A 1% five-stage human-label edit limit allowed both signs of the mean headband-minus-human stage-change rate in 128 BOAS v1.1.3 nights from 100 people with fixed valid edges

We tested the sensitivity of the stored headband minus human sleep stage change rate in BOAS v1.1.3. The study used 128 nights from 100 people. At most 1% of shared valid human epochs in each night could receive any of five stage labels. Valid epochs, adjacent edges, and headband labels stayed fixed. Counts and valid edges were pooled within each person. People had equal weight. The original mean gap was -1.7788 changes per valid edge-hour. Exact attained limits were -4.0481 and +0.4050. Both signs remained possible when edits were restricted to epochs next to original human changes. The gap stayed negative when only N1 and N2 substitutions were allowed. The exact minimum edits to match every night's headband count totaled 1.1739% of valid epochs; a count match does not establish event timing. An exact integer method and proof support the results. Independent count sets checked all tested limits and 2,432 saved sequences. A fresh source download reproduced every result table and array. The package preserves source labels, fixed URLs and hashes, source code, exact cost curves, attaining labels, all planned checks, and reproduction instructions. These are sensitivity limits under an assumed edit set. They do not estimate human errors or true sleep stages.

sleep staging · stage changes · label sensitivity · exact bounds · reference labels · BOAS · wearable EEG