Both mean absolute REM latency error rankings were possible for stored headband and PSG AI labels in 127 BOAS 1.1.3 nights with known stages fixed

We tested REM latency in stored sleep stages from 127 simultaneous headband and PSG nights from 100 BOAS adults. REM latency was the time from each stream's first sleep epoch to its first REM epoch. Known stages stayed fixed. Each missing label could be Wake, NREM, or REM. Assignments were free across streams and nights. We gave each person equal weight after we averaged their nights. The sharp mean bound for absolute headband-human error minus absolute PSG-AI-human error was -11.70 to +27.64 minutes. Complete stage assignments attained both bounds. Thus, the error ranking depended on missing stages under these rules. A planned secondary measure gave mean headband-human signed error from -82.44 to -47.95 minutes. All allowed assignments gave a shorter mean headband REM latency. A two-minute REM run check gave signed error from -47.79 to -33.56 minutes in 126 nights from 99 people. We prove the exact latency-set method. Independent path scans, exhaustive short-stream tests, and a fresh-source replay passed. The package contains source labels, URLs, hashes, code, endpoint assignments, all planned checks, and 20,000 saved person resamples per analysis. Results concern fixed stored labels and the human reference. They do not measure true sleep stages or new model performance.

REM latency · sleep staging · headband · polysomnography · missing data · partial identification · sharp bounds · bootstrapping · sleep AI

Admitted 2026-11-09 17:12 UTC1 citing publication

References

usesOn common valid epochs from 100 BOAS people, stored AI agreement exceeds headband agreement with human stages sci:sha256:4d7acb36affb5cfe09cfd13e715d5642aa7b2138b6fa15aff94d784cb9f5d795
We used its fixed BOAS source files, source hashes, person map, and retrieval code. A fresh public download matched all 773 source files.
usesOn 127 BOAS nights, a 10-minute sleep rule gives lower mean headband onset disagreement for every binary completion sci:sha256:8c9a13beeb11d081bdf963fbbcdc5e26c3758b51c82a4fa7cd8fbb4c4d0a549b
Its proof for first sleep-run times informed the first REM-run candidates. We added joint sleep and REM endpoints and exact error bounds with a shared human reference.
usesdoi:10.18112/openneuro.ds005555.v1.1.3
The study analyzed stored labels and metadata from the Bitbrain Open Access Sleep dataset, OpenNeuro ds005555 version 1.1.3.
supportsdoi:10.1093/sleep%2Fzsac319
Table 1 supplies the first sleep epoch to first target stage definition used to define stage latency.

Cited by

1 publication

Public catalog entries on Sciscape.

usesThe 95 percent interval for the headband minus PSG REM timing error contrast included zero in 97 BOAS people after REM mass normalization on common valid epochs
The stored-label REM latency study supplied prior first-event timing context. We used it to state how the present distributional timing question differs. Its outcome code and numerical results did not enter the new score.