Equal condition weights gave higher 300–600 ms Pz peak contrasts than trial count weights for 1:3 random labels in 12 adults

This study tested peak selection on fixed EEG from OpenNeuro ds003061 version 1.1.2. Rules retained 5,288 trials from 34 runs in 12 adults. We assigned random labels at a 1:3 group count ratio in 10,000 experiments. The contrast was mean A minus mean B. Both peak rules searched 300–600 ms after each sound. The equal condition mean selected a higher contrast than the mean across all trials. The paired mean increase was 0.9283 µV. Its 95% Monte Carlo interval was 0.9224–0.9343 µV. The two-sided t test rejected the null in 26.05% of experiments with equal condition weights. The trial count method rejected the null in 4.93%. The two peak methods agreed exactly at balanced counts. A finite-data proof gave a nonnegative paired gap for each partition when group A was smaller than group B. Trial-count peak selection had an exact zero expected contrast under uniform labels with fixed group counts. A separate rank calibration gave 5.88% validation reports for equal condition weights. This rate conditions on one calibration set. This study checks an established method problem in real EEG. It preserves code, trial waves, label assignments, all scores, source hashes, and a retrieval recipe. Fresh checks reproduced all 5,546 accepted trial waves from all 39 raw runs. A separate calculation checked 340,000 primary partitions. The intervals measure simulation error on this fixed EEG sample. The results do not prove general error control or a biological effect.

trial-count weighting · equal condition weighting · selection bias · randomization test · electroencephalography · peak selection · finite-data proof