Using 100 instead of 500 samples for high gamma raised mean Brier error on later trials in eight SEEG readers with fixed pairs, Hann power, and 500 samples for lower bands

We tested ten fixed signal rules on public SEEG data from eight people. The primary reader kept 500 samples for 8 to 12 Hz and 13 to 30 Hz power. It used the latest 100 samples for 70 to 110 Hz power. The reference reader used 500 samples for all three bands. Each person had eight fixed bipolar pairs, Hann power estimates, and a separate logistic reader. The readers learned from early trials and predicted states in 674 later trials from the same recordings. Mean Brier error rose from 0.190139 to 0.212978. The equal-person mean rise was 0.022839, with a 95 percent person bootstrap interval of 0.011375 to 0.034046. Seven of eight recordings had higher error. The direction held at two additional weight penalties. This is an exploratory extension with prior use of the same cohort and test trials. The result does not measure live command delay or cue-independent intention. The package preserves exact voltage windows, source versions and hashes, code, features, models, predictions, three figures, and reproduction instructions. Independent numerical checks and an exact replay passed. Fresh public source ranges matched 128,000 saved voltage values.

stereoelectroencephalography · brain-computer interfaces · high gamma · signal history · Brier error · Hann periodogram · logistic regression · within-recording prediction