Five-pulse consistency selection showed no clear later SEEG forecast gain over energy selection at 20 to 500 ms in seven offline sessions

Five early good pulses selected a stimulation site by response consistency or mean response energy in public SEEG data. Five final pulses tested relative waveform forecasts at 20 to 500 ms. Seven offline sessions gave 21 candidate groups and 217 target choices. Consistency selection changed 64 choices. Its mean skill gain was -0.32 percentage points, with a descriptive 95% person bootstrap interval of -10.09 to +9.24 points. The geometric later mean-waveform power ratio was 0.807 relative to energy selection. A fixed half-energy constraint gave an uncertain gain of +1.32 points and a power ratio of 0.984. The study proves the exact link between the consistency score and relative leave-one-out error and tests the response-size tradeoff. Candidate timing, retrospective labels, and reuse of the source cohort limit the result to an exploratory offline comparison. The result does not establish equal performance or live brain control. The retained files preserve exact pulse products, provenance, all scores, three figures, recorded checks, and a public retrieval description. GPL analysis and extraction scripts are excluded, so executable reproduction is unavailable from this package.

stereoelectroencephalography · electrical stimulation · waveform forecasting · response consistency · forecast skill · brain-computer interface