{
  "complete": true,
  "checked_utc": "2026-10-06T15:59:45.382575+00:00",
  "scope": "all",
  "split": "confirmation",
  "backend": "FP4 packed weights, BF16 activations/GEMM, Triton per-matrix dequantization on AMD",
  "same_native_prompt_readout_temperature_API": true,
  "hardware_speed_validated": false,
  "full_decision_index": null,
  "overall": {
    "n": 544,
    "agreement_percent": 97.24264705882354,
    "bf16_accuracy_percent": 77.57352941176471,
    "fp4_accuracy_percent": 76.83823529411765,
    "mean_KL_bf16_to_fp4": 0.011581205562651279
  },
  "by_domain": {
    "ESCI": {
      "n": 32,
      "agreement_percent": 96.875,
      "bf16_accuracy_percent": 59.375,
      "fp4_accuracy_percent": 56.25,
      "mean_KL_bf16_to_fp4": 0.008368359788850645
    },
    "HoVer": {
      "n": 32,
      "agreement_percent": 93.75,
      "bf16_accuracy_percent": 81.25,
      "fp4_accuracy_percent": 75.0,
      "mean_KL_bf16_to_fp4": 0.023173074279276965
    },
    "Habermas": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 31.25,
      "fp4_accuracy_percent": 31.25,
      "mean_KL_bf16_to_fp4": 0.02629049770095534
    },
    "When2Call": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 93.75,
      "fp4_accuracy_percent": 93.75,
      "mean_KL_bf16_to_fp4": 0.002222273564954566
    },
    "sarcasm_ar": {
      "n": 32,
      "agreement_percent": 90.625,
      "bf16_accuracy_percent": 62.5,
      "fp4_accuracy_percent": 59.375,
      "mean_KL_bf16_to_fp4": 0.012054604471233589
    },
    "old_general/choice": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 84.375,
      "fp4_accuracy_percent": 84.375,
      "mean_KL_bf16_to_fp4": 0.002750536940483584
    },
    "guard/humor": {
      "n": 32,
      "agreement_percent": 93.75,
      "bf16_accuracy_percent": 40.625,
      "fp4_accuracy_percent": 40.625,
      "mean_KL_bf16_to_fp4": 0.027150870865254688
    },
    "Phishing": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 96.875,
      "fp4_accuracy_percent": 96.875,
      "mean_KL_bf16_to_fp4": 0.006934639060563293
    },
    "old_general/noul": {
      "n": 32,
      "agreement_percent": 96.875,
      "bf16_accuracy_percent": 93.75,
      "fp4_accuracy_percent": 90.625,
      "mean_KL_bf16_to_fp4": 0.0035566811986626184
    },
    "ANLI": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 84.375,
      "fp4_accuracy_percent": 84.375,
      "mean_KL_bf16_to_fp4": 0.009098772482953045
    },
    "VAST": {
      "n": 32,
      "agreement_percent": 93.75,
      "bf16_accuracy_percent": 71.875,
      "fp4_accuracy_percent": 68.75,
      "mean_KL_bf16_to_fp4": 0.009012711907023834
    },
    "SATA": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 96.875,
      "fp4_accuracy_percent": 96.875,
      "mean_KL_bf16_to_fp4": 0.006667878952544886
    },
    "ContractNLI": {
      "n": 32,
      "agreement_percent": 93.75,
      "bf16_accuracy_percent": 87.5,
      "fp4_accuracy_percent": 93.75,
      "mean_KL_bf16_to_fp4": 0.005304457598214596
    },
    "ACOS": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 84.375,
      "fp4_accuracy_percent": 84.375,
      "mean_KL_bf16_to_fp4": 0.011603451698692325
    },
    "RAGTruth": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 90.625,
      "fp4_accuracy_percent": 90.625,
      "mean_KL_bf16_to_fp4": 0.006294842546801113
    },
    "sarcasm_en": {
      "n": 32,
      "agreement_percent": 93.75,
      "bf16_accuracy_percent": 68.75,
      "fp4_accuracy_percent": 68.75,
      "mean_KL_bf16_to_fp4": 0.00877219076521019
    },
    "guard/science": {
      "n": 32,
      "agreement_percent": 100.0,
      "bf16_accuracy_percent": 90.625,
      "fp4_accuracy_percent": 90.625,
      "mean_KL_bf16_to_fp4": 0.027624650743396446
    }
  }
}