{
  "algorithm": "rlvr-fusion",
  "dataset": "arithmetic-smoke",
  "baseline": {
    "accuracy": 0.1953125,
    "mean_reward": 0.359689058826365,
    "entropy": 1.791759469222057,
    "kl_from_reference": 0.0
  },
  "final": {
    "accuracy": 0.28125,
    "mean_reward": 0.45564911291653953,
    "entropy": 0.5678259511881575,
    "kl_from_reference": 1.2239335180338968
  },
  "relative_accuracy": 0.44,
  "training": {
    "steps": 100,
    "learning_rate": 0.08,
    "group_size": 4,
    "train_examples": 512,
    "validation_examples": 128,
    "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": 6,
    "last_diagnostics": {
      "executed_fusion_paradigm": 0.0,
      "task_vector_cosine_mean": -0.10631733125749472,
      "merge_vector_norm": 0.038562437179543003,
      "mix_rl_advantage_std": 0.27089713477857746,
      "mopd_positive_teacher_fraction": 0.75,
      "loss": -0.035213863051630406,
      "policy_entropy": 0.1942487660165422
    },
    "fidelity": "mechanism reproduction on a candidate-policy model"
  },
  "history": [
    {
      "step": 1.0,
      "loss": -0.01473984966018551,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7911789512877143,
      "kl_from_reference": 0.0005805179343405969
    },
    {
      "step": 10.0,
      "loss": -0.022180850242191828,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7510985386716067,
      "kl_from_reference": 0.040660930550447935
    },
    {
      "step": 20.0,
      "loss": -0.015125127663891617,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.6424093883063193,
      "kl_from_reference": 0.14935008091573568
    },
    {
      "step": 30.0,
      "loss": -0.05875658247289874,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.4329210110733896,
      "kl_from_reference": 0.3588384581486656
    },
    {
      "step": 40.0,
      "loss": -0.041921484467702,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.2615163728650098,
      "kl_from_reference": 0.5302430963570456
    },
    {
      "step": 50.0,
      "loss": -0.07821672795493304,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.0938462235513384,
      "kl_from_reference": 0.6979132456707166
    },
    {
      "step": 60.0,
      "loss": -0.0436382342167632,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 0.9385610983059598,
      "kl_from_reference": 0.8531983709160951
    },
    {
      "step": 70.0,
      "loss": -0.05005620648333211,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 0.8766189235396149,
      "kl_from_reference": 0.9151405456824397
    },
    {
      "step": 80.0,
      "loss": -0.039961806600904745,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 0.7248127534170089,
      "kl_from_reference": 1.0669467158050465
    },
    {
      "step": 90.0,
      "loss": -0.012661590528451295,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 0.6475824027798488,
      "kl_from_reference": 1.1441770664422062
    },
    {
      "step": 100.0,
      "loss": -0.035213863051630406,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 0.5678259511881575,
      "kl_from_reference": 1.2239335180338968
    }
  ],
  "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/2608.27409-rlvr-fusion/metrics/arithmetic-smoke-seed42.json",
    "dataset_fingerprint": "not recorded in historical artifact",
    "historical_migration": "historical-metrics-v2-2026-08-09",
    "original_code_commit": "not recorded"
  }
}