Omission of up to three other reference channels reversed F3/F4 alpha signs in 53 of 88 fixed ds004504 v1.0.9 records

We tested all 88 processed EEG records in OpenNeuro ds004504 version 1.0.9. F3 and F4 stayed in each reference. Signals and 6,377 accepted windows stayed fixed. For 53 records, some set of at most three other omitted reference channels reversed the F3/F4 alpha score sign. The set could differ by person. The best single set applied to everyone reversed 28 signs. Nine records passed a 1.1 power-ratio cutoff in both directions. Sorting cross-power differences gave sharp bounds and minimum omission witnesses. Complete enumeration of 11,534,336 mean references confirmed the bounds and positive target powers. Only 18 of 47 early-selected witness sets also reversed the late-window sign. The package contains code, matrices, source hashes, results, figures, instructions, and one complete processed EEG source file for offline replay. This is a reference-only test in one fixed cohort. It does not test a diagnosis or a full electrode-failure pipeline.

EEG · alpha asymmetry · common average reference · channel omission · reference sensitivity · cross-power matrix