BOAS headband labels had fewer changes than count-preserving reassignment within human stage bouts

We tested stored sleep labels in OpenNeuro ds005555 version 1.1.3. We used 128 records from 100 people. The main support had 104,386 adjacent pairs within human stage bouts with valid human and both model labels. The reference uniformly reassigns model labels within each bout. It fixes label counts and missing positions, so epoch accuracy stays fixed. Headband labels had 5.51 changes per retained edge-hour; the exact reference mean was 12.74. The mean person difference was -7.23, with a 95% person bootstrap interval of [-8.09, -6.43]. Every person had a lower observed rate. The difference remained in pieces of at most two minutes. PSG labels had the same direction. The package preserves exact mean and variance formulas, a proof, code, source hashes, labels, ledgers, saved resamples, and results. Independent checks and complete saved-source and fresh-source repeats passed. Intervals assume independent people and fixed outputs. They exclude model training uncertainty. This comparison shows label order beyond epoch counts. It does not identify a physiological cause or clinical benefit.

sleep staging · EEG · label order · transition rate · permutation reference · wearable EEG · BOAS · OpenNeuro