Sleep tracking. What a wrist can see, and what it cannot.

What a night is made of

Sleep is not one state. The National Heart, Lung, and Blood Institute describes it as cycling between non-REM and REM phases, with the cycle restarting roughly every 80 to 100 minutes and typically four to six cycles across a night.

Non-REM has three stages: the first is the transition between wake and sleep, the second is where sleep proper begins, and the third is deep sleep, also called slow-wave sleep. The order matters and it is not even across the night — deep sleep predominates earlier, and REM becomes more prevalent later, which is why a night cut short at the end loses proportionally more REM than deep sleep.

One fact worth carrying: slow-wave sleep peaks in early childhood, drops sharply in adolescence, and keeps declining through adulthood, with some older adults getting very little. Any article that gives you a target percentage of deep sleep without asking your age is giving you a figure that is wrong for most of the people reading it.

How much sleep, and who that is for.

The reference figure comes from a joint consensus statement issued in 2015 by the American Academy of Sleep Medicine and the Sleep Research Society. A 15-member expert panel worked through a formal 12-month process, reviewing more than five thousand articles across nine health domains, and concluded that adults aged 18 to 60 should sleep 7 or more hours per night on a regular basis. The Centers for Disease Control and Prevention publishes the same floor for that age group, with slightly different bands above it: 7 to 9 hours for ages 61 to 64, and 7 to 8 hours from 65.

Two qualifications on that number that the consensus statement makes itself. It is a population recommendation, not a personal prescription — the panel notes that sleep need varies with genetic, behavioural, medical and environmental factors. And it applies to adults 18 to 60; children and adolescents have their own, longer, guidance.

The health associations attached to short sleep are real and worth stating carefully. Regularly sleeping under 7 hours is associated with higher rates of weight gain, diabetes, high blood pressure, heart disease, stroke and depression. Associated with is doing work in that sentence: the underlying evidence is largely observational, so these are patterns seen across populations rather than an established chain of cause and effect in any one person.

Asleep or awake: what a wearable gets right.

The reference standard for measuring sleep is polysomnography — a full overnight laboratory recording of brain activity, eye movement, muscle tone, breathing and heart rhythm. Everything below is what happens when consumer devices are put next to one.

A 2021 study in the journal Sleep did exactly that: 34 healthy young adults, three consecutive laboratory nights including one deliberately disrupted, with polysomnography, research-grade movement monitoring and seven consumer devices recording at once.

Detecting sleep worked well. Every device correctly identified time asleep at a high rate. Detecting wake did not. Correctly identifying wake ranged from poor to mediocre across the devices, and five of the seven were at or above the research-grade movement monitor's own low figure — which tells you this is a limitation of the approach rather than of any one brand. The practical consequence is specific and worth remembering: wearables tend to over-count sleep and under-count how broken the night was. If you lay awake for a while and your device did not notice, that is the expected behaviour, not a fault.

Sleep stages: the weakest claim on the screen.

Stages are the part of a sleep screen that looks most scientific and is supported least well.

In the same 2021 study, every device overestimated light sleep, and on average the devices failed to correctly identify between 30% and 50% of both deep sleep and REM sleep. That is not a rounding error. It means a deep-sleep figure on a consumer device is a rough estimate of something a laboratory measures directly, and that two devices on the same wrist on the same night can reasonably disagree by a lot.

Duration and continuity are measured better, but they are not clean either. A 2025 meta-analysis in the American Academy of Sleep Medicine's own journal pooled 24 studies and 798 participants across a range of wrist devices and found consistent directional bias: total sleep time underestimated by about 17 minutes on average, sleep efficiency underestimated by about 4.7 percentage points, and time awake after falling asleep overestimated by about 13 minutes. Variation between studies was high, so those are directional cautions rather than the error you should expect from your own device.

Its authors also named the structural problem: manufacturers' algorithms are proprietary and undisclosed, which limits how far anyone outside the company can interpret a number. That applies to every company in this category, Omnisio included.

Why the trend beats last night

Put the two findings together and the conclusion is not subtle. On any single night your device may have over-counted your sleep, missed some of your awakenings, and misallocated a third or more of your deep and REM sleep. Reading that night as a verdict is reading noise as signal.

Across weeks, most of that error is in the same direction each night. A device that consistently over-counts sleep by a similar amount will still show you clearly that you slept less this fortnight than last, or that your bedtime has drifted an hour later since the season changed. The systematic bias mostly cancels out of a comparison; it does not cancel out of a single number.

