The 95 percent interval for the resection AUC gain from splitting 300 seconds of overlapping HFO marks across two sleep records included zero in seven SEEG patients

We tested whether 300 seconds split across two sleep records improved the ranking of released resection labels compared with one complete record. We used 23,528 released ripple and fast-ripple overlap marks from 135 records in seven depth-only SEEG patients in OpenNeuro ds003498 version 1.1.1. The mean AUC was 0.52981 for one record and 0.53937 for two records. The mean gain was 0.00955. Its patient bootstrap 95 percent interval was [-0.00095, 0.02387]. The exact sign-flip p value was 0.234375 under independent patients and a symmetric zero-effect null. Prior list agreement increased from 0.70591 to 0.72294 and was reproduced for each patient. The fixed top-one-pair precision check gave a larger positive estimate. All seven descriptive checks are preserved. The primary test did not establish a positive mean AUC gain or equal performance. Resection labels are treatment labels. The test used existing detector marks and did not rerun detection. The package preserves source bytes, source identity, code, every allocation score, patient draws, an independent validation, and an offline repeat. A proof gives equal expected scores under independent constant-rate Poisson counts and fixed labels.

stereo-EEG · epilepsy · high-frequency oscillations · resection labels · AUC · sampling duration · record agreement