{
  "cohort": "32 reserved chronological T16 recordings, calibration80 from two earlier blocks; ROI chosen on separate development8",
  "prediction_sha256": "a368bfdaefb2003ac27445b4da19819c97a18627868975d9202696ef1ef52dac",
  "strictly_novel_target_ids": [
    "t0575fbc6",
    "t1fe83bc3",
    "t3afd73b4",
    "t41450074",
    "t44bb2095",
    "t4539ffbe",
    "t4d2d7415",
    "t514a6795",
    "t52b24bf6",
    "t588781f6",
    "t59dcd856",
    "t5f2d20a0",
    "t6453e6ae",
    "t6f0fdfe0",
    "t8485b6b9",
    "t848cf4da",
    "t9ce92bac",
    "ta2aa0fcb",
    "ta8a65775",
    "tbfeb0ff2",
    "tca769afd",
    "tcd92c113",
    "tced8ecfc",
    "td133fe68",
    "td28e3558",
    "td6e48564",
    "te3243945",
    "te366b91c",
    "tef9d94a6",
    "tf75891ff"
  ],
  "sol": {
    "all": {
      "n": 32,
      "edits": 24,
      "reference_words": 143,
      "WER": 0.16783216783216784,
      "exact": 18
    },
    "strict": {
      "n": 30,
      "edits": 24,
      "reference_words": 137,
      "WER": 0.17518248175182483,
      "exact": 16
    }
  },
  "comparisons": {
    "generic": {
      "all": {
        "n": 32,
        "edits": 55,
        "reference_words": 143,
        "WER": 0.38461538461538464,
        "exact": 6
      },
      "strict": {
        "n": 30,
        "edits": 53,
        "reference_words": 137,
        "WER": 0.38686131386861317,
        "exact": 6
      },
      "paired_all": {
        "net_edit_reduction_with_sol": 31,
        "WER_reduction_percentage_points": 21.678321678321677,
        "descriptive_trial_bootstrap_95pct_percentage_points": [
          14.000000000000002,
          29.927007299270077
        ],
        "sol_lower_error_trials": 20,
        "algorithm_lower_error_trials": 1,
        "equal_error_trials": 11,
        "scope": "Paired within one chronological recording block. Does not account for unknown serial dependencies or support population generalization."
      },
      "paired_strict": {
        "net_edit_reduction_with_sol": 29,
        "WER_reduction_percentage_points": 21.16788321167883,
        "descriptive_trial_bootstrap_95pct_percentage_points": [
          13.076923076923078,
          29.71014492753623
        ],
        "sol_lower_error_trials": 18,
        "algorithm_lower_error_trials": 1,
        "equal_error_trials": 11,
        "scope": "Paired within one chronological recording block. Does not account for unknown serial dependencies or support population generalization."
      },
      "wall_seconds_algorithm": 36.25285691700992
    },
    "t16cal": {
      "all": {
        "n": 32,
        "edits": 51,
        "reference_words": 143,
        "WER": 0.35664335664335667,
        "exact": 4
      },
      "strict": {
        "n": 30,
        "edits": 49,
        "reference_words": 137,
        "WER": 0.35766423357664234,
        "exact": 4
      },
      "paired_all": {
        "net_edit_reduction_with_sol": 27,
        "WER_reduction_percentage_points": 18.88111888111888,
        "descriptive_trial_bootstrap_95pct_percentage_points": [
          13.043478260869565,
          25.3335616438356
        ],
        "sol_lower_error_trials": 20,
        "algorithm_lower_error_trials": 0,
        "equal_error_trials": 12,
        "scope": "Paired within one chronological recording block. Does not account for unknown serial dependencies or support population generalization."
      },
      "paired_strict": {
        "net_edit_reduction_with_sol": 25,
        "WER_reduction_percentage_points": 18.248175182481752,
        "descriptive_trial_bootstrap_95pct_percentage_points": [
          12.121212121212121,
          24.822695035460992
        ],
        "sol_lower_error_trials": 18,
        "algorithm_lower_error_trials": 0,
        "equal_error_trials": 12,
        "scope": "Paired within one chronological recording block. Does not account for unknown serial dependencies or support population generalization."
      },
      "wall_seconds_algorithm": 1.0316567080153618
    }
  },
  "limits": "One participant/session and one evaluation block. Full-block label-free mean correction uses unlabeled recordings. This is within-session generalization, not cross-session or participant transfer. The Sol workflow and numerical inventory were developed before confirmation; no inference about arbitrary language or private thoughts.",
  "usage": [
    {
      "input_tokens": 2833327,
      "cached_input_tokens": 2715008,
      "cache_write_input_tokens": 0,
      "output_tokens": 45737,
      "reasoning_output_tokens": 30720
    }
  ],
  "wall_seconds_sol": 1376.3414537906647
}