Repeated 95% fixed-sample intervals had 39.30% mean miss risk in 128 BOAS nights with fixed labels, known valid epochs, and uniform review

We tested uniform random audits of stored sleep labels in OpenNeuro ds005555 version 1.1.3. The 128 nights came from 100 people. Labels and valid-epoch masks were fixed. We assumed exact review against the stored human reference. We displayed intervals after 20 reviews, then every 10, and at census. Exact probability calculation gave repeated 95% fixed-sample intervals a 39.30% mean chance of at least one true-count exclusion. The established confidence sequence gave 1.04%. The mean gave each person equal weight. Bonferroni gave 0.79%. At an example 10% error threshold, mean review fractions were 29.79%, 51.59%, and 51.73%, respectively. Five planned comparisons, exact enumeration, source checks, and a fresh-source full repeat support the result. The package contains original label tables, provenance, proof, code, results, and repeat instructions. The result concerns one-record guarantees under uniform review. It does not measure physiological truth, new model performance, or reviewer time. Confidence sequences are established; the contribution is an audit application, exact risk comparison, and tool.

sleep staging · label audit · confidence sequences · sampling without replacement · hypergeometric distribution · coverage · sequential review · BOAS · OpenNeuro