{
  "pxr": {
    "gap_rank1_rank2_mae": 0.0031,
    "gap_over_rank1_spread": 0.111,
    "pairwise_comparisons_check": 4950
  },
  "lifescibench": {
    "wilson_95": {
      "GPT-Rosalind": [
        0.327,
        0.396
      ],
      "GPT-5.5": [
        0.227,
        0.289
      ],
      "Gemini 3.1 Pro": [
        0.207,
        0.268
      ],
      "GPT-5.4": [
        0.18,
        0.237
      ],
      "Grok 4.3": [
        0.108,
        0.156
      ]
    },
    "unpaired_diff_95": {
      "Rosalind-GPT5.5": [
        0.058,
        0.15
      ],
      "GPT5.5-Gemini3.1": [
        -0.023,
        0.065
      ]
    },
    "note": "task-sampling uncertainty only; unpaired approximation (paired per-task results not public); pass rates are rounded to 0.1 point"
  },
  "aintibody": {
    "ratio_lab_over_ai": 1.193,
    "relative_improvement_pct": 16.2
  },
  "inputs": {
    "pxr": {
      "mae_rank1": 0.4061,
      "pm_rank1": 0.028,
      "mae_rank2": 0.4092,
      "pm_rank2": 0.0281,
      "n_activity_entries_pairwise": 100,
      "pairwise_reported": 4950,
      "tier1_activity": 28
    },
    "lifescibench": {
      "n_tasks": 750,
      "pass_rate": {
        "GPT-Rosalind": 0.361,
        "GPT-5.5": 0.257,
        "Gemini 3.1 Pro": 0.236,
        "GPT-5.4": 0.207,
        "Grok 4.3": 0.13
      }
    },
    "aintibody": {
      "best_ai_pM": 94.7,
      "best_lab_pM": 113.0,
      "random_beats_model_pct": 39.0
    }
  }
}
