Cross-pulse energy changed 10–90 ms response label AUC by less than 0.007 for the first five pulses in two SEEG recordings

We tested whether cross-pulse energy improves expert response-label ranking with the first five pulses. We used 630 public pulses, six SEEG readout contacts per recording, and 566 retained expert labels from OpenNeuro ds004774 v1.0.0. In the 10–90 ms window, mean-energy AUCs were 0.8880, 0.7385, and 0.8918. Cross-energy AUCs were 0.8910, 0.7411, and 0.8982. Paired 95% site bootstrap bounds included zero in all three expert groups. With two pulses, the correction reduced the UMCU25 point AUC by 0.0500; its paired bounds also included zero. The unscaled score removes independent trial-noise bias under a common mean and zero cross-pulse noise covariance. Shared noise and fixed artifact remain. The proof is for unscaled energy; the measured baseline divisor is random. This is retrospective label ranking in two recordings, not a live interface test. The package preserves exact epochs, source receipts, all scores and bootstrap values, figures, proof, code, and reproduction steps. All 630 fresh source ranges passed. Independent score and AUC checks passed, and a full numerical repeat was exact.

SEEG · evoked responses · cross-pulse energy · signal processing · response labels · AUC · noise correction