A variance stop rule used 46% fewer pulses and raised normalized 20–500 ms centered bipolar SEEG error by 4.5% versus five-pulse scaling in eight sessions

We tested whether early response variance can select the calibration pulse count at a stimulation site. The test used all 886 eligible stimulation-to-readout pairs at 86 sites in eight patients from OpenNeuro ds004696 version 1.0.1. A fixed rule chose two to four early pulses and sought at most 5% extra model error relative to five-pulse scaling. The final five good pulses formed the later test pool. With equal patient weight, pulse use fell by 46.4% and normalized later individual-pulse error rose by 4.51%. The descriptive 95% patient bootstrap interval was 3.10% to 5.96%. Later error exceeded 5% at 39 sites. The rule showed no clear gain over random count allocation with the same pulse budget. The error change from that control was +0.0019%, with an interval of -0.485% to +0.500%. The result applies to centered 20-500 ms bipolar voltage in one clinical session per patient. It does not establish a valid online error limit. The package preserves compact data, source provenance, code, all scores, three figures, and reproduction instructions. Fresh source rebuilds for sub-01 and sub-08 matched all archived arrays exactly.

stereoelectroencephalography · electrical stimulation · response prediction · calibration · trial noise · pulse allocation · signal processing