{
  "algorithm": "spear",
  "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": 1.7276721105580042,
    "kl_from_reference": 0.064087358664051
  },
  "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": {
      "symbolic_milestone_count": 2.0,
      "lcs_alignment_reward": 0.7,
      "neural_prm_calls": 0.0,
      "loss": -0.025924870951007074,
      "policy_entropy": 1.7214208882257205
    },
    "fidelity": "mechanism reproduction on a candidate-policy model"
  },
  "history": [
    {
      "step": 1.0,
      "loss": -0.0,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7917292900581454,
      "kl_from_reference": 3.0179163911054533e-05
    },
    {
      "step": 10.0,
      "loss": -0.0002878191576955874,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7911273020207281,
      "kl_from_reference": 0.0006321672013269301
    },
    {
      "step": 20.0,
      "loss": -0.026543041815749124,
      "accuracy": 0.296875,
      "mean_reward": 0.468677279507009,
      "entropy": 1.7894840298593941,
      "kl_from_reference": 0.0022754393626611817
    },
    {
      "step": 30.0,
      "loss": -0.04197151147495051,
      "accuracy": 0.28125,
      "mean_reward": 0.45625529841796264,
      "entropy": 1.7869793433028995,
      "kl_from_reference": 0.0047801259191561874
    },
    {
      "step": 40.0,
      "loss": -0.026762004370060162,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7828219168792598,
      "kl_from_reference": 0.008937552342796003
    },
    {
      "step": 50.0,
      "loss": -0.07724105828586329,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7749300183316556,
      "kl_from_reference": 0.016829450890399014
    },
    {
      "step": 60.0,
      "loss": -0.058392662234322906,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.76533611979077,
      "kl_from_reference": 0.026423349431284958
    },
    {
      "step": 70.0,
      "loss": -0.013087972825745864,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7579867599092724,
      "kl_from_reference": 0.03377270931278307
    },
    {
      "step": 80.0,
      "loss": -0.0694660197814162,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7481474995143844,
      "kl_from_reference": 0.04361196970767061
    },
    {
      "step": 90.0,
      "loss": -0.2258006448431887,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.735283726461001,
      "kl_from_reference": 0.056475742761053355
    },
    {
      "step": 100.0,
      "loss": -0.025924870951007074,
      "accuracy": 0.28125,
      "mean_reward": 0.45564911291653953,
      "entropy": 1.7276721105580042,
      "kl_from_reference": 0.064087358664051
    }
  ],
  "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/2608.26550-spear/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"
  }
}