So the honest way to use a sleep tracker is as an instrument for noticing patterns you would otherwise have to remember, and not as a nightly grade. If you woke up feeling fine and the app disagrees, the app is the less reliable witness.

Getting something useful out of one

None of this requires ignoring the device. It requires reading it for what it is good at.

• Watch the direction, not the value. Whether this month is worse than last month is a question your device can answer. Whether you got 94 minutes of deep sleep is not.
• Compare yourself with yourself. Deep sleep declines with age, and varies between people for reasons that have nothing to do with health. Someone else's numbers are not a target.
• Wear it the same way. Position, tightness and consistency change what the sensor sees. A month of comparable nights is worth more than a month of careful nights and careless ones.
• Treat a missing or odd night as missing. A device that came off, sat loose, or recorded a very short night has not measured a bad night — it has failed to measure a night.
• Trust the feeling over the figure. Persistent daytime exhaustion matters whatever the app says, and a good score is not evidence that it does not.

What a sleep tracker cannot tell you

It cannot tell you whether you have a sleep disorder. The validation studies behind everything on this page were carried out in healthy people — the 2021 laboratory study excluded people with sleep disorders outright — so nothing in that literature supports screening, detection or diagnostic language for sleep apnoea, insomnia or anything else.

Omnisio does not attempt it either: the app does not detect breathing disorders in sleep, does not screen for heart-rhythm problems and does not measure blood pressure. Omnisio is a wellness product for general lifestyle use, neither the app nor the devices used with it hold FDA, CE-MDR or TITCK medical device clearance, and nothing in it is intended to diagnose, treat, cure or prevent any disease.

If you snore heavily, stop breathing during sleep, wake unrefreshed for weeks on end, or fall asleep during the day, that is worth taking to a physician. A sleep laboratory can measure what a wrist cannot, and a low score in an app is not a reason to wait for a worse one.

Where Omnisio fits

Omnisio's morning Sleep reading shows the stages, how broken the night was, how long you were actually asleep, and how that sits against your own recent nights. Naps count towards it. That last part — *against your own recent nights* — is the design decision this page argues for on the evidence: given how much stage estimates vary between devices and between people, a comparison against yourself is the only comparison that survives contact with the research.

The other one worth naming is what happens on a night the band could not read honestly. Omnisio gives no score at all and says so, rather than printing a low one. Given that a thin or interrupted recording is exactly the case where consumer devices are least accurate, a number produced from one would not be a low reading of you — it would be a reading of nothing.

How any of it is produced is not published, here or anywhere else. The full description of the morning readings is on the Recovery page, linked in the rail beside this section.

Where this comes from

Six published sources, named plainly. Each was checked live on 8 September 2026.

• Source: American Academy of Sleep Medicine and Sleep Research Society, 2015. Joint consensus statement: adults aged 18 to 60 should sleep 7 or more hours per night on a regular basis; regularly sleeping less is associated with a list of health outcomes. The statement itself notes sleep need varies between individuals.
• Source: Centers for Disease Control and Prevention, 2024. Public-health guidance giving the same 7-hour floor for adults 18 to 60, 7 to 9 hours for ages 61 to 64 and 7 to 8 hours from 65, with the associated risks described as associations rather than established causes.
• Source: National Heart, Lung, and Blood Institute, 2022. Sleep architecture: cycles restarting roughly every 80 to 100 minutes, four to six per night, three non-REM stages, deep sleep predominating earlier and REM later, and slow-wave sleep declining through adulthood.
• Source: the journal Sleep, 2021. 34 healthy young adults, three laboratory nights, seven consumer devices recorded alongside polysomnography: high accuracy detecting sleep, poor accuracy detecting wake, light sleep overestimated by all devices, and 30% to 50% of deep and REM sleep missed on average.
• Source: the American Academy of Sleep Medicine's peer-reviewed journal, 2025. Meta-analysis of 24 studies and 798 participants: total sleep time underestimated by about 17 minutes, sleep efficiency by about 4.7 percentage points, and time awake after sleep onset overestimated by about 13 minutes, with high variation between studies and undisclosed manufacturer algorithms named as a limit.
• Source: the journal Sleep, 2010. Meta-analysis of 16 prospective studies covering nearly 1.4 million people: short sleep associated with about 12% higher all-cause mortality and long sleep with about 30% higher. Entirely observational, based on self-reported sleep, with reverse causation raised by the authors.

Questions