A scaled maximum test controlled global null error under independent uniform contact phases in beta models from 16 SEEG recordings

We tested graph error in beta phase models from 16 public SEEG recordings. The source was OpenNeuro ds004100 version 1.1.3. The models used fixed Fourier magnitudes and independent uniform phases for each contact, window, and bin. The same contact had the same phase in all bipolar outputs. At a 5% threshold, separate edge tests rejected an edge in 95.489% of null graphs. Raw and scaled maximum tests gave 5.093% and 5.168%. A rank proof gives a 5% global null error bound for both maximum rules. The scaled rule used means and standard deviations from a separate pilot sample. The synthetic signal copied two contact phases with probability 0.25. Raw and scaled tests detected the selected disjoint edge in 0.000% and 96.263% of simulations. All 16 source rebuilds and the full numerical repeat passed. The package contains code, data, all scores, a proof, source records, and repeat commands. The results apply to the stated model. They do not establish neural connections in recorded signals. The proof does not give error control under partial alternatives.

SEEG · pairwise phase consistency · maximum statistic · global null error · Monte Carlo rank test · contact phase model · bipolar recordings · synthetic signal