Endpoint tests missed the largest accepted error in 19 of 35 shaft cases for eight fixed SEEG readers under ideal constant gains within ±6 dB

Tests at −6 and +6 dB missed the largest error among accepted labels in 19 of 35 shaft cases from eight people. The frozen SEEG readers used fixed early confidence gates. An exact audit of the full ideal constant-gain interval raised the equal-person mean maximum from 37.24% to 37.57%. The gap was 0.33 percentage points, with a descriptive 95% person bootstrap interval of 0.15 to 0.54 points. We used 975 later trials and 1,950 unique windows from OpenNeuro ds004473 version 1.0.2. The code checks every gate boundary and open interval with rational arithmetic. Separate label-aware gains for each window gave a 98.75% maximum error bound, compared with 46.83% for the worst constant shaft and gain per person. This is an offline bound, not a real fault rate. Direct rational checks, an analysis replay, and fresh voltage ranges from all eight people passed. The package preserves code, data, source pins, proofs, witnesses, checks, and three figures. The results apply to an encoded affine gain model. They do not establish live intent, stimulation, or patient benefit.

stereoelectroencephalography · selective classification · confidence gates · voltage gain · robustness audit · brain-computer interfaces