Nutrition & Recovery ·

What Your Watch or Ring Gets Right About Sleep

Sleep-lab studies show consumer watches, rings and straps are good at telling when you slept and roughly how long, poor at splitting sleep into stages, and that a bad sleep score can itself make you feel worse the next day.

What Your Watch or Ring Gets Right About Sleep

Sprint Summary

The short version — read this if you're short on time.

Consumer sleep trackers, whether on your wrist, your finger or under the mattress, are good at one job and poor at another. In sleep-lab studies with polysomnography as the reference, they agree on whether you are asleep or awake most of the time (86 to 89% of the night in a 2022 study of 53 adults) and give a fair estimate of when you slept and for how long on a normal night. They are much weaker at spotting time awake in bed, so they tend to overestimate sleep on broken nights, and their split into light, deep and REM sleep matched the lab only 50 to 61% of the time for wrist, ring and strap devices.

So use the tracker for timing, consistency and trends over weeks, not for a nightly verdict on your recovery. Don't change training because of one night's deep-sleep figure, and judge a low score alongside how you actually feel. Watch for the score itself becoming a stressor: in a randomised study of 63 people with insomnia, being told they had slept badly made them feel sleepier and less alert the next day. Snoring with pauses in breathing or persistent insomnia are for your GP, not your app.

Safety & Context

A consumer sleep tracker is not a medical device and cannot diagnose a sleep disorder. Loud snoring, gasping or pauses in breathing that a partner notices, or heavy daytime sleepiness despite enough time in bed, can be signs of sleep apnoea and are a reason to see your GP, whatever your sleep score says.

Insomnia that lasts for weeks, or worry about sleep that is building, deserves a GP appointment. Cognitive behavioural therapy for insomnia is the recommended first treatment; this article does not suggest sleep medication, supplements or sedatives, and you shouldn't start any to improve a tracker score.

If checking your sleep data is making you anxious about bedtime or leaving you feeling worse in the morning, take a break from the tracker. In a randomised study, people with insomnia who were told they had slept badly reported more sleepiness and less alertness the next day, regardless of how they had actually slept.

Device studies date quickly. The accuracy figures here come from specific device generations tested between 2017 and 2023 and may not describe your device after a firmware update.

Full Distance

The complete research and analysis.

You wake up, reach for your phone and find out how you slept. Seven hours ten, a sleep score of 68, "not enough deep sleep", and a recovery rating in amber. You felt fine until you looked. Now you're wondering whether to cut the morning's run short.

Most triathletes who own a sports watch, smart ring or recovery strap now have a nightly sleep report whether they asked for one or not. Sleep matters for training and racing, and TriForward's "Does More Sleep Make You Faster? What Sleep-Extension Studies Show" covers what extra sleep can do. The question here is narrower: how much of what the device tells you about your sleep is actually right?

This guide looks at studies that tested consumer sleep trackers against laboratory sleep recording, what they get right, where they go wrong, and how to use the data without letting a number decide how you feel.

How a sleep lab measures sleep, and how your watch guesses

The reference method is polysomnography (PSG): electrodes on the scalp record brain waves, with sensors for eye movements, muscle tone, breathing and heart rhythm. A trained scorer splits the night into 30-second slices ("epochs") and labels each one as wake, light sleep (N1 or N2), deep or slow-wave sleep (N3) or REM sleep. Sleep stages are defined by brain activity, so you need brain signals to score them properly.

Consumer devices don't measure brain waves. Wrist watches, rings and straps combine:

  • Movement, from an accelerometer. Still for a while usually means asleep.
  • Heart rate and heart rate variability, from an optical sensor that shines light into the skin.
  • Sometimes skin temperature and breathing rate, estimated from the same signals.

An algorithm, usually trained on sleep-lab data from a particular group of people, turns those signals into a guess at each epoch. Bedside devices (a mat under the mattress or a radar unit on the bedside table) do something similar from breathing and movement.

A 2019 review by de Zambotti and colleagues in Medicine & Science in Sports & Exercise made two points that still apply. The algorithms are proprietary, so researchers can't see how a device reaches its answer. And firmware updates can change the output of the same device overnight, which means a validation study describes the device and software version that was tested, not every later version.

That's the first rule for reading this research: each figure below belongs to a device type and a year.

What validation studies measure

Before the numbers, a short glossary. Validation studies line up the device and PSG epoch by epoch and report:

  • Agreement or accuracy: the percentage of epochs where the device and PSG agree.
  • Sensitivity: how well the device detects sleep when PSG says you are asleep.
  • Specificity: how well the device detects wake when PSG says you are awake.
  • Cohen's kappa: agreement corrected for chance. Roughly, 0.2 is fair, 0.4 moderate and 0.6 substantial.

