{
  "algorithm": "video-opsd",
  "dataset": "arithmetic-smoke",
  "baseline": {
    "accuracy": 0.1953125,
    "mean_reward": 0.359689058826365,
    "entropy": 1.791759469222057,
    "kl_from_reference": 0.0
  },
  "final": {
    "accuracy": 0.65625,
    "mean_reward": 0.7322317743621116,
    "entropy": 1.7915437756945163,
    "kl_from_reference": 0.0002156935275388919
  },
  "relative_accuracy": 2.36,
  "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": {
      "privileged_frame_fraction": 0.3333333333333333,
      "evidence_token_weight_mean": 0.7223407453299683,
      "evidence_teacher_gap": -0.001218720056453182,
      "full_video_student_fraction": 1.0,
      "loss": 0.008317944131254404,
      "policy_entropy": 1.7915145278583224
    },
    "fidelity": "mechanism reproduction on a candidate-policy model"
  },
  "history": [
    {
      "step": 1.0,
      "loss": -0.0,
      "accuracy": 0.5,
      "mean_reward": 0.5962298893894342,
      "entropy": 1.791759003317489,
      "kl_from_reference": 4.6590456644300776e-07
    },
    {
      "step": 10.0,
      "loss": 0.0007829367278381067,
      "accuracy": 0.609375,
      "mean_reward": 0.6836174400973312,
      "entropy": 1.7917581193889471,
      "kl_from_reference": 1.3498331080558839e-06
    },
    {
      "step": 20.0,
      "loss": -5.4228480671344005e-05,
      "accuracy": 0.578125,
      "mean_reward": 0.6774922968851133,
      "entropy": 1.7917534151065158,
      "kl_from_reference": 6.054115538527473e-06
    },
    {
      "step": 30.0,
      "loss": 0.0029506945635640464,
      "accuracy": 0.6640625,
      "mean_reward": 0.7272569161371335,
      "entropy": 1.7917354547362323,
      "kl_from_reference": 2.4014485822806664e-05
    },
    {
      "step": 40.0,
      "loss": 0.0005015369329902692,
      "accuracy": 0.671875,
      "mean_reward": 0.7353747505778383,
      "entropy": 1.7917314472074595,
      "kl_from_reference": 2.8022014595927342e-05
    },
    {
      "step": 50.0,
      "loss": 0.005000518239754609,
      "accuracy": 0.6484375,
      "mean_reward": 0.7261273225271052,
      "entropy": 1.7917087571798063,
      "kl_from_reference": 5.07120422488667e-05
    },
    {
      "step": 60.0,
      "loss": 0.004461484499640549,
      "accuracy": 0.65625,
      "mean_reward": 0.7323861466712372,
      "entropy": 1.7916621992380617,
      "kl_from_reference": 9.726998399366996e-05
    },
    {
      "step": 70.0,
      "loss": 0.003314201020317625,
      "accuracy": 0.65625,
      "mean_reward": 0.7317359481306855,
      "entropy": 1.7916439853716855,
      "kl_from_reference": 0.00011548385036980587
    },
    {
      "step": 80.0,
      "loss": 0.0064455249796798696,
      "accuracy": 0.6640625,
      "mean_reward": 0.7345393143452014,
      "entropy": 1.7916215242969034,
      "kl_from_reference": 0.00013794492515095756
    },
    {
      "step": 90.0,
      "loss": 0.012651779995909858,
      "accuracy": 0.65625,
      "mean_reward": 0.7291470618481051,
      "entropy": 1.7915864240466373,
      "kl_from_reference": 0.00017304517541729596
    },
    {
      "step": 100.0,
      "loss": 0.008317944131254404,
      "accuracy": 0.65625,
      "mean_reward": 0.7322317743621116,
      "entropy": 1.7915437756945163,
      "kl_from_reference": 0.0002156935275388919
    }
  ],
  "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/2608.27065-video-opsd/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"
  }
}