{
  "schema_version": 2,
  "method": "tlm-dre",
  "dataset": "deterministic mechanism mini-suite",
  "seeds": [
    42,
    43,
    44
  ],
  "runs": [
    {
      "seed": 42,
      "accuracy": 0.03125,
      "mean_reward": 0.19929878377259266,
      "entropy": 1.7915431053535127,
      "kl_from_reference": 0.00021636386854292548,
      "turn_weight_entropy": 1.0986040757582243,
      "asymmetric_token_weights": 0.0869180865305372,
      "loss": -0.0002515587475113573,
      "policy_entropy": 1.7916124256403154
    },
    {
      "seed": 43,
      "accuracy": 0.015625,
      "mean_reward": 0.12425064657368294,
      "entropy": 1.7916501009427548,
      "kl_from_reference": 0.00010936827930056243,
      "turn_weight_entropy": 1.098522049874477,
      "asymmetric_token_weights": 0.08854549154317576,
      "loss": 0.0007809760648813946,
      "policy_entropy": 1.7916696398685386
    },
    {
      "seed": 44,
      "accuracy": 0.296875,
      "mean_reward": 0.46671875360459697,
      "entropy": 1.7913996961343843,
      "kl_from_reference": 0.0003597730876707044,
      "turn_weight_entropy": 1.0985512176234855,
      "asymmetric_token_weights": 0.08890838446014426,
      "loss": 0.0004980060726363922,
      "policy_entropy": 1.7914129456114245
    }
  ],
  "aggregate_metrics": {
    "accuracy_mean": 0.11458333333333333,
    "accuracy_std": 0.15806240526555748,
    "asymmetric_token_weights_mean": 0.08812398751128574,
    "asymmetric_token_weights_std": 0.001059986178600548,
    "entropy_mean": 1.791530967476884,
    "entropy_std": 0.00012564289892609654,
    "kl_from_reference_mean": 0.00022850174517139742,
    "kl_from_reference_std": 0.00012564289892591575,
    "loss_mean": 0.0003424744633354765,
    "loss_std": 0.0005335490566194765,
    "mean_reward_mean": 0.2634227279836242,
    "mean_reward_std": 0.1800139195985105,
    "policy_entropy_mean": 1.7915650037067596,
    "policy_entropy_std": 0.00013475761665104934,
    "seed_mean": 43.0,
    "seed_std": 1.0,
    "turn_weight_entropy_mean": 1.0985591144187288,
    "turn_weight_entropy_std": 4.157921274468798e-05
  },
  "manifest_ref": "post-training:tlm-dre",
  "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/2609.16760-tlm-dre/metrics/mechanism-seeds42-44.json",
    "dataset_fingerprint": "deterministic-sep16-mechanism-mini-suite-v1"
  }
}