Most of a night is sleep, so a device that labelled every epoch "asleep" would score high agreement and still be useless at spotting wake. That's why specificity and kappa matter more than headline accuracy.

The key studies

Seven devices in a sleep laboratory (2021)

Chinoy and colleagues, in a 2021 paper in Sleep, tested seven consumer sleep trackers (four wearables, including two wrist-worn sports watches, and three bedside devices) alongside a research-grade actigraphy watch, against PSG.

  • Who: 34 healthy young adults (22 of them women; average age 28), not athletes.
  • Design: three consecutive nights in a sleep laboratory, one of which was deliberately disrupted.
  • Detecting sleep: sensitivity was high for every device (at least 0.93).
  • Detecting wake: specificity was low to medium (0.18 to 0.54). In other words, the devices missed much of the time participants were lying awake.
  • Sleep stages: comparisons were mixed.
  • Bad nights: devices tended to perform worse on nights with poorer or disrupted sleep.

Most devices did as well as or better than research actigraphy at the sleep-wake task, although the two wrist sports watches tested (2017-era models) performed worse than the others.

Six devices on the same people, same night (2022)

A 2022 study by Miller, Sargent and Roach in Sensors fitted 53 adults (26 women, 27 men; average age 25) with six devices at once (a smartwatch, two sports watches, a smart ring, a recovery strap and a forehead EEG headband) plus PSG, for one night with nine hours in bed.

For sleep versus wake:

  • Agreement with PSG was 86 to 89% across the six devices.
  • Kappa ranged from 0.30 to 0.51, fair to moderate.
  • The authors concluded all six were valid for the timing and duration of sleep in the field.

For sleep stages (wake, light, deep, REM):

  • The five wrist, ring and strap devices agreed with PSG 50 to 61% of the time (kappa 0.20 to 0.44).
  • The forehead EEG device, which measures brain activity, reached 65% (kappa 0.52).
  • The authors concluded that all six "require improvement for the assessment of specific sleep stages".

Inference: if roughly 40 to 50% of epochs are labelled with the wrong stage, a nightly deep-sleep or REM figure from a wrist, ring or strap device is a rough estimate at best. A change of 15 minutes from one night to the next is well within what measurement error alone could produce.

Young athletes and a wrist tracker (2018)

Athletes were the focus of a 2018 study by Sargent and colleagues in Chronobiology International. Well-trained young athletes wore a commercial wrist-worn tracker (2017 model) during PSG-recorded night-time sleep and daytime naps. The abstract reports sleep periods rather than the number of athletes or the sex split.

  • The tracker automatically detected 24 of 30 night-time sleep periods but only 6 of 20 daytime naps.
  • It overestimated night-time total sleep by an average of 52 minutes, with a very wide spread (standard deviation 152 minutes).
  • For naps, the overestimate was small (4 minutes on average).

That device generation is now old. The lesson that dates less quickly is that automatic detection can miss sleep periods entirely, especially naps, and that the error on any single night can be large even if the average looks acceptable.

Eleven trackers, 75 participants (2023)

The largest comparison here is a 2023 multicentre study by Lee and colleagues in JMIR mHealth and uHealth. It tested 11 consumer trackers (five wearables, three bedside "nearables" and three smartphone-based "airables") against PSG in 75 participants recruited from a hospital and a sleep clinic in Korea.

  • Sleep-stage classification varied hugely between devices, with a summary score (macro F1, where 1 is perfect) ranging from 0.26 to 0.69.
  • Wearables showed a high proportional bias in sleep efficiency, meaning their error changed with how well or badly the person slept.
  • Performance varied with body mass index, sleep efficiency and sleep apnoea severity, while differences between men and women were minimal.

This group included people being assessed for sleep problems, not athletes, which is a different population. But it reinforces two points: devices differ a lot, and accuracy drops for the nights and people where you'd most want an accurate answer.

What the trackers get right and wrong

Pulling the studies together, the pattern is consistent across device types and years:

  • When you fell asleep and woke up: usually good on a normal night, especially if you check that the device detected the sleep period correctly.
  • Total sleep time: fair, with a tendency to overestimate because quiet wakefulness is read as sleep.
  • Time awake during the night: poor. Specificity of 0.18 to 0.54 in the 2021 study means many minutes of wake go uncounted.
  • Light, deep and REM sleep: poor to moderate for wrist, ring and strap devices. Devices that measure brain activity do better.
  • Naps: often missed unless you log them yourself.
  • Disrupted nights: less accurate, which is exactly when athletes look hardest at the data.

The composite "sleep score" or "recovery score" that many devices show is built from these inputs plus proprietary weighting. No independent study has validated a composite sleep score against a training or performance outcome in triathletes.

