Mean pulse-specific SEEG read windows were 17.1 ms shorter and gave three fewer correct decisions than fixed windows in five reused offline eight-site tests

We tested whether a SEEG rule could reduce the read time needed to identify one of eight known stimulation sites. We used OpenNeuro ds004457 version 1.0.2 and five reused offline sessions. Calibration used the first four valid pulses per candidate site. Test used the last four pulses at eight calibration-selected sites per patient. The rule used voltage from 20 ms after each pulse and a margin between its two closest templates. Thresholds and fixed comparison cutoffs were selected from calibration data. On the same 160 test pulses, the mean cutoff was 42.875 ms for the new rule and 60 ms for fixed cutoffs. Correct counts were 142 and 145. The 17.125 ms reduction had an observed accuracy cost. A 30 ms start removed the mean time saving. The split was ordered within sites and overlapped global calibration time. These results do not establish live interface performance or a purely neural signal. The package contains four figures, voltage bins, fitted models, event decisions, source hashes, and reproduction code. Independent decision checks, full numerical replay, and a raw-source rebuild passed.

stereoelectroencephalography · electrical stimulation · brain-computer interface · adaptive readout · decision margin · offline evaluation