Later response windows gave lower mean Brier error at both fixed rest positions in eight SEEG recordings with three DPSS tapers

We tested whether moving a 250-sample response window closer to a recorded key press changed reading error when rest position stayed fixed. We used public OpenNeuro ds004473 version 1.0.2 data from eight people. The readers used eight bipolar pairs, 13–30 Hz and 70–110 Hz power, three DPSS tapers, early training trials, and 674 later test trials. We crossed first and last response halves with first and last rest halves. Mean Brier error fell by 0.029637 with the first rest half fixed and by 0.038308 with the last rest half fixed. The primary response change averaged across rest positions was −0.033973. Its descriptive 95 percent person bootstrap interval was −0.061771 to −0.008354. The exact two-sided sign-flip reference was p = 0.05469. Error fell in six people and rose in two. Fixed Hann and single-band checks kept the mean direction. The joint-position mean change reproduced the parent result. A post hoc timing audit showed that cue position also changed. The evidence concerns offline task state in a reused clinical cohort. It cannot isolate motor preparation or establish live reading delay or stimulation benefit. The package preserves exact digital inputs, source versions and range hashes, every model and prediction, code, four figures, and repeat instructions. Independent checks covered 192 fits and 32 feature arrays. Fresh retrieval matched selected samples in 16 source ranges. An exact replay matched 28 output files and 1,428 arrays.

stereoelectroencephalography · brain-computer interfaces · window position · Brier error · DPSS · logistic regression · OpenNeuro