{
  "bespoke_eval": {
    "n": 324,
    "accuracy": 0.8271604938271605,
    "nll": 0.7308274886209167,
    "brier_to_reference": 0.31239001056754634,
    "hard_label_ece10": 0.15284857863329301,
    "hard_label_n": 324,
    "soft_target_n": 0,
    "soft_target_tvd": null,
    "hard_label_nll": 0.7308274886209167,
    "hard_label_brier": 0.31239001056754634,
    "mean_confidence": 0.9726037056817401
  },
  "synthetic_jevbench_proxy": {
    "n": 534,
    "accuracy": 0.8258426966292135,
    "axes": {
      "intelligence": 78.5377656826666,
      "calibration": 80.4126946325609,
      "speed": 88.53288735436024,
      "cost": 42.600857357940114
    },
    "composite": 69.86047453614847,
    "invalid": 0,
    "latency": {
      "n": 242,
      "p50_seconds": 0.1018778674999794,
      "p95_seconds": 0.12314256449994894,
      "endpoint_kind": "gpu"
    },
    "estimated_usd_per_1000": 0.08190393258426966,
    "environment": "Modal H100 SXM80GB; GPU-resident HTTP loopback client and native scorer; serial concurrency1; no internet transit; no KV/prefix cache; no quantization; no generative decoding; one excluded synthetic warmup",
    "reference_prices_per_million": [
      0.1,
      0.15
    ],
    "task_sha256": "1ed70789ec11f468e78e1da4151ebe69d509a46c11f4e69c2c27b1551bce4e78",
    "input_sha256": "af2dab21b44fae8ea791678f3a96fff1f4a3cd5dbc1ee0814fac21bc2331dbe7"
  },
  "temperature": 1.0,
  "candidate_orders": 1,
  "official_jevbench": false,
  "probability_source": "native allowed-token logits and softmax"
}