Loading...

When you fell asleep

Checked against an FDA-cleared home sleep test and a validated smartwatch
OPTICALMOTION

Accuracy

96%

Sleep Onset Agreement

of nights within half an hour, against a validated smartwatch worn on the same nights

Reference standard

Validated smartwatch

Sample size

56 participants

The Ultrahuman Ring places the moment you fell asleep a typical 8.3 minutes from an FDA-cleared home sleep test reading brain activity, and a validated smartwatch agrees.

-12-8-40+4+8+12vs an FDA-cleared home sleep test±8.3vs a validated smartwatch±9typical difference from the reference on a single night (minutes)0 = no difference
8.3 mintypical night-to-night error, against the home sleep test
9.0 minaverage per-night difference, against a validated smartwatch
96%of nights within half an hour, against the smartwatch
−1.4 minthe ring calls sleep early, on average

An FDA-cleared home sleep test, reading brain activity at the forehead, can see the minute the brain stops surfacing. The Ultrahuman Ring has to find that same minute from the outside, from your heart rate, your movement and your temperature, across a boundary the brain crosses, comes back over, and crosses again.

Against that home sleep test — a home sleep study rather than a laboratory polysomnogram — the typical night-to-night error is 8.3 minutes, across 20 paired nights from 4 people. A separate comparison against a validated smartwatch, on 106 nights from 56 people, puts 96% of nights inside half an hour, with an average per-night difference of 9.0 minutes.

These are per-night figures, what a single night actually looks like, rather than an average across nights. Averages come out far smaller, because nights that run early and nights that run late cancel each other out.

The small difference that remains points the same way each night: the Ultrahuman Ring calls sleep about a minute and a half early against the home sleep test, and a validated smartwatch shows the same direction. A steady offset leaves your night-to-night patterns intact, where random error would scramble them.