Separate pair tests called false links in 99.43 percent of 100-pair SEEG beta searches under an independent phase null

We tested whether a calibrated SEEG pair test controls a search of many pairs. We used two-second beta windows from 35 public recordings and 21 participants in OpenNeuro ds006519 version 1.0.0. Each graph searched 100 fixed pairs from different source groups. The controls held channel magnitude spectra fixed. Each channel had one random phase vector shared by all its links. Separate 5 percent pair rank tests called false links in 99.43 percent of complete null graphs. The descriptive participant bootstrap 95 percent interval was 99.40 to 99.46 percent. Bonferroni gave 4.77 percent graph error. A graph maximum rank test gave 4.88 percent, with an interval of 4.65 to 5.12 percent. The paired decrease was 94.55 percentage points. Pair error remained near 5 percent. The graph correction is an established method. The contribution is a checked SEEG benchmark and graph test tool. A fresh rebuild matched all 35 feature archives. A separate replay matched all 44 scientific output files. The package preserves data, source hashes, code, graph traces, decisions, figures, and repeat instructions. The cohort and features are reused. The result applies to the complete independent phase null and offline beta projection. It does not prove neural connection truth, power under partial alternatives, useful live BCI control, or stimulation benefit.

multiple testing · graph error rate · familywise error · phase randomization · SEEG · intracranial EEG · statistical calibration · Fourier spectrum · brain-computer interface