{
 "limit": "held-out sets contain 40-43 cases; the exact test is valid but low-power, and it is reported as descriptive evidence, not a substitute for an independent evaluator.",
 "rows": [
  {
   "baseline_correct": 11,
   "baseline_only": 0,
   "held_n": 26,
   "mcnemar_exact_p": 0.00048828,
   "procedure_correct": 23,
   "procedure_only": 12,
   "skill": "S1_software_repair"
  },
  {
   "baseline_correct": 36,
   "baseline_only": 0,
   "held_n": 40,
   "mcnemar_exact_p": 0.125,
   "procedure_correct": 40,
   "procedure_only": 4,
   "skill": "S2_science_causal"
  },
  {
   "baseline_correct": 1,
   "baseline_only": 0,
   "held_n": 37,
   "mcnemar_exact_p": 0.0,
   "procedure_correct": 36,
   "procedure_only": 35,
   "skill": "S3_data_quality"
  },
  {
   "baseline_correct": 13,
   "baseline_only": 0,
   "held_n": 39,
   "mcnemar_exact_p": 3e-08,
   "procedure_correct": 39,
   "procedure_only": 26,
   "skill": "S4_math_validity"
  },
  {
   "baseline_correct": 26,
   "baseline_only": 0,
   "held_n": 36,
   "mcnemar_exact_p": 0.00195312,
   "procedure_correct": 36,
   "procedure_only": 10,
   "skill": "S5_ops_invariant_gate"
  }
 ],
 "schema": "cain.e111.skill_stats.v2"
}
