Stored headband labels had a lower mean wake-sleep change rate than human labels on common valid edges in BOAS 1.1.3

We tested stored sleep labels from BOAS ds005555 version 1.1.3. The sample had 128 recordings from 100 people. The main analysis used 114,040 common valid adjacent edges. People had equal weight, with counts pooled across each person's recordings. Headband AI had a mean wake-sleep change rate of 3.962 per eligible hour. Humans had a mean rate of 4.533. The paired difference was -0.571, with a 95% bootstrap interval from -0.930 to -0.223. The interval assumes independent people and fixed stored model outputs. Binary epoch agreement was 95.85%. A two-minute complete Wake bout rule gave a higher headband rate. Missing AI labels allowed mean differences from -0.787 to +3.589 on all human-valid edges. Thus the main result applies to the retained grid. The package includes source labels, code, ledgers, results, and all bootstrap draws. It supplies an exact bound tool with a proof and 29,524 exhaustive sequence tests. Fresh source downloads and full analysis repeats matched the published files.

sleep staging · wake-sleep transitions · headband EEG · polysomnography · BOAS · OpenNeuro · label agreement · temporal agreement · missing-label bounds · bootstrap · complete Wake bouts