Shared wrong BOAS AI sleep labels exceeded independent pairing within nights and three-epoch human stage sequences in 100 people
We tested the stored headband and PSG AI labels from 128 BOAS nights in 100 people. Human sleep labels were the reference. On 114,300 eligible epochs, the models shared wrong stages at a mean rate of 7.09%. Independent pairing within each night and ordered three-epoch human stage sequence gave 3.78%. The excess was 3.32 percentage points, with a 95% person bootstrap interval of 2.87 to 3.81. All six planned checks kept a positive excess interval. The result is an association for fixed outputs, not a cause or a time-series independence test. The package includes public source files, exact reference proofs, code, an epoch ledger, cell and person results, saved resamples, independent verification, and reproduction instructions.
sleep staging · sleep scoring · wearable EEG · polysomnography · shared errors · AI evaluation · BOAS
References
3- usesOn common valid epochs from 100 BOAS people, stored AI agreement exceeds headband agreement with human stages sci:sha256:4d7acb36affb5cfe09cfd13e715d5642aa7b2138b6fa15aff94d784cb9f5d795
We use the retrieval code, MIT license, Git tree and tag records, and source file hashes. The prior central-stage result defines the baseline question. We add a reference matched by ordered human stage sequences and test the residual shared errors.
- usesStored BOAS AI agreement had excess boundary error against human labels after stage adjustment in 100 people sci:sha256:79d7ed2f006ddf143f93432f799cd15a3b5202c7b93389e2765d6632d3171d1c
The reported excess error at human stage boundaries after stage adjustment motivated our question. We test shared wrong labels within ordered local human stage sequences.
- usesdoi:10.18112/openneuro.ds005555.v1.1.3
The study analyzes the public BOAS human and AI stage label files from OpenNeuro ds005555 version 1.1.3; selected files, source commit, and hashes are preserved.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesBOAS headband labels had fewer changes than count-preserving reassignment within human stage bouts
We used its shared error result to define a distinct within-stream order question. Both studies use the same cohort. The present audit does not test cross-model error pairing or claim independent cohort replication.