{
  "banking": {
    "source": "massive/evidence/banking_historical_sam_v1_test.tar.gz",
    "sha256": "d9a1b6a1c085bdc06a32c8480d67e9136dcf454b7d9233844a21aa78b5a4c925",
    "correct": 2899,
    "n": 3080,
    "row_id_note": "artifact_row_index is the preserved prediction artifact order, not an official dataset ID",
    "selection": "Best observed of three SAM seeds after test evaluation",
    "three_seed_mean_percent": 94.0367965368,
    "checkpoint_sha256": "0ff739f4413497119476af928b6cff482f1c0ef0cfe9bfb96527b769df172d29"
  },
  "clinc_single": {
    "source": "artifacts/round2/compact_test_predictions.npz",
    "sha256": "0c8228d72eef1cd898119e3a27ee915f041ce0d42fe2bb875dedd5f53bc8c358",
    "correct": 4399,
    "n": 4500,
    "row_id_note": "artifact_row_index is the preserved prediction artifact order, not an official dataset ID"
  },
  "clinc_ensemble": {
    "source": "artifacts/hybrid/compact_test_predictions.npz",
    "sha256": "318c11741292661b87e706773097bec2cc6459c048f63c961cb77f5d0280e177",
    "correct": 4411,
    "n": 4500,
    "row_id_note": "artifact_row_index is the preserved prediction artifact order, not an official dataset ID"
  },
  "massive": {
    "source": "massive/evidence/final_test_20261002.tar.gz",
    "sha256": "52275a090dfa61da732d933df4de18b6822a156157b328bf86cd13cf6ad845d7",
    "correct": 391467,
    "n": 455022,
    "row_id_note": "seed + locale + sample_id identifies each prediction; 151674 test rows repeated across three seeds, not 455022 independent samples"
  }
}