Contact name group removal from the average reference gave greater mean high gamma correlation edge turnover than matched random removal in 19 SEEG records

We tested whether reference contact groups change strong correlation edges more than random contact sets of the same size. We used two interictal minutes from each of 19 public SEEG records in OpenNeuro ds003876, version 1.0.2. The primary measure used the strongest 10 percent of absolute cross-group correlations from 80 to less than 150 Hz. Bins from 118 through 122 Hz were excluded. Mean turnover was 9.81 percent for group removal and 5.41 percent for random removal. The mean difference was 4.41 percentage points, with a participant bootstrap 95 percent interval of 3.39 to 5.50 points. All 19 participant averages were positive. Of 217 individual group averages, 31 were negative. Both conditions used the same evaluation edges. Contact name groups were unverified shaft proxies. This software reference effect does not show a neural cause or clinical accuracy. The package includes code, covariance data, pinned source recipes, exact removal sets, results, and validation. All 19 covariance files matched fresh raw-data rebuilds. A repeated analysis gave identical hashes for all five numerical output files.

intracranial EEG · SEEG · common average reference · high gamma · correlation networks · edge turnover · reference sensitivity · electrode sampling · reproducibility