Early error weights gave three more correct sound selections in 38 SEEG passages with template shifts up to two seconds

We tested early error weights for sound selection from public SEEG data. The source was OpenNeuro ds004703 version 1.1.0. Ten fixed readers used eight SEEG pairs each. We kept 57 early fit passages and 38 later test passages. Each query had six seconds of readout. The query time was supplied. Candidate templates could move from -2 to +2 seconds. Early passage errors fitted one time correlation value per person. Mean accuracy, with equal person weight, increased from 12.5% to 20%. Three file decisions improved. No correct file decision was lost. The descriptive 95% person resampling range for the gain was 0 to 15 percentage points. The sign-flip reference was p = 0.25. With supplied template timing, accuracy decreased from 47.5% to 37.5%. Effects also changed across query positions. One added correct file used a template 1.6 seconds from the supplied time. These results do not establish a population gain or a live interface benefit. All 95 selected SEEG ranges and 39 sound ranges were freshly checked. A separate score calculation and a clean replay passed. The package preserves input arrays, source records, code, models, scores, three figures, and reproduction instructions.

SEEG · brain-computer interface · temporal correlation · sound selection · error weighting · exploratory study