Added SEEG power gave an uncertain mean AUC gain over contact number in 17 ds004100 test patients under fixed models, independence, and null symmetry

We tested whether fixed power improved seizure onset label ranking after a model used contact number. The public source was OpenNeuro ds004100, version 1.1.3. We fitted models on 17 patients and tested them on 17 separate patients. The added feature used local Laplacian power from 80 to below 115 Hz in two short records. Mean AUC increased from 0.687 to 0.742. The mean paired gain was 0.055. Its 95 percent patient bootstrap interval was -0.039 to +0.144. The exact sign-flip test gave p = 0.268 under independent patients and symmetric differences under the zero-effect null. Inference held the fitted models fixed. The mean gain remained uncertain. Thirteen patients gained AUC, while four lost AUC. Contact number was a name-based position surrogate, not measured anatomical depth. Prior results from this cohort informed the feature and comparator. This was not external validation. All 68 pinned public signal blocks passed a fresh power check. A separate reader and spectrum calculation checked all four sample rates. The package preserves code, source metadata, compact powers, retrieval locks, features, models, predictions, inference draws, and reproduction commands.

stereoelectroencephalography · SEEG · intracranial EEG · high gamma power · seizure onset · area under the curve · patient-level inference · reproducibility