Uniform selection of 20 windows without replacement changed the expected mean C3/C4 beta contrast sign in six ds002778 pairs under CAR and fixed power weights

We tested uniform window selection without replacement in public Parkinson resting EEG from ds002778 version 1.0.5. The main rules selected six medication ON/OFF pairs. The score used C3/C4, 13 through 30 Hz, and average reference. We derived exact selection means and variances with fixed full-pool power weights. The formulas do not require independent EEG windows. At 20 selected windows, the mean session score increased by 70.7 percent. The expected mean ON-minus-OFF contrast changed from -0.000159 to 0.001382. Three of six expected person contrasts changed sign. Recalculation of subset power also gave a positive simulated mean contrast at 20 windows. At five windows, it gave a negative mean where the fixed-power result was positive. The fixed-power formula thus cannot replace the changing power ratio. A distinct-window product score had an expectation that did not depend on the selected count. This property does not prove population unbiasedness. The tool passed 656 exhaustive checks. Fresh extraction reproduced all 30 source derivatives. A full numerical repeat matched all 46 output files byte for byte. The package includes code, proof, derivatives, subset assignments, source hashes, and repeat instructions. These results describe window selection in fixed records. They do not establish a medication effect or neural coupling.

EEG · window selection · finite population · imaginary coherency · Parkinson disease · uniform sampling · power normalization