Convex reference changes at total variation 1/19 reversed 359 of 11,138 tested mean-reference imaginary coherency signs at 10 Hz in ds004504 v1.0.9

We tested common reference weights in 88 processed EEG records from OpenNeuro ds004504 v1.0.9. All weights were nonnegative and summed to one. The baseline gave each of 19 channels equal weight. The primary test used the first 100 accepted seconds at 10 Hz. It included 11,138 pairs with absolute baseline imaginary coherency at least 0.05. A total variation budget of 1/19 allowed 359 signs to reverse (3.22%). This distance is half the sum of absolute weight changes. The 19 omission references detected 52 reversals (0.47%). The exact audit found 307 additional reversals. A different reference can attain each pair's bound. A mass transfer rule gives sharp bounds and the minimum distance to zero. Independent linear programs, direct Fourier checks, two fresh source replays, and a full offline repeat passed. The package includes all Fourier inputs, weight witnesses, code, proof, source hashes, and instructions. These results describe fixed estimated spectra. They do not show neural direction or disease effects.

EEG · imaginary coherency · common reference · reference weights · total variation · sharp bounds · sign reversal · spectral analysis