{
  "experiment_date": "2026-11-04",
  "cohort": "nine previously exposed TRAIN development cases; no held-out accuracy claim",
  "model": "Qwen/Qwen3.5-9B",
  "model_revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
  "cumulative_checkpoints": [
    {
      "name": "base",
      "completed": 9,
      "planned": 9,
      "strict_grade_matches": 6
    },
    {
      "name": "c0-source-a4",
      "completed": 9,
      "planned": 9,
      "strict_grade_matches": 3
    },
    {
      "name": "c1-source-damped",
      "completed": 2,
      "planned": 9,
      "strict_grade_matches": 0
    },
    {
      "name": "c2-source-selective1024",
      "completed": 9,
      "planned": 9,
      "strict_grade_matches": 5
    },
    {
      "name": "c3-source-selective256",
      "completed": 1,
      "planned": 9,
      "strict_grade_matches": 0
    },
    {
      "name": "c4-source-selective512",
      "completed": 1,
      "planned": 9,
      "strict_grade_matches": 0
    },
    {
      "name": "c5-source-selective768",
      "completed": 1,
      "planned": 9,
      "strict_grade_matches": 0
    },
    {
      "name": "c6-source-delta090",
      "completed": 1,
      "planned": 9,
      "strict_grade_matches": 0
    },
    {
      "name": "c7-loop-rollback16",
      "completed": 9,
      "planned": 9,
      "strict_grade_matches": 5
    },
    {
      "name": "c8-loop-random-control",
      "completed": 1,
      "planned": 9,
      "strict_grade_matches": 0
    }
  ],
  "anonymized_case_results": {
    "headers": [
      "Case",
      "Critic",
      "Label",
      "Original",
      "c7"
    ],
    "rows": [
      [
        "Paper-trading explanation",
        "Whole run",
        "GOOD",
        "GOOD",
        "GOOD"
      ],
      [
        "Drawdown versus volatility",
        "Whole run",
        "BAD",
        "REPAIR",
        "REPAIR"
      ],
      [
        "QQQ condition omitted",
        "Whole run",
        "REPAIR",
        "REPAIR",
        "GOOD"
      ],
      [
        "Duplicate portfolio creation",
        "Executor",
        "REPAIR",
        "GOOD",
        "GOOD"
      ],
      [
        "Return calculation from dates and P/E",
        "Executor",
        "BAD",
        "BAD",
        "Invalid JSON"
      ],
      [
        "Authorized research",
        "Executor",
        "GOOD",
        "GOOD",
        "GOOD"
      ],
      [
        "Financial screening plan",
        "Plan",
        "GOOD",
        "BAD",
        "GOOD"
      ],
      [
        "Bollinger/SMA plan",
        "Plan",
        "REPAIR",
        "REPAIR",
        "REPAIR"
      ],
      [
        "SSO plan",
        "Plan",
        "BAD",
        "BAD",
        "BAD"
      ]
    ]
  },
  "public_four_problem_control": {
    "cases": 4,
    "base_correct": 1,
    "edited_correct": 3,
    "rescues": 2,
    "regressions": 0,
    "paired_results": [
      {
        "id": "gsm8k_train_6071",
        "gold": "1200",
        "base_correct": false,
        "edited_correct": true,
        "base_tokens": 32768,
        "edited_tokens": 1335,
        "base_finish": "length",
        "edited_finish": "stop",
        "base_prediction": null,
        "edited_prediction": "1200"
      },
      {
        "id": "gsm8k_train_5179",
        "gold": "1360",
        "base_correct": false,
        "edited_correct": true,
        "base_tokens": 32768,
        "edited_tokens": 5047,
        "base_finish": "length",
        "edited_finish": "stop",
        "base_prediction": null,
        "edited_prediction": "1360"
      },
      {
        "id": "math_train_algebra_1730",
        "gold": "8",
        "base_correct": false,
        "edited_correct": false,
        "base_tokens": 32768,
        "edited_tokens": 32768,
        "base_finish": "length",
        "edited_finish": "length",
        "base_prediction": null,
        "edited_prediction": null
      },
      {
        "id": "math_train_geometry_186",
        "gold": "65",
        "base_correct": true,
        "edited_correct": true,
        "base_tokens": 8954,
        "edited_tokens": 12915,
        "base_finish": "stop",
        "edited_finish": "stop",
        "base_prediction": "65",
        "edited_prediction": "65"
      }
    ],
    "purpose": "four-case public practice control; not full replication or grader accuracy"
  },
  "reserved_validation_and_test_used": false,
  "paid_model_api_and_gpu_rental_usd": 0,
  "loop_case": {
    "parent_seconds": 2216,
    "parent_output_token_limit": 32768,
    "child_seconds": 151,
    "child_output_tokens": 961,
    "child_intended_grade": "GOOD",
    "label": "BAD",
    "child_format": "duplicate JSON keys"
  },
  "public_reproduction_archive_sha256": "74b48453e8f77db7cdb68c2605d45a844ace8db0ea1455dafca0769bfc708dd0"
}