Complete human sleep paths from Wake to Wake contain more N2-to-REM than REM-to-N2 changes in 100 BOAS people

We tested time direction in BOAS sleep labels from 100 people and 128 recordings. We kept complete human paths that leave Wake and return to Wake. We removed invalid labels and incomplete paths. The paths contain 548 N2-to-REM changes and 274 reverse changes. The mean person score is 2.881 net changes per 100 stage changes. The 95% person bootstrap interval is 2.471 to 3.332. Planned checks give positive mean scores. A chosen-edge whole-person reversal test gives a Monte Carlo p value of 0.00001. This is its lowest reportable value. The reference model has fixed reversible first-order Markov transitions within each recording. It also needs independent people and a reversal-invariant observation rule. The local plan followed review of prior BOAS work. This exploratory result does not correct for question selection. A time-dependent model can give a positive score even when each time-specific matrix is reversible. The result therefore does not identify a neural cause. The package preserves source files, hashes, retrieval code, a proof, all primary draws, derived results, and model checks. A fresh download and full repeat match the saved output hashes.

sleep stages · Markov chains · reversibility · sleep paths · transition direction