You wake feeling good. Your wearable gives you a 61. The next morning you feel flattened, but the app celebrates an 88. Which one is wrong?

Possibly neither. The device is estimating sleep from a limited set of signals. You are reporting function from inside the system. Both observations contain information, and both can be noisy.

A sleep score is a model output, not a direct measurement of recovery.

What the device can see

Consumer wearables commonly use movement and optical pulse signals. Some add skin temperature, oxygen-related estimates, respiratory rate, or other sensors. Software combines those signals to infer sleep and wake, sleep timing, and sleep stages. The score then compresses several estimates into one number.

Polysomnography, the clinical reference standard, measures a different and richer set of signals, including brain electrical activity, eye movements, muscle activity, heart rhythm, breathing, airflow, oxygen, and limb movement. A wrist or ring does not become a sleep lab because its app shows a hypnogram.

Where wearables help

A 2023 multicenter study compared 11 consumer devices with polysomnography. Performance varied by device and metric. Across the category, the useful lesson is not that wearables are worthless. It is that broad sleep-wake and duration estimates tend to be more dependable than precise sleep-stage claims.

Wearables are strongest when the same person uses the same device under similar conditions and watches longer trends:

  • Bedtime and wake-time consistency.
  • Estimated sleep duration across weeks.
  • Resting heart rate and heart-rate variability relative to personal baseline.
  • Changes that track alcohol, illness, travel, training load, or schedule disruption.
  • Adherence to a planned sleep window.

That is why public figures such as WHOOP founder Will Ahmed and Andrew Huberman can make wearable data useful in a protocol discussion. They can surface patterns and hypotheses. The device still does not supply the diagnosis.

Where scores fail

Algorithms change. Hardware changes. Fit changes. Tattoos, skin contact, movement, arrhythmias, unusual schedules, and medical conditions can affect signals. A model trained on one population may perform differently in another. A composite score can also hide disagreement among its inputs.

The American Academy of Sleep Medicine states that consumer sleep technology should not be used to diagnose or treat sleep disorders in place of validated testing. A normal-looking score does not rule out sleep apnea. An abnormal score does not establish it.

There is also a behavioral risk. "Orthosomnia" describes sleep problems driven or worsened by a pursuit of perfect wearable data. If checking the score increases anxiety, delays getting out of bed, or makes a person distrust a normal day, the measurement is no longer neutral.

A better way to use the score

Start with the question. If the goal is schedule consistency, use bedtime and wake time. If the goal is reducing alcohol-related disruption, compare matched nights and include how you felt. If the goal is diagnosing apnea, stop trying to solve it with a consumer score and seek clinical evaluation.

Use weekly averages rather than reacting to one night. Pair device data with a short morning check: sleep quality, energy, mood, soreness, and daytime sleepiness. Record obvious confounders. When the device and symptoms disagree for several weeks, investigate the disagreement instead of choosing the preferred answer.

The score is most valuable as a prompt: What changed? Is the pattern persistent? Does it affect function? Is this still an optimization problem, or is it now a clinical question?