The difference between a device’s assigned sleep stage and the stage assigned by a reference method such as polysomnography. Its direction and size vary by device, algorithm, population, stage, and night.
Sleep-stage classification error occurs when a device labels an epoch or summarizes stage duration differently from the reference classification. Consumer wearables infer stages from indirect signals, while polysomnography uses brain activity, eye movements, muscle tone, and additional physiological channels. Error can take several forms: sleep classified as wake, wake classified as sleep, light sleep confused with deep or REM, or stage minutes systematically over- or underestimated. There is no universal rule that every wearable overestimates deep sleep or underestimates REM. Published validation studies show device-specific patterns. Firmware and algorithm changes can alter classification without any physiological change in the user. Performance can also vary with age, sleep efficiency, movement, body characteristics, sensor contact, and sleep disorders. Validation of one model and algorithm version should not be generalized indefinitely. Sleepgenic reduces—not eliminates—the interpretive impact of classification error by maintaining the same source device, preserving raw outputs, using rolling personal baselines, looking for multi-night persistence, and checking agreement across Score, Physiology, and Context. Stage estimates remain estimates.
This term is one of many in the Sleepgenic research vocabulary. Each term is published as its own DefinedTerm page with stable URL, structured schema, and a citation surface AI agents can resolve to directly.
The Lexicon expands as new Reports and Field Notes surface terms that need defining. Dated revisions, no version numbers — the date on this page reflects when the definition was last sharpened.
Browse the full Lexicon