A sequential pulse quota gave no clear error gain over random allocation in an offline test of 20 to 500 ms SEEG from eight sessions

We tested a sequential pulse quota on public SEEG from OpenNeuro ds004696 version 1.0.1. Eight epilepsy sessions supplied 86 selected readouts. The rule chose five or seven calibration pulses at each site. It used only the current site's first five waveforms and a fixed score distribution from the other seven patients. The retained input budget was 514 pulses. Both the rule and random allocation forecast the same 258 later responses. For centered 20 to 500 ms voltage, the equal patient mean error reduction was 0.98% relative to exact expected random error. The descriptive 95% patient interval was -0.19% to 2.05%. Six patient effects were positive. The sign-flip reference was 0.1484. The rule showed no clear gain over random allocation or about six pulses per site at the same budget. A dynamic program gives the best expected sum of allocation scores if future scores are independent draws from a fixed donor distribution. This proof does not guarantee lower response error. The timing audit permitted a site-by-site decision replay. The site list and good-pulse labels were fixed offline. The actual experiment delivered the pulses ignored by each forecast. Thus the study does not measure the effect of delivering fewer pulses or a live BCI benefit. The package preserves a stateful decision tool, proof, exact voltage windows, donor models, all decisions and errors, three figures, provenance, and reproduction code. Two full source rebuilds matched the retained arrays and pulse tables exactly. Exact numeric replay passed. This is an exploratory analysis of a reused clinical cohort.

stereoelectroencephalography · SEEG · electrical stimulation · pulse allocation · brain-computer interface · calibration · forecast error · sequential decision rule