A fixed posterior EEG rule gives 68.6% expected accuracy across one-month eyes-closed visits of 60 ds004148 v1.0.1 adults with balanced random labels fixed per person

A fixed rule matched 23 of 60 adults to their own posterior EEG record across eyes-closed visits about one month apart. The data are OpenNeuro ds004148 v1.0.1. The rule used relative log power density from 1 to 30 Hz. It had no fitted scale or parameter search. For balanced random binary labels held fixed per person, the match rate gives 68.6% exact expected accuracy. Removal of each test person's template gives 49.2%. The expected difference is 19.5 percentage points. All templates were available in the first test; each control had 59 templates. A test of random identity linkage gave p=0.0001 from 9,999 permutations. The audit used 9,999 balanced label assignments and 10,000 query bootstrap draws. Separate visits retained person information in this cohort. The study does not test disease labels or performance in new people. The package preserves all 360 spectra, source hashes and metadata, the local plan, source and analysis code, features, matches, all random draws, results, and retrieval instructions. All input hashes passed. Three complete raw recordings reproduced every saved array. Independent feature, distance, and proof checks passed. Two offline runs gave identical result files.

electroencephalography · EEG · person information · cross-visit validation · random labels · spectral analysis · test-retest · data leakage