Mean STE HFO AUC exceeded a linear SOZ run reference in 18 SEEG sleep records under uniform allowed labels

We tested first-hour released STE HFO counts in 18 SEEG sleep records from OpenNeuro ds005398 version 1.1.3. The label reference fixed contact gaps, SOZ counts, and positive run counts within each observed segment. It selected allowed labels uniformly and drew segments independently. The mean observed AUC was 0.719. The exact mean reference AUC was 0.648. The mean excess was 0.071, with a descriptive patient bootstrap interval of [0.012,0.143] and a sampled upper-tail p of 0.00089. A fixed check also kept both segment-end labels. Its mean excess was 0.005, and only five patients had variable reference scores. An exact calibration audit found that count-only tests rejected more than 5% of run-model labels in seven patients at a nominal 5% threshold. The largest fraction was one third. These are finite model results, not clinical error estimates. The package contains exact test code, compressed source tables, pinned retrieval instructions, results, saved draws, and figures. Independent checks and a full numerical repeat passed. This is a secondary analysis of a known cohort. It does not establish anatomical control, new detector validity, or surgery benefit.

high-frequency oscillations · stereo-EEG · seizure onset zone · epilepsy · exact permutation test · label runs · calibration · AUC · OpenNeuro