How your watch decides you were in deep sleep

A watch can turn an apparently uneventful night into a coloured chart of sleep, waking and changing stages. Those labels look like a direct view inside the night. Understanding how they are produced makes the chart more useful—and explains why the score beside it answers a separate question.[3]

The stages come from sleep physiology

Sleep researchers distinguish three non-REM stages and REM sleep. Deep sleep is the third non-REM stage; it tends to occupy more of the earlier part of the night, while REM periods generally become longer later. The pattern changes with age. This gives the colours a biological meaning, but it does not turn the largest possible amount of one colour into a universal goal.[1]

The reference test, polysomnography, records signals including brain activity, eye movements and muscle activity. A wrist device supplies a different set of measurements. Researchers assess its stage estimates by comparing them with the sleep-study record, so the key question is how reliably its accessible signals predict the reference labels.[6]

Different wearables listen to different signals

Apple’s October 2025 technical paper describes an accelerometer-based classifier. It uses movement, including small motion associated with breathing, to assign stages in thirty-second intervals. Heart-rate measurements are not inputs to this sleep-stage model. Apple combines the first two non-REM stages into “Core” sleep. Its 2025 update particularly targeted quiet waking, when lying still can resemble sleep.[2]

Oura describes a different approach for its ring. Optical sensing follows the pulse; movement and temperature add other information. Changes in heart rate and the intervals between beats provide clues about the body’s activity during the night. A trained model combines those clues to infer a stage. Consequently, “the tracker uses AI” tells us very little about which physical evidence a particular device actually uses.[4]

This combination has an intelligible purpose. Stillness alone leaves several possibilities open: someone may be deeply asleep, dreaming or awake without moving much. Other signals can help distinguish those possibilities when they have useful relationships with the reference stages. They remain indirect measurements, interpreted together. A ring’s explanation cannot automatically be applied to a watch, even when both apps display similarly named colours.[4]

Better at finding wakefulness can involve a trade-off

The updated model recognised more waking, with a small reduction in sleep sensitivity. Reporting both values shows the trade-off that a single accuracy headline can conceal. These are Apple’s own validation results for specified algorithms.[2]

Why one accuracy number travels badly

A 2021 Oura research paper provides a useful experiment in adding information. It analysed 440 nights from 106 people using research hardware and five-fold cross-validation. Within that study, four-state accuracy rose from 57% for its accelerometer model to 79% for a model combining movement, temperature, heart-related and circadian features. That isolates a design choice inside one research programme. It is not a current ring-versus-watch league table.[5]

The study also assessed how estimated time in stages agreed with the sleep study, alongside interval-by-interval classification. Those are different achievements. Imagine two charts that each contain sixty minutes of deep sleep, but place that hour at entirely different times. Their deep-sleep totals match perfectly, while the detailed night does not. That original example explains why duration agreement alone cannot establish that a tracker correctly recognised when a stage occurred.[5]

Independent testing adds another perspective. A 2025 six-device study involved 62 adults, mostly men, with one laboratory night each. All six devices detected more than 90% of sleep, but identified only about 29–52% of waking. The usable Apple sample was just 20 people, with 15 failed recordings. Measurements were collected in 2023–24. Its participants, devices and software therefore differ from Apple’s later validation; the numbers do not measure an identical system under identical conditions.[6]

The score is an extra layer of interpretation

Apple’s current sleep-score documentation assigns up to fifty points to duration, thirty to consistency and twenty to interruptions. It is therefore a summary of selected sleep habits and recorded disruption. It does not simply award more points for more deep sleep. In the iPhone Health app, Browse → Sleep provides day, week, month and six-month views; Show More Sleep Data separates stages, amounts and comparisons. Those views let the reader see what sits behind a single morning number.[3]

A practical way to read the chart is to give each display its own job: the timeline describes estimated changes during the night, totals summarise estimated duration, and a score applies the company’s weighting. Looking across several nights also makes an unusual bedtime or missing recording easier to recognise in context.[3]

What the chart can contribute to a health conversation

Sleep staging and screening for a breathing disorder are separate functions. The American Academy of Sleep Medicine’s 2025 advisory acknowledges that some consumer technologies have regulatory clearance to assess sleep-apnoea risk. It also explains that risk assessment cannot diagnose the condition or establish that it is absent. A reassuring stage chart therefore cannot settle that different clinical question.[7]

The chart can still make a conversation more concrete: when sleep seemed disrupted, how the recorded pattern changed and whether that matched the person’s experience. The useful starting point is to identify which feature produced the information and what that feature measures. That preserves what a wearable can contribute without asking a coloured timeline to answer every question about a night’s sleep.[7]

Report an error or suggest a correction