112403a29bade3efdd8190ea1e865644c9e5ac3112231a09a7cf2d690311a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "b24bfc5ec0d9d28ee3b4f9f231b0c51517d5dd11589102472e27e722c3f5e673",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ec7fe7f58cacf5402ee31675ec1398282824da0d48c7e3ba2c3fd77265fa83de",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "47b374a4ebdf17f633d506a1310f43b8b7fb62a73cda7d27524d2adaff4a503c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "670cef4997f20b49065239184a5039ead25f72f177afb0b1ec31b6f488922d7a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "a32a68f9a675b960a5c6f7f7a7d73212bae0f5e70ed76bc31bfe58a284a327fa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "cd8fa929b3aa618b45ec191996ee52f026ce7bcd79b0c352b150f36f048b1903",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "f7194c4396f12c12486f50ccf00d1289a9bf75d720aad3ebc56ba433d07ca51e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "e28ce5e602a3388593460e1a264a3902ec661f83a9d721dcf3bcf01549612379",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "98363ba2706d80c4fbc6b9926314aa7034e734ff917baec840b872a5593117dd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c3f2de9cfe48f054ece6c67592390754dbffa6700fe762ed7676122d25e710b2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "2a6ddd6b48a552d8e8df28006dde0c1948ba7bea9b38e8f4057eae914781f93b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "7b5c8b54216c71d0e11d01d8d2d2d70a6509f2921a532dc24f56c7e6e1eb44b9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "32e719ce0b365cb320e0f5d855c6703fdccd9e1b216e05eff235d11deaf02d22",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "4f75cfb2c4bb0dd9dad08de05c168091473b800cc9d2016d8923c6bf53779b00",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "d6ad26eac9d3a652d77d2af29bbb32ccb9ecf34f026a2750e34eaa2452f9111f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "a8e323520981f5b6f9632566de74dec1d33ba6c4113d52ea59d0ff51a06b7773",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d2ee67322cffb0c362506ba8073300ef72d6426181aaddb9d1be861aff80b16b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a8d0a66c4bdb5af2fbc856b9d35a15b1f6656e8339f6091e3bd674d06715711b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a87d077cea2fbb8d8f32ab5340462c05f40cce475c079dae2f5efede659d5cdc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "41ba7d1c5b74129ae680329cf72186bef97cbae72114ef47d1094d74b0069e08",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "8b42c7d263439b240e3614817b6da11e46b70a6ccd3b03c9a80624c4279330db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "0b019759eb41cc3cc98c5d599174771562263640447d1847b45e6825fa410555",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "915c47762d5e464d3f87d2146cb5addcbc4bd8ad7624e09e08908eaae729c078",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "ca2153be9db3b707952509c728bd88058f70406945f5e57d871b07e36db58a7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "67ddc90480edaaca672a3c26f73ded8409f0269499ccc4d19224ffa9e705b6e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "4ef08007129fd850c623b27d61a3495a72de34de43de4196a100854ca7b1376c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "b8ac22b3080bbb46b1911afce75503f25d796d9a540140dad7c784f07a10de01",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "243f37a0f51f35d041a3501c0ff4aec37b2854a4bc0e4908d56bea2d14d4eabc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "b63912f6c5dc8655ba2c2c365d62fa5c190d91e035980cb3756bd41167171028",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "graphify",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/graphify/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "31d9adf518e3e02e50141234abead97b713c7e80230dec0fda26fc7d5fcefcda",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|graphify|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "c16252da13d17294a7133344fce0c7d43583d7d9f237c5688c3d8922ed72dbcf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "e2c2767929f72a01bff072844f3522603b5f54393e1eb69776a058367dcf0188",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "f724c0f5a7e22d9e2d97a0adf5bc15d2944ee4b47c06641a3261568d712b32b9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "bddfe89d878379c33ed9bf982a7369c69b249c2a9f30f4eb3c1ca69bb6b878cc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "fddf134033c60af010aad6613d8f156a350c0d2cec3a72d5e469c9c007fed85b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "a63b68a32c6ced52358335a91179b47b5d084880a93da204a6acd1ca102b27fc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "5a71916fa7a2017f5f5e890d2873140435f689cc9d099b53ad5a055f4f25d08b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "e5d77bd1f24ba8046cba9aec5ff1fc2ab37ff05a0277a3606c28366e8561ce83",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "79225a9aaf794dac472ed90690b8e808aa2634073cfe0ce970c7c6ff8d31c2e2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "aa9defd5a3d34ad06157ba5ad70a2772f83df1188e4c3e229509e55b8f508f26",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "95d4db32e487b7b1b53e098a6e9079f8a76291e9266fb2813dad5c957790ece3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "0bcb7f14b2733c16fd84a96ecf0dfb0a7865d8440283cae176255d5baf32579d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "c621170dde643fa55656c0bdb6785af29f6a8293c88ce5494c4f60b449