Stored headband labels had a lower mean transition rate than human labels across five sleep stages in 100 BOAS people on common valid edges

We tested stored sleep labels from OpenNeuro ds005555 version 1.1.3. The data had 128 nights from 100 people. The main rule used two adjacent 30 second epochs with valid labels in all three streams. The mean headband rate was 8.287 transitions per eligible hour. The human reference rate was 10.066. The paired difference was -1.779, with a 95% person bootstrap interval of -2.364 to -1.212. With N1, N2, and N3 combined into NREM, the difference interval included zero. The result therefore depends on the five stage definition. Headband epoch agreement was 86.47%, but directed transition recall within 30 seconds was 32.82%. An exact sequence control kept epoch confusion counts fixed and raised the expected headband rate by 18.069 transitions per eligible hour. The package includes unchanged source files, an audit command, matched times, all results, bootstrap samples, and reproduction instructions. Independent calculations passed. Full fresh source retrieval reproduced all numerical files byte for byte. The intervals assume independent people and fixed outputs. These are label transitions against a human reference, not measured EEG arousals or a test of new model predictions.

sleep staging · sleep stage transitions · polysomnography · wearable headband · label agreement · transition matching · BOAS · OpenNeuro