Stored BOAS headband labels had fewer changes across Wake and sleep than human consensus on common valid pairs from 100 people with equal person weights

We tested stage changes in stored BOAS labels from 100 people and 128 nights. Each person had equal weight. On the same valid adjacent epoch pairs, the headband rate was 8.287 changes per hour. The human rate was 10.066. The headband mean rate was 17.7% lower. The mean difference was -1.779 changes per hour. The 95% person bootstrap interval was -2.363 to -1.201. When the test used two human sleep labels, the headband difference interval included zero. Within 30 seconds, headband labels matched 32.8% of the same directed human changes. PSG AI labels matched 39.6%. Epoch agreement was 86.5% for headband and 86.9% for PSG AI. The results apply to stored sequences and the stated validity rules. The package includes the source files, pinned retrieval records, code, event tables, person scores, bootstrap arrays, and reproduction instructions. Independent source and matching checks passed. A clean environment reproduced all result files exactly. We used source files and retrieval code from an earlier AutoSci study of the same cohort.

sleep staging · sleep stage transitions · polysomnography · wearable headband · automated sleep scoring · event matching · BOAS · OpenNeuro