536e1f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "bb5a9de64ad2c485ed092ef99d85ceb65519f07cbdfe27c90642570a072be22a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "d82152f464228b65887274cd128190d090ff60f39c7c231b0bb667da134c3c43",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "2b6c56315f17e4ce219df794a16f04b6e7ba63e9d7ed8298a3c7d84c9ad4d16a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "242d03c2cdb35fb496e5b02014c861f2a9f201d96d1a43f108f240e308c715b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "33ae7842c5e05c2156dd49dab43e44691570b5e6488b27e26ed46f9aced2b630",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "eef61f5a1e2e00b5618034cdb3d4ff13ab1685fea9e95ec45a59222ab9c9b5bb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "29d28b2cdb4547165ced14dfe598d740add9abd8b1a020401e7dc0ece25c1404",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "c9f0713ed66aeb0d9dd75a07350b3467e63cae42a6a722330bf6daedfc9a8a51",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "f934088267198171d7552a1b2d8f8de2984128e7906f3bc839ae12c1dfb28449",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "3eaa68d07947e4079b63779634c3886bbc2223a163351dc0dae4736c4b6cae1c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "c5c5bab9305f6d0a650ce21249c34d712aa1c537904ae83e55f5705cc463c858",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "b4daa78e2e32b3541c43c0efa2e20c200727555aec766f510727098eb889cc10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "3ba52dcee490d94f5480944c0a52d0396315c734e9d0f750424da4a40cb49bf5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "8e040fc79b6167718bd5e360c2faf32acc738884b5b266df4957d302caae6a94",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "050db01af5b8a278410a377e41d8c7f89047ca7b24c0aec2d482febff7200186",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "70454dd8d4699a8a88f2d03074450e806199363919f7d95afb5ae502af1e0b6a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "f70b9ed389f5d01172fcae33e531bd10562bcf5b51daeb82e1190646cf3be34a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep11",
      "rep": 11,
      "run_json_sha256": "85d97de71d871aed2d77fb94551c909c8fab682f34725deb0b4591d6e5d3efbe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep12",
      "rep": 12,
      "run_json_sha256": "14714830edc578610e1946a77977fcd4f6cddcb2e6571f7bdc1377f09947a814",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep13",
      "rep": 13,
      "run_json_sha256": "230d81bfe44e1cc41dc7718a320339d8cd0de353b4a59b9c0856dcd1b6679b63",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep14",
      "rep": 14,
      "run_json_sha256": "14604d4428d478d461a8d3d9722b0134ff707296970e98435432d736351d8c6c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-fable-5",
      "path": "runs/serena/claude-fable-5/ZeroShotExam/memos/rep15",
      "rep": 15,
      "run_json_sha256": "cb37dc511a615404cf5f92b97f02072bc763d2411b69fd079b582bff511f3e7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-fable-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "c4d81bb341b481649a680fa1305043fceee76153ad2d30cef31b5d9cacf964e8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "bbaa4e5bed4b8645524375f5f3764bcef346e6d11551e967dd9a5c5bbacb7e1b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d9f7984f0d8f27def103e01abf18e84cd373c8b25cadadf26b95e49299d2d5db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "64beb2fae6fd393e9e0f383283c4c65f6970bca73507cdaa7002821c70e46cdf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ffdacd9b599e92cbeda73f6c20addec2c5a2d511a63c085ebade6f17c33a0f4b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "4254e60d7b1e1b8cbe45b1cb3266949818efe857569e379ae32fc103c77c5297",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "c1e9bd1e2156070d51ec5c0f7295c57f072fcb2b144e1c3daeb03ea4179cb868",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "13f1d8983fe96505f85cf9bf3a8086ddf4cd8f5f9d2788e25b09cfba8dad4f69",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "0fd97eab9ac0073f8120ced861d68455d0041b666f25ed2b175bee5c48940579",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "991d68905e0d9bc01c9f6148977cdb8250dcb92ff2e9375196bdb912314186b5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "7e74706f9d4b1732ab3bff47bab9ad421929bc693f67904dfe86dfdff2538972",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "8b7c7da3d1d225d465efc0fbaad7ac0f35eab9384215dbf93476a72c47fefbb4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "ce18b93b0b1355e461baf6e0b4e03efb61ffcb856881cc4e43616e81606ccd3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "037630dd5f6898970483095e76c6ab3fd73030240ea54a1b93e5c7e02ae44349",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "1b3a8d12640473747a7a90eee51cb267a51cde36b175111fb68926f4cedcd042",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "0a63601ec5de9c65f066baddec29206ccd5d7242a3dba05d5b2683ea90777827",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "bc40fc5a88295ccc46d51e4217fd5cd9e70214cd60507c73331d94c70cd3d307",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "ead43ba7ae43516cb6cee4ff9a641cb03f9c1a79d59cf6828a498664d4a781eb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "bf66178d4c3cc04e3ed302dfbb08e607472fb62dbfc92710b6f9ad89ea2d8ed3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "6d36965dc1f7f9e9478768b0804d2f