Your sleep score says you slept badly and you feel fine. Which is wrong?

The symptom

• Your sleep score says 45 minutes of deep sleep and you woke up feeling excellent
• The stage breakdown swings wildly between nights that felt identical
• It logs you as awake when you know you were asleep, or vice versa
• Two devices on the same night tell two different stories

Why the tracker might genuinely be wrong

Robbins et al. (2024) measured three consumer devices against clinical polysomnography, n = 35, single night inpatient.

Sleep versus wake was fine — 91-93% agreement across the board. Your total sleep time is broadly trustworthy.

Four-stage classification was much weaker— 76.3% (Oura), 75.0% (Apple), 70.9% (Fitbit), with Cohen's kappa of 0.65, 0.60 and 0.55.

Deep sleep specifically was all over the place. Sensitivity of 79.5% (Oura), 61.7% (Fitbit), 50.5% (Apple Watch). The Apple Watch overestimated light sleep by 45 minutes a night and underestimated deep sleep by 43.

So if you are on an Apple Watch worrying that you only got 45 minutes of deep sleep: the published evidence says the device identifies roughly half of it, and the true figure was plausibly close to double what you saw. The anxiety in that case is a measurement artefact.

Why "how you feel" is not a reliable check either

We should be even-handed. How rested you feel is affected by a great deal your wearable does not measure: where you woke in a sleep cycle, circadian timing, sleep fragmentation, alcohol, illness, caffeine timing, stress. It is not a gold standard either.

The reasonable position is that a single night's stage breakdown is weak evidence from both directions. Trends over weeks are far more informative than any one night, from a device or from your own impression.

What actually fixes it

1. Buy the device with the best measured result

If stage data is why you are here, the Oura Ring is the only consumer device with a strong published polysomnography result. A finger is a better optical site than a wrist. The caveat we keep repeating: that result belongs to the Gen3.

2. Fix compliance instead — often the bigger problem

A large share of "my sleep data is bad" is actually "my sleep data is missing", because the watch was on the charger at bedtime. That is a genuine and underrated failure mode.

The Garmin Venu 3 states 14 days, so it survives the nights. The Garmin Index Sleep Monitor is a night-only upper-arm band — the placement the 2025 study found beats the wrist — which means your watch can charge overnight while something else records.

3. Recalibrate what you expect from it

Free, and often the right answer. Use total sleep time and week-over-week trends, which are well supported. Treat the four-color hypnogram as an indication rather than a measurement. Stop optimizing a number with 50-80% sensitivity.

This won't fix it if…

You suspect a sleep disorder. No consumer wearable diagnoses sleep apnea or anything else, and none of these devices is a medical instrument. Persistent unrefreshing sleep, loud snoring or witnessed breathing pauses are reasons to see a doctor — a better tracker is not.

You want certainty about your sleep stages. That requires polysomnography — electrodes, a lab, a technician. Every consumer device is inferring stages from heart rate, movement and temperature, and the best of them agrees with the clinical standard about three quarters of the time.

Your sleep is genuinely poor. Then the tracker is right and the fix is not a purchase.