Correlation magnitudes require balance in a mean 66.8 percent of 1–80 Hz SEEG triangles from the strongest 10 percent of links between named shafts after average reference in 35 patients

We tested whether measured correlation magnitudes require a balanced sign product in strong SEEG triangles. We used public OpenNeuro ds004100 version 1.1.3 and preserved covariances from 36 epilepsy patients. The primary graph used 1–80 Hz, equal-weight average reference, and the strongest 10 percent of links between different named shafts. Each triangle contained three shafts. There were 418,198 triangles in 140 blocks from 35 patients. One patient had no selected triangle. At fixed magnitudes, an unbalanced product was impossible in a mean 66.8 percent of triangles (patient bootstrap 95 percent interval 59.7 to 73.8 percent). A uniform shuffle of selected signs was expected to give an invalid three-channel correlation matrix in 31.1 percent of triangles. All 7,000 shuffles in the 35 tested blocks with triangles gave at least one invalid triple. The fraction with required balance changed with reference, band, and edge density. This fraction is a mathematical constraint. It does not measure the share of a signal caused by an artifact or by neural activity. The package includes an audit tool, all input covariances, source metadata, pinned raw retrieval recipes, results, bootstrap draws, and checks. Two fixed raw excerpts reproduced 12 covariance matrices. The complete offline repeat gave identical result files.

SEEG · correlation matrices · signed networks · covariance constraints · sign shuffle · epilepsy