Pair score thresholds gave 44.6% mean false acceptance for one 20-person EEG list under uniform splits of ds004148 v1.0.1

We tested unknown-person rejection with closed-eye EEG spectra from 60 adults in OpenNeuro ds004148 version 1.0.1. The fixed similarity rule compared visits about one month apart. For one fixed 20-person enrollment list, we computed exact means over all 137,846,528,820 uniform calibration subsets of 20 remaining people. All-pair thresholds gave 44.61% mean false acceptance. One randomly selected pair per calibration person gave 35.47% after exact averaging over candidate selections. Calibration on each person's largest list score gave 4.76%, the finite-cohort rank fraction 1/21. Across 10,000 allocations that also changed the list, the respective means were 46.82%, 36.91%, and 4.73%. The largest-score rule accepted 14.86% of genuine queries correctly and rejected 81.48%. The standard future-query bound requires exchangeable calibration and unknown person records conditional on a fixed list. It is marginal, not conditional on one calibration set. The package preserves exact count tools, spectra, provenance, raw retrieval recipes, decisions, proofs, and figures. Two complete raw replays and an offline repeat passed. The result concerns this cohort and score rule.

electroencephalography · unknown-person rejection · score calibration · false acceptance · person exchangeability · closed-eye EEG · exact finite-cohort average · Pearson correlation