8afe8160391a6586e111658e4d662e7ded",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "b0e2b06e90bd25ee9a59b35e5c614fd8cd34637cf40b3016dc822a71570f0366",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "63387b9551a156818926e7c6bf3482ff127bca2a99b5a53f763fdf096ca995dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "cacc410ce1e356841916b360b726a6b644530d2b5eb7623a928bd0c0b13f20eb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "494b2fb1e9c9369698f272bc2121ce07816c6941dd97126fa45d0b7d247132de",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "dea34d2600b5e686a63cfbcae30ede2cb066ea3ddcf873522a0a60cf9437e7ab",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "9853d6bf3efb016d01b6bd7ae7544ebfc93ee0f64d87bf8b9eb204f7ad20303d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "6b4222d807d57362af466a323f6ec780a6d962cca83b87b828d28869c87aec5b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "789fde0de5734b1ed87b0b9a2de472fa6b03cca241981bff689ac97e7438123a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "a437600fe23667e09a786442d248569c07dee3171c8b0546a5e906fd35e0d6a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "28f90b87928b89e46fc08dc5c1af413f65f5b9fe9dd290a20ebe2b04fd89ce98",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "7cdb212a56c8c99dc8e609e1d005c595221cd438ca4c0336515d2757340d0bd5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e92d84ac6ce31c1929b0dc6f30fb06a00f02d5827a9ba8e533397f31eea55d1e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "96e2c60a1c87fc3fd9f00b8895f18c1558968ca3619d6206787bbedf1025fdf5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "86caa403841c6bdc71b8e2d926d3392e3a0d8ad7e2c145394c9124b5d946db6d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "100f7ff9459c71f490ce5a6d0938529b2e10ab2a68041a219703e9b44d549233",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "4e3a415c3621a27cd695b419f948f859d72b3ae155df6b8734be2423c3468fdf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "cdb3e4f06e2e0cb70a0cd2dcf4f3646a7da545c856fbf26e4b7b5833b88400db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-haiku-4-5",
      "path": "runs/serena/claude-haiku-4-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "274a9fa5b7186401215640d2c4b9b00dfd3d99619814237ebd7c4cd8d46b9508",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-haiku-4-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "e85b3f80d6cd4462d3e6cc0b3ba3e33bb67b389a902fd9fb480ecef5d961ace3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "8dae09da535e72af87988148c2d63669040ce47a7873dc421f94d98dbdd39100",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "0c937007cecfe0a0d39dede208eef89f957ba3fc93d4e4bc84c0d996592a86c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "cb78730928ed546c8b3bcd306d59eaad218330c4c0d824f4ddac9cf3fad7437b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "af5f2c7e2c10be2c498d8c9a45de1a6b1b3b9b61f9d31273182c3d46cb37a0c0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "cd221a0d31312e12c0ae3a92624d6e5bcbcb1321d0be154aec79825dd95f1f61",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "cb5eca60e471975d5585a7a5c502aeaea7eabf701eb4cf755f3ffb90e4fbbb87",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "a53f735a3c0774d52f5e712fd32237ecbe25753903e7d1dffc1dff7b7e5c2fdc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "e740e9b02b9f8ebe067821a0812cb8ae0cd124e05ce71637b21377a386d111c0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "02777607bf654a0292e6d3a07bc00546cf1e1cc9f5f8f015f590ddb4e7491c7e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "8ae6d8ede55b05524d37ffabc4d997b27d6d81096856478e90e8e4ccd8f5663d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "3215ba7f258611347f8ea26a7aedbf9b5edbbbbff3cb3a822137a7cce8e2e1ea",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "e2f48a58d38665fd9179dd0e6e1d91a7c8ff6544cdc98b3193c3b5e77eb82e81",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "e7385ceb27bab28a87fb1a31a656c7dfab68ae070a1db0a889b31599b36fd2c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "6e128ec1e892334fa8f3523e712c5e126db7377f455d37338ba6b6058ed1625c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "7223e8e14351632b6d50d8236e5101d96bcbc67468fb578cb2d4e1ac060fde47",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "4612f05359523848c350685b29124cb552c88aae413d238ba285332113556047",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "e9a56cc152abfee0957e5681907ed4317ab04d7bd6727074d5fd99c92081f2c1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d2306fd91d6ec2586dbaa26bfbbaaa51047cc1f62a070c8cd57a2645c804d5a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "a5aa99fc067b70d8c7d5b605cd141ac91e47fff70ba32d5132f72585517ded02",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "7fa66384bdd814f8a09a3f3119782f7197057dd41a2ca7371cd9fa37bd3fb03c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "c49200392adabfdd374638631d7d317c02ea4373166e57780d2d71d288c6c6a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "5aa7393e8a7dd1842770b90874296cfdeb8f0c4eac94b10b03e3e37e1f70bc47",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d182e51d8e3c2f69e1d07519adcc8e14cc3debb87004cb862f7fc75f436206bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "e689473947e94cae123a175095b0d483fe21fb74b944f9c5b08edcb793983192",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "f49ca3bd2d987e29de28b68da176bc604cfad0b44ccbbad40421d12d29a71d0f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "2746b097031670c1a3d6b22524266f02a0bb23921aca3d341327d8250158dca5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "1e80bc7a1792e37182852ee0617b0e08f0b095f84c075f015ebaf994c4802c88",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "d3259c2d3d46bc73002a93a3568cefcd1a06ab5d33008a089adce9fba9b1336d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "d22b19e7769da2d541339759237f9010d5117936a873149cf0bf1eb1c75846dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "236ee763df8079cf7253b02bc24834c409696b1ce2b1409281da1230fb0252a7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "43cc85ff34e9e059be23b7ed9369773e74cb385c9d7f5c79624a49bdb9db2bbd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "96612fd5ef08cf95f5185bf8b7e41fcb69851839f41ab0d01b4071613cfd3e4d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "11c7420d4d29233520900d3771242e051ece484833b3f34f177e6790435e85e9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "65489567fcce579a647dfa16e4c02448e2823338f4975d9377eec569361693da",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "920b7a3cb01c303f686305806858b071cfcb547700671552a78b74b4156fbb73",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "c94b00fc7f65d909ad0a885c68ae3c5d87b91e13dc1ffacaa573cc7c893196b4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "cb7a0598c94e0a23169b62ae363d152b31dc932654451d099557d135777ef182",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "f7849d51ad7a29d8a3ce1118f44165a265e653e7dac84af41d53c320e36caaf2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "776482443f07116aba3b43ba28b009a2596f4690fe72893c2fe1bb85d2a6ae69",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "7bc9148659a42073e339018f4c9b3e861c34d7933371479994693d4179adc6b1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "658a54e1680a1a9eb9e2a58cf2ef73174b3c4cdef3d0bb142a42f740abf608cd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "11ddfe8c196ed211ea3929e81e78d7b3dcb155a9ae5937c6e208397d113dd8cb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "d899b89f015edc14534536375ea6fcda4de61e9f8de90b3c529ec4a25c6c6d81",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "c9c1a3755b9be08210d132e7b40e4ac145dbc42761ef93c98db169504cffbdb8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "37104b4bc8f77ae085e91a928758846a01bf739f56206684b225b0b02b265e6d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "fea3c7130a8c8cbae5bc3dfdc8b282f14a04bace7da6ffe10d81c08a6c47252e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "2ae1f9a5e032993a958b36ea661f71a49946eb4d3daf4596c07079bac6833b54",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "f959579b03fb7ff441c710f523e2b2662708b03c08e439ae00892553b2b3969c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "df5e1c6b678ba7a39eaee838c1f46636f6b0deb23086d610a9d384e084a6b099",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "89df37d4370eb9f90cd427e0d5f212633d471e5d3d896f782ccab22ab57cacba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "a5a143748ad459747df92323d2cedb82e9d25d6a29008f8e87000616cb406e42",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "97cd3baf41937846f7eb01ac3a70bc26aca568113a996f1c249291290a115e9e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "7d9a263022416098dc4d37afcd812e4b75e97d588c1a10da2468911afb428b91",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "3a50c50f77ad37aec5ae3aef428c7be03e90fafacac2320c6b465d85f8c9ae57",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "5f701cc91ca6935c4aee85533f2c08aaec9023d9f40f2cfc678aecf6f8c55823",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "8f9a5f5129e89001ca658698351094e9e95e6fbe056a3a908e5b8a74f66b3964",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "a1c7212dd3618a59299ac6da40767fd919577eddf9f4fd230542fc1c66b1b93d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "753e2f32cb60ae5b03d81a416c74dfeb58d1a30438d2d1769b76e5e3a9d1732a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-6",
      "path": "runs/serena/claude-opus-4-6/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "c3ada29696ce1b259a48a7e451fd4ab124b9d89b7171ee2bf8bd71c945842c21",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-6|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "51790e40f497fedda3eeb2c8489ada8d3ac16059aa4542dc0010c89476945ce1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "1308e5276fc9c0705fecc5727f1b0c60e456e61004981f5fa7f8367a3857a38b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "c4ef8b7810524d8715865155e4e80a4e83167dfa3044953f45257921a4870e9f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "2c01d5c29c09a335dd05d50e8e16b34112251e20ab056c23e0d4a3a872b36ca2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9f84552ad9b1020570e4d19ec7f2bb13967c08070952dc932145ed1497774975",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "035b91fb1040a34dfaa8aea98ce363295baa741c65b593e30a227423749d8f30",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "ff590051716d80ca84ddd58399198da6146f4b82e4717357b0c4118c587eb056",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "670aa4157c7bee18a38c2bdfbf07c9aa7389de121cbc5e30514f8341682df561",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "dd3799782c3be00ecadb3bac4def35113e2abfbff72d338337ebaba068681a