Selected removal of at most two of the first 20 accepted ten-second windows reversed 10 Hz imaginary coherency beyond a 0.05 margin in 32 of 88 ds004504 v1.0.9 records under a fixed mean19 reference

We tested window removal in all 88 processed EEG records from OpenNeuro ds004504 v1.0.9. Selected removal reversed 69 channel-pair estimates in 32 records. Each record used one shared set of at most two of its first 20 accepted ten-second windows. Each score used a fixed mean19 reference and the 10 Hz bin. Both scores had an absolute value of at least 0.05. Uniform removal of exactly two windows gave an exact expected total of 2.042 reversals. Pair-specific removal bounds allowed 104 reversals. With all accepted windows, shared removal reversed five estimates in three records. A reusable tool gives attained bounds and minimum strict-sign removal counts. We rebuilt the Fourier input from all 88 complete public files and checked each source hash. We checked every primary score and repeated all result files byte for byte. These results apply to fixed spectra, a stated reference, and selected window removal. They do not give population rates or causal brain direction. The package preserves code, data, provenance, complete results, proofs, figures, and reproduction instructions.

imaginary coherency · EEG · window removal · sensitivity analysis · signal processing