{
  "algorithm": "rrc",
  "seed": 42,
  "dataset": "arithmetic-smoke",
  "baseline": {
    "accuracy": 0.234375,
    "mean_reward": 0.38409619661640243,
    "entropy": 1.7917594692220566,
    "kl_from_reference": 0.0
  },
  "final": {
    "accuracy": 0.453125,
    "mean_reward": 0.5870797230492214,
    "entropy": 1.7274924293110652,
    "kl_from_reference": 0.06426703991098916
  },
  "relative_accuracy": 0.9333333333333333,
  "training": {
    "steps": 120,
    "learning_rate": 0.08,
    "group_size": 4,
    "train_examples": 256,
    "validation_examples": 64,
    "data_source": "deterministic arithmetic smoke suite",
    "teacher_cache_entries": 0,
    "teacher_prefill_calls": 0,
    "online_teacher_calls": 0,
    "drift_events": 0,
    "critic_updates": 0,
    "rollout_policy_refreshes": 0,
    "last_diagnostics": {
      "ranked_responses": 4.0,
      "anchor_reward": 1.0,
      "loss": -0.28165563736511134,
      "policy_entropy": 1.734382553119111
    },
    "fidelity": "mechanism reproduction on a candidate-policy model"
  },
  "schema_version": 2,
  "manifest_ref": "post-training:arithmetic-smoke-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/post-training/2608.06310-rrc/metrics/arithmetic-smoke-seed42.json",
    "original_code_commit": "working tree",
    "dataset_fingerprint": "deterministic built-in mini-suite",
    "historical_migration": "historical-metrics-v2-2026-08-09"
  }
}