Late response selection gave a mean SEEG repeatability gain of 0.024 over early selection when half the available pairs were kept in eight sessions

We tested whether early SEEG responses select repeatable pairs for a later response to brain stimulation. The test used public OpenNeuro ds004696 version 1.0.1 and separate first-five and final-five pulses. The sample had 884 available bipolar pair instances at 84 sites in eight sessions. Both rules kept half the pairs at each site, rounded up. Selection at 100–500 ms gave a mean final-pulse late-window correlation of 0.489. Selection at 20–100 ms gave 0.464. The mean change was 0.024, with a 95% person-resampling interval of -0.007 to 0.062. Four people had a gain and four had a loss. The exact two-sided sign-flip probability was 0.23438. The evidence does not show a clear advantage across people or establish equivalence. The rules shared 65.6% of selected pairs on average. Before-pulse scores were close to zero. Keeping one pair gave a larger secondary mean change. The study measures waveform repeatability, not BCI decoding or clinical benefit. It reuses exact public trial products and adds no recordings. The package preserves a five-pulse selection tool, data provenance, a local plan, numeric results, figures, and replay instructions. A fresh full source check covered one person. Independent checks reproduced all split correlations, forward site outcomes, and primary statistics.

SEEG · brain-computer interface · stimulation · waveform repeatability · response-window selection