Higher thresholds increased unmatched command counts at some steps in eight fixed SEEG replays with a 1 s wait

We tested whether higher forecast thresholds always reduce commands with no recorded keypress in the next 100 ms. We used fixed task plus SEEG forecasts from eight people in public OpenNeuro ds004473 version 1.0.2. Each emitted command started a 1 s wait. All eight test streams had an unmatched count increase at some steps: 286 of 26,879 steps. Total command counts never increased. A two-candidate proof explains the result. All main increases occurred at low thresholds and above 9.89 unmatched commands per minute. At an early limit of one unmatched command per minute, first feasible and tail feasible rules chose identical thresholds. The later equal-person mean was 1.29 unmatched commands per minute; five of eight people exceeded the limit. Thus the stricter rule gave no benefit at the tested limit. The package contains an exact audit tool, full curves, three figures, source hashes, raw retrieval recipes, and repeat instructions. An independent scheduler checked all 402,360 curve thresholds. Fresh raw ranges matched for all eight people. The result applies to fixed recorded streams. It does not show intent, live BCI benefit, or stimulation effects.

brain-computer interfaces · SEEG · threshold scheduling · unmatched commands · fixed-stream replay · calibration