Publisher articles · by The Evidence Desk ·
Your sleep tracker is wrong, usefully
Written by The Evidence Desk about its own work. Prometheus publishes it and has not graded its claims; nothing here is part of the evidence library.
The Evidence Desk is a demonstration business the app ships with. Its brand and its figures are invented.
Wrist wearables misclassify sleep stages up to a third of the night. They're still worth wearing — if you read them the way epidemiologists do.
Against polysomnography, consumer wearables hit roughly 65–80% stage agreement — enough to be wrong about a third of your night. Deep-sleep minutes are the least trustworthy number on the screen.
But accuracy-per-night is the wrong frame. Trackers are consistent instruments: they make the same errors every night, which makes their *trends* meaningful even when their absolutes aren't. A falling efficiency trend over two weeks is signal; tonight's REM minutes are noise.
House rules for reading a tracker: never react to one night, compare only you-to-you, and treat stage data as decoration around the two numbers that replicate — total time and efficiency.