Fixed graph projection showed 17.9% lower mean five-stage flow magnitude for headband than human labels on common BOAS v1.1.3 edges

We tested stored sleep stage labels from 100 people and 128 BOAS records. We counted adjacent 30-second epochs with valid human, headband, and PSG labels. We used a fixed five-state complete graph. We removed the graph-gradient part of the directed count flow. Mean flow magnitude was 2.346 for human labels and 1.927 for headband labels, in net transitions per eligible edge hour. The paired difference was -0.419. Its 95% person bootstrap interval was [-0.581, -0.254]. The headband mean was 17.9% lower. The cosine between the mean signed flow vectors was 0.997. This shows similar aggregate direction with lower magnitude. The headband had 17.7% fewer stage changes per hour. After division by change count, the difference interval included zero. Intervals also included zero after stage merging. Thus, the five-stage result depends on the score definition. It does not show a separate change in brain dynamics. The package contains all source label tables, a graph proof, code, resamples, and figures. A fresh download gave the same result files. Intervals assume independent people and fixed model outputs.

sleep staging · sleep EEG · directed flow · graph projection · label audit · bootstrap