5c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "8e3245822609e0fe4c7c0e1556f3ec159d9f4087a46f0284a8e61df8579c2afd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "1bdd4cd3f0e285f19eeda9bc63bc7866082530fb124d72295bd28785eed64115",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "601f09ad50745da8bf134117a69f831a89a22efc4b7f426c5d4a83aca7116dd0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "ced1d8fd7f4f2b4dbd8060550fb7f2429cc3c4804d3b57e049dec312ead2ac26",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "ccdaa9606dc29351eed456ad9dc878581c8e7ab5688f11ee934b42faac4dbc22",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "1d0af60aa682eccdbcca0e6f18f9f267eb4c2c05d43cda637b05888f04a5b942",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "a354f4d7e3418c6eac917be28806c665d0d5d6d3d6e39462a326f7a2fbb7ea89",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "01a01f6024d9071494184cfee0a818637dd74f176abb5d65e974ed562c0e6d2a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "bab7135979a5d29ed9de1a273dd682fdf73b96650f95bf0d0e3b2dc022bd4d7f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "d72a82da5d7fd408d0997d5f6ce59083ee54b2bfe0c96f0a103be1b3bac5061d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "97ee805c265c93427b95e58913ee4bfe8af59c2642ec1438a06585ce168f4c69",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "aad9bcb91a5da2a16fc7652773d5bcf2f5adefaf1e50aac7be75208fddcc5b6e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "94aab77d10bca1f21e8a10b8f132c1bcf3ca08515b042397ad12d9b74220c6db",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "a738567b1b0709bc33176508c0e48f2fa3324412871fdd1a57face0ade77e76b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "2b1c43e8a24ae4fcfcd1a03b3a1748b2d7e38cb5845f22cc96137e69fa4b644b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9977bfc8343ec6b80e0f584ceba4791e7b284f726276a8f2fdfdf0dd16bdf111",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "ef344e79985ab8393df1dcfb2dbe50648b6fa6736052b0ae7ea5431c52378b3b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "7eb415b74d7185db8456e0d8c7410f749161d54cde2cd6b57738c4a736d15bd0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "4643edfaaa73a694ef80a0d7c2f36a532d8bcfce945e3b954010fa7f9ed44847",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "2a52cea33d4533b97b6af611be60df725192457497434350d99e33c8f221beff",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "f57eabdec0b26979cbad16ffb4b1be4abf646c04ae4d0719dc81c55792aaed8f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep11",
      "rep": 11,
      "run_json_sha256": "a78b1b1397f8b9aba4fb42437690968ef5ac003397a16070d87c05c69ec4848f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|11",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep12",
      "rep": 12,
      "run_json_sha256": "a0122a73fdc1c08ddc9e74bd5277f15a3bc1751c50cd765da3fcf8e7a5de1996",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|12",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep13",
      "rep": 13,
      "run_json_sha256": "c19aa420dd343a23eb89453ece77b8f0e62dd4365b293b55f0f0cbe256c1c2c3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|13",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep14",
      "rep": 14,
      "run_json_sha256": "9544300d73d0d679242f733cfbfd948841b2a81ac7fa11a788a338a3e9812f4e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|14",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep15",
      "rep": 15,
      "run_json_sha256": "bd73b013d77f7b81b954c9f37004d8a13c822b3fd26f5cd0d5687ec6683c80f2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|15",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep16",
      "rep": 16,
      "run_json_sha256": "7b33df2209e60055aa9e8f8ef3b782957a9d9296ab0e80dac54fd5f59171fa20",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|16",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep17",
      "rep": 17,
      "run_json_sha256": "8f1842cd002ebbf81fe17fbc744b125f61a4a862935bdb91ce21b175db1e648f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|17",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep18",
      "rep": 18,
      "run_json_sha256": "0cb5136d09da1771f10a2dfd91a62d6be2204a1c2e2949ec2bd6050d91727ede",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|18",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep19",
      "rep": 19,
      "run_json_sha256": "3091a17b9ecc0a2b387ea00afcb2f7bf6b0caafbf7a2027a47fffc0777d09e2f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|19",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos-hard/rep20",
      "rep": 20,
      "run_json_sha256": "bbfc481fb92400fbd7f394e33013208253a2443a95e3e3c053d9888a8cbdc284",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|ffc82068051276c8ecade67885ff45af|20",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "c4ff44c595200cc6d9ed8832d274ce25f69caf6923e46fa420707f48709fad39",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b0899896b32e3730ff620d235804e0fc3b4abe1c01d02a506984932d372837f5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "c7a25626987c666c02ec1086dcacdbe86e815747dc1e8765deca679648887bfa",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "de5a08f674665129d1699698a3e6e25e427bb62e03fee5ed9cf1e6159c44a447",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "7a4dd16b11493a736de7e7a86774681f8380a5f0476b60903ebadcf29ced4f6d",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "bab41a6835b02a32105a9a15bf3a733c60a833ecab71b8abb082a35f64e5374f",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "af1795b484ee19eda2ce775b6ac9a28fa31f2488efaef503cbb8594f3911e6e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "3f15691ac89b4e7c7548f004c724674eb11db445dcabfacea465e31698c1b6c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "ac2f4e182435c85dd5772260b2d2450c19965e70a834e3ef7a353109266246a1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-4-8",
