Fixed zero-sum alpha filters with the lowest session 1 ratio had a positive mean session 3 contrast in 60 ds004148 adults

We tested whether a low training alpha power ratio kept a negative mean in a later EEG record. We used 60 adults from OpenNeuro ds004148 version 1.0.1. Each person had eight posterior channel weights that summed to zero. Session 1 set the filter with the lowest closed-eye to open-eye 8 to 13 Hz power ratio. Session 3 tested the fixed weights about one month later. The selected training mean was -1.42 dB. The later mean was +3.84 dB. Its 95% whole-person bootstrap interval was +2.87 to +4.84 dB, from 20,000 draws. Twelve people had a negative later value. All planned sensitivity intervals were above zero. Complete-visit covariance had both positive and negative directions in 38 of 60 session 3 records. This fitted audit is separate from the fixed-weight later test. The package preserves all covariance inputs, code, source versions, masks, weights, bootstrap draws, figures, and checks. Two fresh raw crops replayed exactly. A full analysis repeat gave identical results. Fixed state order and different source processing limit causal neural claims. This study reuses the prior cohort. It is not an independent cohort replication.

EEG · alpha power · zero-sum spatial filters · generalized eigenvalue · test-retest · power ratio