Baseline Occupancy Entropy in Pediatric Epilepsy: A Cross-Subject Signal, a Fragile Cross-Dataset Comparison, and the Cost of an Unverifiable Field Convention

Prior WCM-PL work on the CHB-MIT pediatric epilepsy EEG database anchored analyses on seizure onset,comparing a patient's pre-ictal window against their own background -- yielding weak,normalization-inconsistent results (Mann-Whitney rank p=0.040, robust p=0.465, n=22 seizures). This paperreframes the question: instead of asking whether a seizure differs from a patient's own baseline, we askwhether patients' baselines differ from each other, and from a healthy population. We find (1) a real,cross-normalization-agreeing, pseudo-replication-corrected difference in baseline occupancy entropy acrossCHB-MIT's 4 subjects (Kruskal-Wallis, file-level, rank p=1.9e-6, robust p=0.0024); (2) that this difference isconfounded with subject age in a way the dataset cannot resolve; (3) that a healthy-control comparison againstPhysioNet's eegmmidb reverses sign depending on which eegmmidb run (eyes-open vs eyes-closed) is usedas the comparator; and (4) that this fragility traces to an undocumented, unverifiable field convention --CHB-MIT records no eyes-open/closed state, and prior published work (independently of this project) simplyassumes eyes-open throughout. We report this chain of findings, including the convention-fragility result itself,as the paper's central methodological contribution. ※ The License for codes is AGPL 3.0-or- later