      "path": "runs/serena/claude-opus-4-8/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "0b3699efb30a8d340a161d056ae9cfa1b5ff915159aa96b254495a4daa20e08c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-4-8|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": false
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "4b3e687822f1d365d5166e5bd2c0a531d3568cc2cf27c5d794760492f1976fe8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "25fc9975a33fdf446d89339ae872d599b572b483375d834689f4aa519b1e1360",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "63c0595144edf9ce5c92b51306a7ab0cb2c65c94089581e6faa3afe36ab70762",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "fb142b6ead4a0226599c27eea3844a6e5b70a2dfffd54a3c4a3f03755591b135",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "ad012e34e289672b689cccb479869442827b8e07b05d4dac6c1c9c19752016f7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "e3af543ab639d9880332f30539f2857201e2d08e16e76857987844f874ef4fb6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "a874aefa08f3cffd66f56780e19df911cec7a2e7a4ca067082e847dfefb907b6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "3b6b1b331855104b9f27cb679193ffd851e8569738bd12c64ccbd652e0cd549a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "e673a1813fceb4dd148f95c02731c0e8e9a04b8a4801f258db89e01b942b6809",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "b5cd60a255db4f841ec51dd836eb5566c35c2cbe00bdbf9d535aa4c17edff879",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "176e0c8d560c1acb375d1cc9bcfb454130bc61171b8a73567f29f1285718e8f8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "ef88ee73955250a8c96bdcb4012d55dd1e98cb99acf823fd34225feb135e6dc6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "74405dfa4fac4343db25cdbcc8c5212ebbf2902c8cac6d88409c83d2f035a623",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "c8d33a412984f998c5f40fc612a5546f48bcd5c80c23c99036a1446cb1bb1a73",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "a2039cfa318d2b6bccaddc5d225122048634b7bebb702232deaadc935c134d66",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "21c53eb21d66bc889ddcaa21e44598e39e8130678aa604050ecbd4ad66f9f9bb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "30cd07542664a5f60c5748c4b0d1c4d0a8558ef8ad50e00f07dbb056ef3436d2",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "3c9c1147118c4d9dd4998b6cd1faf81214a638505f58b7f60b7af58ad23e4379",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "c623effb23e07d2b8071bf04f733a51e31feeeb0dab559b5002d7e73a21dd118",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "HumanExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/HumanExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "b1a4b9170ffd53446afc4da5d3ac69b061a62d1ca40443ad142643fe598b4252",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|HumanExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "9a8dbababc06e7c2736cb93f0b19fab516524d82a8000bc14822f49974d21292",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "f6170fb8b26fef673cf353e95720ef47b8fceb6f578a86dfe4bc85d7180cd2e4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "956490808aa37c5c94f545962523929656d386260ea8ea416fb54b9310925221",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "c94d9d2bb0e49a4b57a7d008102f7855a10b5aeccd241a4275cf706ccc5544c7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "4c52dcbad831e4378b737148717600f9470053e147c70ca37ce98b795647d0e5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "41ffe55f0a1f7a197606cb89581a1015fd55af3a400d7957f542e6b86946f8a6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "aec0cde0b41ade2a0d73403396c8c8ec5931ed394c474a59602218b776b740b5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "910e789d41668ebf70aac3d6bf5493464255b08f63023dab887b02f4e35d8075",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "250ee666b7d18d9fcfe83410c1ad37b71e5765b8ae7a00d033dde29ffbf7f8df",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "087604e243495bcd990b7bd3349737839dbd7315575bfdcbaf0e5af3dd673af7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "4bf44bbdb5b3f0f9d8e6be78a5f002a1a2a8798996a30969f05600b17a9c5281",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "9227d1fc3a12d3b69aa9e83b1e51865450c65856c89f0cd43498b57cd9ab9309",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "d92feedc1a5bc34ddba14355b50b489217cb7f65ec733fc6b53f6ba195c42f5c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "fedc966fe3376d6c5bf6592726921b0cb35b65c0a72018d5f2dfaa83f4190bbf",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "54af87f66d1eb6c3dd0ee61cdea9ae7b623cd9f57fb6722f6f57fbf7ce739b4c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "26e4c3bda6828e7e2582f3ebf14334782d35a75dbfcaf838e78a55034c177895",