{
  "schema_version": 2,
  "method": "retire-opd",
  "dataset": "deterministic mechanism mini-suite",
  "seeds": [
    42,
    43,
    44
  ],
  "runs": [
    {
      "seed": 42,
      "accuracy": 0.21875,
      "mean_reward": 0.40066753689370793,
      "entropy": 1.7912896522968134,
      "kl_from_reference": 0.0004698169252424484,
      "retired_fraction": 0.0,
      "student_teacher_ratio": 0.14330037615578525,
      "discrepancy_improvement": -0.0005878137044841858,
      "loss": 0.03957718756807123,
      "policy_entropy": 1.7912994827995612
    },
    {
      "seed": 43,
      "accuracy": 0.609375,
      "mean_reward": 0.7027521932326924,
      "entropy": 1.791405201384315,
      "kl_from_reference": 0.00035426783774033035,
      "retired_fraction": 0.0,
      "student_teacher_ratio": 0.15820507535911515,
      "discrepancy_improvement": 0.0034689658349575103,
      "loss": 0.059274086393203744,
      "policy_entropy": 1.791444998295649
    },
    {
      "seed": 44,
      "accuracy": 0.71875,
      "mean_reward": 0.7688085059891876,
      "entropy": 1.7915648575614136,
      "kl_from_reference": 0.0001946116606414355,
      "retired_fraction": 0.0,
      "student_teacher_ratio": 0.41000098730725626,
      "discrepancy_improvement": -0.0010755772706999012,
      "loss": 0.29333215282395164,
      "policy_entropy": 1.7915659900749181
    }
  ],
  "aggregate_metrics": {
    "accuracy_mean": 0.515625,
    "accuracy_std": 0.2628531850196988,
    "discrepancy_improvement_mean": 0.0006018582865911412,
    "discrepancy_improvement_std": 0.0024949363908838293,
    "entropy_mean": 1.7914199037475138,
    "entropy_std": 0.00013819046287153177,
    "kl_from_reference_mean": 0.0003395654745414047,
    "kl_from_reference_std": 0.00013819046287191176,
    "loss_mean": 0.13072780892840888,
    "loss_std": 0.1411634564825714,
    "mean_reward_mean": 0.624076078705196,
    "mean_reward_std": 0.1962763142345442,
    "policy_entropy_mean": 1.791436823723376,
    "policy_entropy_std": 0.00013344155903775967,
    "retired_fraction_mean": 0.0,
    "retired_fraction_std": 0.0,
    "seed_mean": 43.0,
    "seed_std": 1.0,
    "student_teacher_ratio_mean": 0.2371688129407189,
    "student_teacher_ratio_std": 0.14986246323474173
  },
  "manifest_ref": "post-training:retire-opd",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42,
      43,
      44
    ],
    "formal_comparison": false,
    "diagnostic_only": true,
    "claim_policy": "paper-scale claims excluded"
  },
  "provenance": {
    "commit": "working tree before commit",
    "command": "PYTHONPATH=src python scripts/generate_sep_19_2026_artifacts.py",
    "artifact_path": "docs/post-training/2609.20784-retire-opd/metrics/mechanism-seeds42-44.json",
    "dataset_fingerprint": "sep19-public-fixture-v1"
  }
}