Three tapers increased mean SEEG gate score at error cost 4 in eight reused recordings, but the change interval included zero

We tested a fixed 80% confidence gate in public SEEG data from eight people. We scored a correct output as +1, a wrong output as -4, and a withheld output as zero. Three DPSS tapers gave mean score 0.136429 per available test window. One Hann taper gave 0.068157. The paired change was +0.068272. Its 95% person bootstrap interval was -0.002205 to +0.129043. The interval included zero. Output retention rose from 49.93% to 56.83%. Mean wrong outputs per available window stayed near 8.64%. Observed error among accepted outputs was 20.19% for Hann and 18.12% for three DPSS. The probabilities predicted only 9.34% and 7.87%. Each method had negative score in four people. An offline control that retained 50% of outputs favored three tapers. The test used 674 later trials with two task-state windows per trial. The trials came from OpenNeuro ds004473 version 1.0.2 and had been used in earlier work. Response markers defined the windows, and an inherited quality rule used post-response samples. This study does not establish a live error bound or a brain stimulation benefit. The package preserves exact inputs, source records, source retrieval code, every output, figures, and reproduction checks.

stereoelectroencephalography · brain-computer interface · selective classification · confidence gate · DPSS · Hann taper · calibration · OpenNeuro ds004473