{
  "files": {
    "data/sft_train.jsonl": "39ee5e9920f480736f42c80acd3ff76a1e62cc1cb250b0dace94fda239e3e68f"
  },
  "count": 7546,
  "filter_stats": {
    "raw_chains": 13245,
    "kept": 7546,
    "surplus_held": 0,
    "decontam_dropped": 0,
    "rejects": {
      "no_assertions": 3894,
      "bad_arithmetic": 1211,
      "non_composing": 387,
      "wrong_answer": 206,
      "dup_problem": 1
    },
    "keep_rate_pct": 56.97,
    "cap": 20000
  },
  "verifier_ref": {
    "verifier/step_verifier.py": "d5d146cfbb9a69e162c1a80a330c942b195bfd6d0a0961ae480cc1875989613b"
  },
  "eval_ref": {
    "eval/eval_set.jsonl": "e1573cabaf44010555ef1138cbeda17bc43deb8121524cde23d94413a5d69753",
    "eval/_eval_meta.json": "8d99764e00d564035fcc1cf361c7e44d45783623288763fe679477e8670fc88f"
  },
  "criterion": "verifiable AND correct, decontaminated vs EVAL.lock, dedup by problem text"
}