A delay search changed mean AUC from 0.9348 to 0.9363 under simulated independent discrete uniform ±9.77 ms timing errors in eight SEEG sessions

We tested a template delay search for response monitoring with public SEEG voltage from eight patients and 86 stimulation sites. Five early pulses fixed each readout pair, template, and threshold. Five later pulses supplied the test data. We imposed integer timing errors from -20 to +20 samples at 2048 Hz. The primary AUC used independent discrete uniform errors for the compared after and before windows. Mean AUC was 0.9348 for a fixed template and 0.9363 for a search over 41 delays. The paired difference was 0.00155. Its 95% patient bootstrap interval was -0.00101 to +0.00445; the patient sign-flip test gave p = 0.390625. The test did not establish a positive mean effect or equivalence. Search background calls rose from 14.1% to 17.5% when the search used the fixed-score threshold. A threshold fitted with the search gave 14.0% background calls. The errors were simulated. No natural clock error or live BCI benefit was measured. The package preserves voltage inputs, all scores, code, source hashes, verification records, and three figures. This study reused a clinical cohort and prior readout choices. Fresh checks passed for all eight complete source voltage files.

stereoelectroencephalography · SEEG · evoked response · template matching · timestamp error · area under the curve · AUC · brain-computer interface