{
  "schema_version": 2,
  "method": "trajectory-learnability",
  "dataset": "deterministic mechanism mini-suite",
  "seeds": [
    42,
    43,
    44
  ],
  "runs": [
    {
      "seed": 42,
      "accuracy": 0.203125,
      "mean_reward": 0.38792858424974397,
      "entropy": 1.7916892567362355,
      "kl_from_reference": 7.021248581927212e-05,
      "learnability_entropy": 1.3862269751957492,
      "positive_learnability_fraction": 0.0,
      "max_trajectory_weight": 0.2525174534616529,
      "loss": 0.00995548092024349,
      "policy_entropy": 1.7916917381113617
    },
    {
      "seed": 43,
      "accuracy": 0.59375,
      "mean_reward": 0.690636681277262,
      "entropy": 1.7917171174992201,
      "kl_from_reference": 4.23517228346308e-05,
      "learnability_entropy": 1.3860977814245667,
      "positive_learnability_fraction": 0.0,
      "max_trajectory_weight": 0.25470673413322836,
      "loss": 0.0147467188253002,
      "policy_entropy": 1.7917225572374864
    },
    {
      "seed": 44,
      "accuracy": 0.734375,
      "mean_reward": 0.7788216538811203,
      "entropy": 1.791741660834862,
      "kl_from_reference": 1.780838719277837e-05,
      "learnability_entropy": 1.3861521637317276,
      "positive_learnability_fraction": 0.0,
      "max_trajectory_weight": 0.2565500913320996,
      "loss": 0.07462340471983955,
      "policy_entropy": 1.7917412025108397
    }
  ],
  "aggregate_metrics": {
    "accuracy_mean": 0.5104166666666666,
    "accuracy_std": 0.27525438045257944,
    "entropy_mean": 1.7917160116901059,
    "entropy_std": 2.621954420965292e-05,
    "kl_from_reference_mean": 4.345753194889377e-05,
    "kl_from_reference_std": 2.6219544209648487e-05,
    "learnability_entropy_mean": 1.3861589734506812,
    "learnability_entropy_std": 6.48655288443568e-05,
    "loss_mean": 0.03310853482179441,
    "loss_std": 0.03603265612128009,
    "max_trajectory_weight_mean": 0.2545914263089936,
    "max_trajectory_weight_std": 0.0020187902241906113,
    "mean_reward_mean": 0.6191289731360421,
    "mean_reward_std": 0.20502283339708147,
    "policy_entropy_mean": 1.7917184992865627,
    "policy_entropy_std": 2.4980631060955077e-05,
    "positive_learnability_fraction_mean": 0.0,
    "positive_learnability_fraction_std": 0.0,
    "seed_mean": 43.0,
    "seed_std": 1.0
  },
  "manifest_ref": "post-training:trajectory-learnability",
  "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/2609.18321-trajectory-learnability/metrics/mechanism-seeds42-44.json",
    "dataset_fingerprint": "sep19-public-fixture-v1"
  }
}