Your Sleep Tracker Knows How Long You Slept. Not How Deeply

On August 20, a California buyer named Madison Surber filed a proposed class action against Oura in federal court in the Northern District of California. The target was one line of marketing: “95% Sleep Staging Accuracy compared to clinical sleep lab.” Three days later, Oura published a long defense of its science, and in the middle of it sat a concession that matters more than the lawsuit. The 95%, the company wrote, describes telling sleep from wake. Sorting the night into light, deep and REM sleep is “a harder problem,” with agreement of roughly 76 to 79% in healthy adults.

That gap, between knowing that you slept and knowing how you slept, is the whole story of consumer sleep tracking. The research now says something clear: duration has earned some trust, the stage breakdown has not.

Two different tests behind one number

Validation studies all work the same way. A volunteer sleeps wired to polysomnography, the clinical reference that reads brain waves, eye movements and muscle tone. A technician labels every 30 second slice of the night. The device produces its own label for the same slice, and the two are compared.

The easy version of the test asks only whether each slice was sleep or wake. Since most of a night is sleep, leaning toward “asleep” scores well by default. The hard version asks which of four states the slice belongs to. A ring or a watch never sees a brain wave. It infers stages from pulse, heart rate variability, movement and skin temperature. Oura’s own account of its foundational 2021 study makes the point: a motion only model reached 57% on four stages, and adding heart and temperature signals lifted that to 79%, against 96% for sleep versus wake.

What the lab studies found

The study Oura cites first was run at Brigham and Women’s Hospital and published in Sensors in October 2024. Thirty five healthy adults aged 20 to 50 spent one night in the lab wearing an Oura Ring Gen3, a Fitbit Sense 2 and an Apple Watch Series 8. All three caught at least 95% of sleep. On nightly totals, the ring did not differ significantly from the lab for total sleep, light, deep or REM sleep. The Apple Watch underestimated deep sleep by 43 minutes and overestimated light sleep by 45; the Fitbit missed 15 minutes of deep sleep. Two caveats sit in the paper itself. Oura funded the study, and the lead author is a member of Oura’s medical advisory board. And person by person, agreement with the lab on deep sleep was poor for all three devices, with intraclass correlations between 0.13 and 0.36.

A Belgian team funded by the Flemish innovation agency VLAIO tested six wrist devices on 62 adults, 52 of them men, average age 46, one night each, in SLEEP Advances in March 2025. Every device detected more than 90% of sleep epochs but only 29 to 52% of wake epochs. Agreement across stages, measured by Cohen’s kappa, ranged from 0.21 to 0.53, fair to moderate; the Apple Watch Series 8 did best. The Withings ScanWatch does not separate deep from REM sleep, so the researchers compared its “deep” figure with the lab’s deep plus REM. It overshot by 73 minutes on average, 217 minutes against 144, and the limits of agreement ran from 37 minutes under to 186 minutes over.

Then come the people the marketing rarely shows. At the Charité sleep lab in Berlin, researchers tested three rings on patients referred with a mix of sleep disorders and other medical conditions, reported in Scientific Reports in March 2025 with no competing interests declared. On average, Oura’s total sleep came within 12 minutes of the lab. Its four stage accuracy was 53%, and the authors stress that the good average masked large errors in individual patients.

A University of Massachusetts study funded by the National Institute on Aging, published in SLEEP Advances in January 2026, recorded polysomnography at home in 13 young adults and 19 healthy adults aged 56 to 80. In the young group, no device differed significantly from the lab on total sleep. In the older group, the lab scored an average of 17 minutes of deep sleep on the nights the Oura Ring recorded; the ring reported 88. The Withings Sleep Mat overshot deep sleep by 97 minutes. Even in the young group, the ring added 81 minutes of deep sleep. The authors put it plainly: an inflated amount of deep sleep “may give a false impression of sleep quality.”

And the Oura Ring 4?

In September, a Mint journalist in Mumbai spent a night wired to a hospital polysomnograph while wearing an Oura Ring 4, a Garmin Forerunner 265S and a Fitbit Air. The lab counted 5 hours 56 minutes of sleep. The ring came closest at 5 hours 29 minutes; the Garmin said 7 hours 22, the Fitbit 4 hours 23. That is one person on one night, with a ring supplied by Oura, and it speaks to duration only. The peer reviewed staging studies I could find tested the Gen3 ring and its algorithm, not the Ring 4 hardware. Results also hold only for the firmware tested.

Why the errors land where they matter

The pattern across these papers is consistent. Error is smallest in young, healthy volunteers sleeping well, the group least likely to need the data, and largest in older adults and patients, whose deep sleep is scarce and whose nights are broken. Wrist and finger signals carry real information, as with the light and temperature rhythms in a recent study of 809 adults, but a clue about the body is not a measurement of the brain. Funding deserves the same scrutiny it gets in any health study: the most flattering staging result here was paid for by the company whose ring won.

What comes next is partly legal and partly scientific. The Surber complaint is an allegation, and courts have not been uniformly receptive to accuracy suits against wearables. The more useful outcome would be independent, multi night studies of current hardware in older adults and patients, reported with the stage numbers next to the sleep versus wake numbers instead of hidden behind them.

I read a sleep tracker the way I read a bathroom scale that rounds generously: useful for the trend, misleading for the decimals. Duration and regularity are numbers these devices have earned the right to show. The stage chart is a model’s guess drawn with the confidence of a measurement, and the industry chose to market the easy score under the hard score’s name. Whatever a judge decides, the fix is cheap. Print the sleep versus wake figure as what it is, print the four stage figure next to it, and let buyers see that the second one is the weaker of the two. Until that happens, the 95% will keep saying more than the evidence does.