On common valid intervals in 100 BOAS people, two-minute sampling retained a larger fraction of headband than human stage change counts

We tested how endpoint spacing changes counts in stored BOAS v1.1.3 sleep labels. We used 128 recordings from 100 people. Each selected start had 11 consecutive epochs valid in the human, headband AI, and PSG AI streams. At two-minute spacing, headband labels retained 53.31% of their internal stage change counts. Human labels retained 47.23%. The mean paired difference was 6.08 percentage points. Its 95% person bootstrap interval was 4.41 to 7.73. All planned checks kept a positive difference. On the same intervals, the human minus headband rate gap was 1.76 changes per hour for internal 30-second edges and 0.26 for endpoints. The package provides a general audit tool, source tables, hashes, a local plan, every interval, results, all bootstrap draws, and reproduction instructions. Independent checks and a fresh-source repeat passed. The retained count fraction is not recall of matched events. The result applies to fixed supplied labels and observed valid windows. It assumes independent people for its interval. It does not test physiological arousals or a model scored with two-minute epochs.

sleep staging · temporal resolution · stage change counts · endpoint sampling · headband EEG · polysomnography · bootstrap · BOAS