Global phase shuffle gave 35.55% rejection under channel isolation after average reference in 88 EEG magnitude models at 10 Hz with shared background phases

We tested two phase shuffle methods after common average reference. We used magnitudes from 88 public EEG records at 10 Hz. In 7,040 generated records, target input phases were independent and uniform. The 17 background channels shared a phase in half the segments. A shuffle of all channels rejected 35.55% of records at a stated 5% level. A shuffle of one target channel rejected 5.23% (95% Monte Carlo interval 4.71% to 5.75%). The paired increase was 30.33 percentage points (29.35 to 31.30). The full shuffle tests a complete independent phase null, which this background violates. It cannot supply a 5% test of channel isolation in this model. The one-channel method leaves other input data fixed and repeats the reference. A conditional rank proof gives its 5% error bound. Target phases must be independent and uniform given all magnitudes and all other input data. Seven planned checks test background dependence, frequency, identity reference, and one target dependence alternative. All numerical outputs matched in two full runs. A fresh source replay matched all selected coefficients. The package contains code, fixed inputs, source hashes, saved copy banks, proof, figures, and retrieval instructions. These model results do not establish neural connections.

phase randomization · conditional randomization test · channel isolation · common average reference · EEG · squared coherence · Monte Carlo · conditional rank test