Best constant shifts gave 38.06% mean headband same-pair recall in 100 BOAS people within 60 seconds on common valid edges

We tested fixed sleep labels from BOAS, OpenNeuro ds005555 version 1.1.3. We used 128 recordings from 100 people. We asked how much one constant time shift could improve human change recall. Events came from common valid edges among Wake, N1, N2, N3, and REM. A match required the same ordered stage pair within 60 seconds. Each event could be used once. We proved that an integer-epoch search gives the maximum over all real shifts for these grid events and tolerances. Headband mean recall rose from 36.86% without a shift to 38.06% with the best shift per recording. The paired gain was 1.20 percentage points, with a conditional 95% person bootstrap interval of 0.87 to 1.58. The minimum mean unmatched fraction was 61.94%. Each person had equal weight. PSG mean recall rose from 44.07% to 45.27%. The bound uses fixed event sets and liberal matching rules. It does not measure clock errors or new-person accuracy. Independent dynamic programs checked every saved shift score. A fresh-download repeat gave identical bytes for all 14 numerical files. The package preserves source files, provenance, proof, code, match witnesses, all shift scores, bootstrap samples, and repeat commands.

sleep staging · stage transitions · EEG · headband EEG · recall · constant time shift · BOAS · event matching