{
  "algorithm": "clue-opsd",
  "dataset": "arithmetic-smoke",
  "baseline": {
    "accuracy": 0.125,
    "mean_reward": 0.3039061471113459,
    "entropy": 1.7917594692220566,
    "kl_from_reference": 0.0
  },
  "final": {
    "accuracy": 0.59375,
    "mean_reward": 0.6843854637614489,
    "entropy": 1.7833193753804875,
    "kl_from_reference": 0.00844009384156765
  },
  "relative_accuracy": 3.75,
  "training": {
    "steps": 100,
    "learning_rate": 0.08,
    "group_size": 4,
    "train_examples": 64,
    "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": 6,
    "last_diagnostics": {
      "clue_teacher_process_reward": 0.6118168173506832,
      "student_full_context_fraction": 1.0,
      "inference_clue_annotations": 0.0,
      "ema_teacher_update_rate": 0.99,
      "loss": -0.009812570258071793,
      "policy_entropy": 1.784171065798132
    },
    "fidelity": "mechanism reproduction on a candidate-policy model"
  },
  "history": [
    {
      "step": 1.0,
      "loss": -8.326672684688674e-17,
      "accuracy": 0.203125,
      "mean_reward": 0.38792858424974397,
      "entropy": 1.7917479919662007,
      "kl_from_reference": 1.1477255854350929e-05
    },
    {
      "step": 10.0,
      "loss": -0.0046547628835259675,
      "accuracy": 0.234375,
      "mean_reward": 0.41366344934761173,
      "entropy": 1.7913040582526625,
      "kl_from_reference": 0.000455410969392857
    },
    {
      "step": 20.0,
      "loss": -0.0051974266203446765,
      "accuracy": 0.328125,
      "mean_reward": 0.48961081352779734,
      "entropy": 1.7909264042347166,
      "kl_from_reference": 0.0008330649873378875
    },
    {
      "step": 30.0,
      "loss": -0.018729012705827868,
      "accuracy": 0.53125,
      "mean_reward": 0.6398879034889423,
      "entropy": 1.79037109946577,
      "kl_from_reference": 0.001388369756284624
    },
    {
      "step": 40.0,
      "loss": -0.024811952597804332,
      "accuracy": 0.578125,
      "mean_reward": 0.6732114862463144,
      "entropy": 1.7893376616008378,
      "kl_from_reference": 0.00242180762121793
    },
    {
      "step": 50.0,
      "loss": -0.012853307933088046,
      "accuracy": 0.578125,
      "mean_reward": 0.6734559395682362,
      "entropy": 1.7881245064340061,
      "kl_from_reference": 0.0036349627880490525
    },
    {
      "step": 60.0,
      "loss": -0.008235096482005903,
      "accuracy": 0.578125,
      "mean_reward": 0.6734559395682362,
      "entropy": 1.7872041286461537,
      "kl_from_reference": 0.004555340575901599
    },
    {
      "step": 70.0,
      "loss": -0.021656507786097304,
      "accuracy": 0.578125,
      "mean_reward": 0.6734559395682362,
      "entropy": 1.7862798034318261,
      "kl_from_reference": 0.005479665790228908
    },
    {
      "step": 80.0,
      "loss": 0.00026424079948296486,
      "accuracy": 0.59375,
      "mean_reward": 0.6843854637614489,
      "entropy": 1.785320600850309,
      "kl_from_reference": 0.006438868371745859
    },
    {
      "step": 90.0,
      "loss": -0.02068613232934828,
      "accuracy": 0.59375,
      "mean_reward": 0.6845040857412398,
      "entropy": 1.7840748978423364,
      "kl_from_reference": 0.007684571379719067
    },
    {
      "step": 100.0,
      "loss": -0.009812570258071793,
      "accuracy": 0.59375,
      "mean_reward": 0.6843854637614489,
      "entropy": 1.7833193753804875,
      "kl_from_reference": 0.00844009384156765
    }
  ],
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed mechanism result; do not claim a stable improvement"
  },
  "schema_version": 2,
  "manifest_ref": "post-training:clue-opsd-arithmetic-smoke-seed42",
  "provenance": {
    "artifact_path": "docs/post-training/2608.25356-clue-opsd/metrics/clue-opsd-arithmetic-smoke-seed42.json",
    "dataset_fingerprint": "not recorded in historical artifact",
    "historical_migration": "historical-metrics-v2-2026-08-09",
    "original_code_commit": "not recorded"
  }
}
