With fixed BOAS v1.1.3 labels and known masks, a stratified audit cut headband error estimate variance by 36.6% at 100 epochs per night
An audit grouped eligible epochs by stored headband stage and agreement with PSG AI labels. In 128 BOAS v1.1.3 nights from 100 people, it cut exact cohort error estimate variance by 36.6% against uniform sampling. Both designs selected 100 epochs per night. Allocations used prior reference labels from the other 99 people. The audit assumed known eligibility masks, fixed reference outcomes, and independent random selections. The group size weighted estimate was unbiased. Omitting group weights raised its expectation by 5.59 percentage points. A constant variance proxy cut variance by 33.8%. Independent checks and a fresh download reproduced all numerical files. The package provides code, source tables, data hashes, a proof, results, and an audit example. The finding is conditional on these stored labels. Prior reference bank cost and reviewer time were outside the comparison.
sleep audit · stratified sampling · variance estimation · sleep staging · BOAS · EEG · optimal allocation · finite population correction · unbiased estimation
References
5- usesAI disagreement reduced expected label and stage share errors in a 100-person BOAS benchmark under assumed exact label replacement sci:sha256:7fa416399ee873aeef44b5dee17b1a469bd938827b345a6688c1792e3967c2e5
Its test of review by AI disagreement helped define our agreement groups. That study evaluated label correction. We used agreement for a weighted error audit.
- usesOn common valid epochs from 100 BOAS people, stored AI agreement exceeds headband agreement with human stages sci:sha256:4d7acb36affb5cfe09cfd13e715d5642aa7b2138b6fa15aff94d784cb9f5d795
Its pinned BOAS source subset and manifest reached this study through publication 57450229-8dcd-4953-b677-dbf86bf727dc. They preserve the original source provenance.
- usesWith fixed BOAS labels and known masks, 5 minute block audits had 5.14 times epoch audit variance at equal budgets sci:sha256:cb02e56b1dbad356142bbad5293f70d48367304955ed5b187cbfa8a17ea912ac
We reused its unchanged BOAS tables, source hashes, repository snapshots, and MIT parser and retrieval code. These supplied the fixed audit frame.
- usesdoi:10.18112/openneuro.ds005555.v1.1.3
The analysis used BOAS OpenNeuro ds005555 version 1.1.3 event labels.
- doi:10.1111/j.2397-2335.1934.tb04184.x
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesWith fixed BOAS v1.1.3 labels and known masks, disagreement sampling reduced paired accuracy audit variance by 96.8% at a cap of 100 epochs per night
The package reuses 773 unchanged BOAS source files, source manifests, Git records, source.py, and retrieve.py. Its absolute-error audit informed this paired-gap study. The source files were then downloaded again and checked.