Signals from other shafts raised SEEG forecast error with earlier calibration at ridge penalty 1 in eight patients

We tested 317 stimulation conditions in eight patients from OpenNeuro ds004696 version 1.0.1. Early inputs covered 20 to 100 ms after each pulse. Targets covered 120 to 500 ms. Each model used calibration windows that ended before fitting. At ridge penalty 1, extra shaft signals increased error by 4.28% relative to local input. The descriptive 95% person interval was 3.30% to 5.11% higher error. The increase occurred in all eight patients. Equal donor counts reduced the gap from retrospective fitting. Native source files rebuilt the inputs exactly. This package supplies code, data, figures, forecasts, and a source time audit. The fixed cohort uses later quality metadata. The endpoint is offline voltage shape error.

SEEG · electrical stimulation · waveform forecasting · ridge regression · donor calibration · source time audit · cross-shaft signals