{
  "schema_version": 2,
  "method": "growmtp",
  "dataset": "deterministic mechanism mini-suite",
  "seeds": [
    42,
    43,
    44
  ],
  "runs": [
    {
      "seed": 42,
      "accuracy": 0.484375,
      "mean_reward": 0.5932800534207904,
      "entropy": 1.7915373241416839,
      "kl_from_reference": 0.0002221450803712562,
      "draft_acceptance_depth": 2.0,
      "depth_coupled_loss": 1.7917594692280552,
      "detached_backbone": 1.0,
      "loss": -0.0005657581304911036,
      "policy_entropy": 1.7916388681099593
    },
    {
      "seed": 43,
      "accuracy": 0.28125,
      "mean_reward": 0.45429363230689995,
      "entropy": 1.7915509840559742,
      "kl_from_reference": 0.00020848516608080938,
      "draft_acceptance_depth": 3.0,
      "depth_coupled_loss": 1.7917594692280552,
      "detached_backbone": 1.0,
      "loss": 0.003284238310573853,
      "policy_entropy": 1.7914783954164504
    },
    {
      "seed": 44,
      "accuracy": 0.4375,
      "mean_reward": 0.545742335800438,
      "entropy": 1.7887851093723903,
      "kl_from_reference": 0.0029743598496646665,
      "draft_acceptance_depth": 1.0,
      "depth_coupled_loss": 1.791759469228055,
      "detached_backbone": 1.0,
      "loss": -0.017561978467025646,
      "policy_entropy": 1.7887766839205061
    }
  ],
  "aggregate_metrics": {
    "accuracy_mean": 0.4010416666666667,
    "accuracy_std": 0.10635717633678196,
    "depth_coupled_loss_mean": 1.7917594692280552,
    "depth_coupled_loss_std": 1.2819751242557092e-16,
    "detached_backbone_mean": 1.0,
    "detached_backbone_std": 0.0,
    "draft_acceptance_depth_mean": 2.0,
    "draft_acceptance_depth_std": 1.0,
    "entropy_mean": 1.7906244725233496,
    "entropy_std": 0.001592949857699355,
    "kl_from_reference_mean": 0.0011349966987055774,
    "kl_from_reference_std": 0.0015929498576992991,
    "loss_mean": -0.004947832762314299,
    "loss_std": 0.011092480427921386,
    "mean_reward_mean": 0.5311053405093761,
    "mean_reward_std": 0.07063984383387924,
    "policy_entropy_mean": 1.7906313158156386,
    "policy_entropy_std": 0.001608161208079737,
    "seed_mean": 43.0,
    "seed_std": 1.0
  },
  "manifest_ref": "post-training:growmtp",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42,
      43,
      44
    ],
    "formal_comparison": false,
    "diagnostic_only": true,
    "claim_policy": "mechanism execution only; not a paper-scale capability result"
  },
  "provenance": {
    "commit": "working tree before commit",
    "command": "PYTHONPATH=src python scripts/generate_sep_16_2026_artifacts.py",
    "artifact_path": "docs/post-training/2609.16648-growmtp/metrics/mechanism-seeds42-44.json",
    "dataset_fingerprint": "deterministic-sep16-mechanism-mini-suite-v1"
  }
}