What Your Sleep Score Can—and Cannot—Tell You

A sleep score looks precise because it is displayed as one number. Underneath, a wrist device uses signals it can measure, such as movement and pulse-related changes, to estimate sleep and wake. A product algorithm may also estimate stages before compressing its outputs into a proprietary score. The score is not a direct recording of every process involved in sleep.

Imagine lying awake but still for half an hour while your watch labels that interval as sleep. Your score may look reassuring even though the night felt interrupted. Your memory is not a lab test either. Ask a narrower question: did the device mark bedtime, wake time and obvious awakenings plausibly? One isolated score is weak evidence about your health.

Treat a single-night result as a prompt to look at context, not as a verdict about sleep health.

Schyvens and colleagues compared six commercial wrist-worn devices with polysomnography during one laboratory night in 62 adults. Their analysis aligned the device labels with the lab's labels in 30-second intervals. Sleep sensitivity was above 90% for all six tested devices: when the lab labeled an interval as sleep, the devices usually did too. Wake specificity was much lower—about 29% to 52%—meaning many lab-scored wake intervals were labeled as sleep by the devices.

That distinction matters more than a single accuracy headline. A device can correctly identify most sleep intervals and still miss some time spent awake. Agreement on sleep-stage labels ranged from fair to moderate by the study's reported kappa values. The finding concerns interval labels, not a validated percentage error for a brand's overall sleep score.

The boundary matters as much as the headline number. Participants wore only two to four of the six models, the sample was mostly men, and the comparison covered one laboratory night. The findings do not rank every device sold today or tell us exactly how the same algorithms perform at home.

The 29–52% figure describes wake-interval specificity in one lab study—not the percentage of your night, your score's error rate or every tracker sold today.

Most devices in the six-device study also differed from the laboratory measure on at least some overnight outcomes, including total sleep time, sleep efficiency, time awake after sleep onset or light sleep. A separate 2025 meta-analysis pooled 24 studies involving 798 people and found differences between consumer wrist devices and polysomnography across common sleep measures. Its subgroup results were not identical across device groups, so neither paper supports a universal ranking or a claim that all wearables are useless.

The six-device comparison covered one lab night. Each participant wore only two to four of the six models, and 52 of the 62 participants were men. The broader review combined different hardware, populations and study methods, with its search updated through March 2024. Those details limit how confidently either result applies to a tracker running a newer algorithm at home.

These studies test agreement with laboratory measures in studied settings. They cannot tell you whether one personal score proves a health problem or improvement.

First, check the clock: do the estimated bedtime, wake time and obvious awakenings match what you remember? Second, compare several nights rather than trying to optimize one stage percentage. Third, note context—travel, illness, alcohol or an unusual schedule—alongside how rested you feel. This is a reading routine, not a way to validate the device or prove what caused a change.

If the score repeatedly disagrees with your experience, keep the pattern as a question rather than a verdict. An American Academy of Sleep Medicine position statement published in 2018 cautioned that consumer sleep technology was not a substitute for diagnosis or treatment of sleep disorders, but could support a conversation during clinical evaluation. That statement predates many current devices and algorithms, so it cannot rate your model today. Persistent sleep concerns deserve a qualified clinical evaluation regardless of a reassuring score.

What changed consistently across nights, and what would independently confirm or challenge that pattern?