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "5ca2df9e5f21217eca8c816e491c3fcef7b0ddf74f7c3fd3459b747587bcb9f0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "6d1e09d69d5a0f9427cbdbed2ed15a3231d4a33c947bf4ad276870a73a65f4b3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "5e1a4754a52d4ebaf12c4281c9300863c2989eb63a63ebe644e44126239849ba",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "95458ffb5d5447063d7bc2d4abcff8eb7ebc4ca7a0431b0f27104af73f2522a4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "96efb82a753d47d7f0fe33b1c5843d5b7b55ed2ceee898c30eb0c93b136712d8",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "bfe83188cdbd43dcebff7001685596f8f6dae751e15953845c74a252d0bbebfe",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "8a38a7a2f21d913e155a590d05e83a731a38bc8cd1e0e9fa59b10acbf6b97019",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "5e607d574de6758aa95d5668f31d5d89bfbc7daa154f6e9430efbc3a07dcf370",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "7ec38f2555e95ced4cb7d5bd78d3b7fee5af9ee29502a4ebcd484afc77080a10",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "7bd90ce5cdc2897f2591284f4a6a9d67fcfce0bb55be23175e5bb3c754494b27",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "d9a3190b23a2ee3e64507ba3f8bc6947d39ac80ed45877b3112da485c7c14064",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "856b062761415a799fffca68fae9357d01d8f53312af69d823bd09806bd22478",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "882ea1f5d05749bf73279d48c20497b02cd370a43afcb17a069049a89c419f51",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "fbc93ddba5754b7c5b47966ce18c2ad9d53ad75a1a57311c6440d2f855e6db7a",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "a73a2b6de03f44cc012746daba34f7e970d35d48676ab0037f0adb2e32fc5696",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "b9ebb2fc5466366a84c499573dcaa2d876a72b962df5c24b5cf49935bed3f224",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "a3d110d7b61e87c73041c51e7d7acd532d9f40362fd986df1bbca05b85d15ae6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "df8e639d963eadb2193cdb416e07ff6d693c9adef591d040319f520b8038b7e6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "6f775b1ba3348de396f66eb3e5e4e32c37822313a82647a020d088c2bd4d4ca6",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "bdb84a54025ede4d919ccd77608b616fec2362a32bfc09504271e07f75a2fb75",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "ab82a1f02c54653aec58efa83d846e69abb9c4cb6cc06b5f012ca255b6010731",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "706ba02ce8595ef62e8116c51a6911d6e91f94b54c1168c514510a0b8b4e07eb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "3050a60713a731ff30cc4844c18e90c319f502e05dc5d65c7c89c2f751d56692",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-opus-5",
      "path": "runs/serena/claude-opus-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "d1c9eb03a1cdada832200bd707fbf4604ca94d42d92639a21159c2649412e133",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-opus-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "882d19fc65719a850866ae8c2aa80dbaffac678afe8f37082bb6d5afc514c60e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "69cdb588a0dcef1f7a52ae8c6c83feae5c3f9c9d8ce209a545e311c1e2b5e737",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "33f03bbb6d51e86dcb03ac24e659ad1e47c27b6057b00dec1d1c61254008e3dc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "71b2b1f4879527480d3fc3e91eb0fbc7529cdb45b3578f7d2b88e9b54819e80c",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "9a38e8b214e9899482d77d954a60e06af5320e7730fb47b843c447e4af3f40a3",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "7bfe7022255661c5a3179fd7e6c0a0bd28035e5f16411cfa036a9a9ee0cfb774",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "c874d63042fcf8e82ee6f91d0f7f576fefbcbc9430c39f8eb6ee41c7cafd85b9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "ac0501208a6846b017f827e463a1cd908143663e1caeea6f209f6641b43dd4bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "a37493e61e9f800a8b3c46c34c1a7cac46fe712fdbab1ea9b56dcaeb6066f8af",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "d4b643407ef6bce452afa75df2d47c26c46338df1ec55783c067921c4f3ce937",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "939e84a58a7c9d692e786d3e09d4d878ef4dcb53ea16342e0fc38598d47083a9",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "d3e458aee3782d68683919b3726b8d93737c44f5dccfa749d4104959c626d374",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "8a6b813703d2cd786ef2ffcaff091b3368782a0aa7de4ceea4b1eb464107b4ec",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "f3096a45e45bb7b4bdf7d477a9e3b716566c9a13b5ca2be356af4319ddf458d4",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "60a072c65906fd03829286d39235b10cbd51c97585b3373b7b914a80979aa040",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "d2569c1418c8cc29b63100edb74c3f1d695f0e1dc51df485a605711da2e8762b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "7a4ff60a9222117ac9d3c5ec6b3d1607a8fda5cff729053d974e84788e3f314b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "71da19308173424a15da5c2ea57346b5028fddd0ccd094060e801888e16b29fd",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "a273247de6fcc98753576e2062465fd8d4eadd9e9924fd509cbe6691253479d7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "MultiTurnExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/MultiTurnExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "2ef5eca3d834044f337b1e37fcf325a552ba20fb0bda91c8a8a0ff0eccccb581",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|MultiTurnExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep01",
