Mean voltage from four pulses at the same site gave 43.8% accuracy and label votes gave 40.1% at 20 to 100 ms in five offline SEEG sessions

We tested whether mean voltage from four later pulses identified the stimulation site more accurately than a vote from four single-pulse labels. We used public OpenNeuro ds004457 version 1.0.2, selected ventral temporal SEEG pairs, and 20 to 100 ms features. Five patients supplied 138 sites, 552 calibration pulses, and 552 test pulses. Each reader used only the first four valid pulses per site to fit a fixed distance rule. Each test group had four later pulses at one site. Mean patient accuracy was 43.78% for mean voltage and 40.14% for votes with equal probability in a tie. The paired difference was 3.64 percentage points. Four patients had a positive difference; one had no difference. The descriptive patient resampling interval was 1.38 to 5.90 points. The exact two-sided sign reference gave p = 0.125. These five sessions do not establish a population benefit. Tie groups supplied the observed net gain. Mean accuracy from one pulse was 33.71%, but four pulses required more measurements and time. The median group span was 14.34 seconds; one group spanned 2355.2 seconds. Source time alone gave 82.35% accuracy. Only 32 test pulses occurred after all calibration pulses. The result is an offline benchmark, with source schedule and artifact limits. The package includes a score accumulator that exactly implements voltage averaging for the fixed affine reader, three figures, all decisions, fitted models, code, derived data, and a pinned raw retrieval recipe. Fresh raw input reconstruction and an independent replay passed. This work does not test live brain control or a stimulation policy.

stereoelectroencephalography · electrical brain stimulation · response aggregation · voltage averaging · label voting · site identification · brain-computer interface · intracranial electrophysiology