A 25 ms search raised window AUC under uniform simulated ±25 ms errors at 86 sites in eight SEEG sessions

We tested a time search for reading late windows after a known stimulation pulse. We used public SEEG data from OpenNeuro ds004696 version 1.0.1. The test retained 86 fixed recording pairs in eight clinical sessions. At each site, five early pulses fitted a signed template. Five later pulses tested the reader. Target windows had strict 125 to 375 ms bounds. The main test enumerated uniform simulated clock errors from -24.90 to +24.90 ms. Search used the same bounds for pulse and background windows. AUC is the area under the receiver operating characteristic curve. The equal-patient mean AUC rose from 0.8498 to 0.8733. The mean paired gain was 0.0235, with a descriptive 95% bootstrap interval of 0.0103 to 0.0361. Seven of eight patient means increased. Energy AUC was 0.8726. A post hoc paired comparison did not show a clear search advantage over energy. At exact timing, the same search increased mean AUC by 0.0183. Thus the full gain cannot be assigned to clock error recovery. Searching pulse windows only gave diagnostic AUC 0.9101 because background processing differed. The package includes selected data, exact scores, four figures, source audits, exact result files, and source records. This is an exploratory offline window test. Real clock faults, continuous commands, and stimulation benefit were not tested.

stereoelectroencephalography · SEEG · brain-computer interface · electrical stimulation · time search · clock error · AUC · signal energy