      "rep": 1,
      "run_json_sha256": "ab6ae067b18eef96329808dde02a9d17b0f42c4f0dab3820ce0e4665f190a881",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep02",
      "rep": 2,
      "run_json_sha256": "180fe64e8163bd97f6567b797e347ef1882da72bd409518df1db58d144d64270",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep03",
      "rep": 3,
      "run_json_sha256": "cb9bc53efbfff8320ec72f236c900eca96dad01a803d198d0604a25b3cc7efbb",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep04",
      "rep": 4,
      "run_json_sha256": "ddb201895c1ef303c0bc7b43e5643ed00132ebd04629f325adb42ce82440484e",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep05",
      "rep": 5,
      "run_json_sha256": "b80be2f8567faa52b9bd3560b5015a4b1d8fc2dda7b18024623dfc04be54e9c5",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep06",
      "rep": 6,
      "run_json_sha256": "5bd5f8cf56ccaffa16a37e7379b16d027cc4e7653c8602548f3995fd2dc80018",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep07",
      "rep": 7,
      "run_json_sha256": "9f51daa9bad0aa55f94a8f5ca5175f6348c13939ac48cc8c531ab97ad1924bce",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep08",
      "rep": 8,
      "run_json_sha256": "30e24b8b4ed4319bd82cc0fe7dc388ba641ec79c6551a0f0e682084ba8e1c492",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep09",
      "rep": 9,
      "run_json_sha256": "24e0b0ee41e99321d124589dbd8d9f6976d40347f551ffdb349e659c8a161930",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos-hard",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos-hard/rep10",
      "rep": 10,
      "run_json_sha256": "612dcf959261f56eaaeeb82064143b4da308307d8bcbf1ab531aa35efa36047b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|ffc82068051276c8ecade67885ff45af|10",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep01",
      "rep": 1,
      "run_json_sha256": "8c71ba3b5ccfaf0e7932ad5f270afe754a0f2e6ad2c5d38972811f353e22bdee",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|1",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep02",
      "rep": 2,
      "run_json_sha256": "7a4dde7222351d3abbc51ec3021e427b0103924d9b505a1a9448c036bab489bc",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|2",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep03",
      "rep": 3,
      "run_json_sha256": "efd75139b9c57e41d359db7df3572506fd02a0b33fc1db02df448375eb72ae4b",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|3",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep04",
      "rep": 4,
      "run_json_sha256": "575702bcdd08384725340437e84c48339b04d0c4291b4437db02182cc171ade1",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|4",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep05",
      "rep": 5,
      "run_json_sha256": "802b2072cfc0a09d6defdcaed648508712ed2e7f026e6f629fd70acf944463d0",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|5",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep06",
      "rep": 6,
      "run_json_sha256": "b4ce65ce998da89f6ea14aa46e57b7c1820c6ef3baee275d37e4ce61f3e0f513",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|6",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep07",
      "rep": 7,
      "run_json_sha256": "2dbbffb812b2c253770b3e289e74196c7a07383ff351ef5a284135d071d6c0da",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|7",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep08",
      "rep": 8,
      "run_json_sha256": "fc48808da4127ce983b9809bbb191d30d9094a5a6f3f6ca5d20aa46c8f838e29",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|8",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep09",
      "rep": 9,
      "run_json_sha256": "b14044eb83ff276cc9d56477dbe7e708a68bd86f5ae19e88fc692aa9f3198587",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|9",
      "side": "comparison",
      "validity_ok": true
    },
    {
      "arm": "serena",
      "battery": "memos",
      "exam": "ZeroShotExam",
      "model": "claude-sonnet-5",
      "path": "runs/serena/claude-sonnet-5/ZeroShotExam/memos/rep10",
      "rep": 10,
      "run_json_sha256": "7dd1fcb6f097e0b52b75699af23c68dce0ca72bc23ed550d53400a64e6ab1ce7",
      "run_key": "ef773dfc921e2e957e945faf5ffc8679|serena|claude-sonnet-5|ZeroShotExam|33ee333a178d3d40bd6a3f99f45d4162|10",
      "side": "comparison",
      "validity_ok": true
    }
  ],
  "schema_version": "public-manifest/1",
  "scope": {
    "comparison_predicate": "family IN (cc_bare, cc_competitor, harness_bare) AND no founding-import row AND not #pilot AND exam×battery pair present on the Corograph side",
    "corograph_predicate": "exam_epoch = N AND concept_arm = 'cg-ask'",
    "run_selection": "a declared list of 2692 run keys supplied at export time; store rows the predicates match but the list omits are held, not published",
    "runs": 2692
  },
  "scorer_eras": [
    "tabench-1.1"
  ],
  "store": {
    "db": "bench.Mini-M4-48GB.db",
    "hostname": "Mini-M4-48GB",
    "migration_head": "0012_harness_bare_family.sql"
  }
}
