{
  "determinism": {
    "class": "best_effort",
    "notes": [
      "MPS: bit-identical across runs is best-effort"
    ],
    "seed": 0
  },
  "null_stats": {},
  "probes": [
    {
      "base_value": null,
      "ci_95": [
        0.00845183397355073,
        0.02976923823248102
      ],
      "evidence": {
        "divergence_kind": "js",
        "max": 0.036150663063872254,
        "num_prompts": 4,
        "per_prompt": [
          0.01062496373830732,
          0.006711906030413639,
          0.013671617802962006,
          0.036150663063872254
        ],
        "raw_ci_95": [
          0.00845183397355073,
          0.02976923823248102
        ],
        "weight": 1.0,
        "z_by_rank": null
      },
      "ft_value": null,
      "kind": "delta_kl",
      "message": "mean js=0.0168 (\u2265 0.0) (no calibration for delta_kl)",
      "name": "dk_golden",
      "raw": 0.016789787658888805,
      "score": 0.0242225433930576,
      "verdict": "pass",
      "z_score": null
    },
    {
      "base_value": null,
      "ci_95": [
        0.0,
        0.0
      ],
      "evidence": {
        "fraction_regressed": 0.0,
        "mean_delta_nats": -0.03518791794776917,
        "raw_ci_95": [
          0.0,
          0.0
        ],
        "regressed_count": 0,
        "regression_nats_threshold": 1.0,
        "total_items": 20,
        "weight": 1.0,
        "worst_offenders": [],
        "z_by_rank": null
      },
      "ft_value": -0.03518791794776917,
      "kind": "calibration_drift",
      "message": "0/20 items regressed >1.0 nats (frac=0.0%), mean_delta=-0.035 nats/tok (no calibration for calibration_drift)",
      "name": "cal_golden",
      "raw": 0.0,
      "score": 0.8572832218805948,
      "verdict": "pass",
      "z_score": null
    }
  ],
  "schema_version": 1,
  "score": {
    "band": "partial",
    "components": {
      "ablation": 0.0,
      "adherence": 0.0242225433930576,
      "attribution": 0.0,
      "baseline": 0.0,
      "calibration": 0.8572832218805948
    },
    "findings": [
      "adherence score is 0.02 \u2014 below the noise threshold"
    ],
    "overall": 0.3574468147880725,
    "weights": {
      "ablation": 0.15,
      "adherence": 0.3,
      "attribution": 0.35,
      "baseline": 0.0,
      "calibration": 0.2
    }
  },
  "spec_path": "<memory>"
}