{
  "date": "2026-08-02",
  "seed": 42,
  "post_training": {
    "paper": "2607.28590",
    "algorithm": "vad",
    "dataset": "gsm8k-candidate",
    "steps": 120,
    "baseline_accuracy": 0.171875,
    "final_accuracy": 0.796875,
    "final_mean_reward": 0.8153364653766422,
    "kl_from_reference": 0.009165395625760278,
    "online_teacher_calls": 240,
    "projection_active_rate": 1.0,
    "mean_teacher_correction_alignment": 0.5145374600623333,
    "fidelity": "branch-aware counterfactual target reconstruction with symmetric JSD on candidate-policy evidence axes"
  },
  "agent_evaluation": {
    "paper": "2607.28609",
    "method": "os-shepherd",
    "benchmark": "osreward-mini",
    "episodes": 120,
    "accuracy": 1.0,
    "success_recall": 1.0,
    "fail_recall": 1.0,
    "balanced_accuracy": 1.0,
    "leniency_rate": 0.0,
    "official_dataset_used": false
  },
  "schema_version": 2,
  "manifest_ref": "experiments:latest-20260802-seed42",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "provenance": {
    "artifact_path": "docs/experiments/latest-20260802-seed42.json",
    "historical_migration": "historical-metrics-v2-2026-08-09",
    "original_code_commit": "not recorded",
    "dataset_fingerprint": "not recorded in historical artifact"
  }
}