Population and sex splits

  • Ages: the studies are mostly in adults in their 20s. Sleep architecture changes with age, so accuracy in masters triathletes is less certain.
  • Sex: the 2021 and 2022 studies included roughly half or more women, and the 2023 study found minimal differences in performance between men and women. That's reassuring, but none of these looked at menstrual-cycle changes in skin temperature or heart rate, which some devices use.
  • Athletes: only the 2018 study was in athletes. Athletes have low resting heart rates and high heart rate variability, which the algorithms may not have been trained on. Inference: algorithms that lean on heart rate might be less accurate in very fit people, but this hasn't been tested directly.
  • Triathletes: no published validation study was found in triathletes specifically.

When the score becomes the problem

Accuracy isn't the only issue. The information can change how you feel.

Sham feedback changes next-day symptoms

A 2018 randomised study by Gavriloff and colleagues in the Journal of Sleep Research recruited 63 adults with insomnia disorder from the community. Each morning they received sleep feedback from a wrist device, but the feedback was fake: half were told they had slept well and half that they had slept badly, regardless of how they had actually slept.

  • The negative-feedback group (n = 32) reported lower alertness (effect size d = 0.79) and more sleepiness and fatigue (d = 0.55) in the evening than the positive-feedback group (n = 31).
  • Mood and alertness improved more through the morning in the positive-feedback group.
  • There was no difference on a reaction-time test (d = 0.12) or a test of attention to sleep-related words (d = 0.20).

These were people with insomnia, not athletes, and the feedback was deliberately false. But the finding is direct: being told you slept badly made people feel worse, without any measurable change in performance.

"Orthosomnia"

In 2017, Baron and colleagues described a case series in the Journal of Clinical Sleep Medicine of patients who came to a sleep clinic worried about light or restless sleep shown by their trackers. They coined the term orthosomnia for a perfectionistic quest for ideal sleep data. The patients often trusted the tracker more than validated clinical tests, and the effort to improve the numbers fed the worry that was keeping them awake.

A case series can't tell you how common this is. Inference: athletes who already monitor heart rate, power and training load may be well placed to add sleep data to the list of numbers they worry about. TriForward's "Tracking Everything: Does More Data Help or Hurt Motivation?" looks at that wider question.

Practical application

None of this has been tested as a package in triathletes. It's built from the studies above, and the specific steps are inference.

Use it for what it measures well

  1. Look at timing and consistency. When you went to bed, when you got up and how much that varies across the week are the most accurate outputs and the easiest to act on.
  2. Look at weekly averages, not single nights. A seven-night average of total sleep smooths out the night-to-night error. Compare it with your training log: does sleep shrink in big training weeks, after late sessions or when you travel?
  3. Treat stage figures as rough. If you want to watch deep or REM sleep at all, look at a monthly trend, never one night.

Check your own device

You can't run a sleep lab at home, but you can sanity-check the output:

  • Keep a short sleep diary for two weeks: lights out, roughly how long it took to fall asleep, any long wake-ups, time up. Compare it with the device. If the device regularly shows you sleeping through periods you clearly remember being awake, it's overestimating sleep for you.
  • Check detection. Does it pick up naps, early nights and lie-ins correctly, or does it need you to edit the sleep window? The 2018 athlete study found missed sleep periods were common.
  • Note firmware updates. If your numbers jump after an update with no change in your life, the update is the likely cause.
  • Wear it consistently, on the same wrist or finger and with a similar fit, so changes reflect you rather than the sensor.

Decide how much weight to give it

  • Let how you feel lead. A recovery score is one input. Combine it with your mood, muscle soreness, how a warm-up feels and the previous days' training. TriForward's "Reading the Trend, Not the Number" makes the same case for heart rate variability.
  • Don't cancel a key session on a score alone. If you feel fine and the warm-up goes normally, a low score after an ordinary night is weak evidence of anything.
  • Don't check the score before you've noticed how you feel. Inference from the sham-feedback study: forming your own view first may stop the number from setting your mood for the day.
  • Turn off sleep alerts or the tracker for a while if bedtime is becoming a performance.

A triathlon scenario

The following example is illustrative, not drawn from a study. A 41-year-old age-group triathlete is eight weeks out from a middle-distance race. Her ring tells her she's averaged 38 minutes of deep sleep this week against "her normal" of 55, and her recovery score has been amber three mornings running. She feels fine and her sessions have gone well. She's considering dropping Saturday's long ride.

Using the evidence above:

  • She looks at what the device measures best. Her bedtime has drifted later on three evenings because of work, and her weekly average total sleep is down by about 40 minutes a night. That's worth acting on.
  • She discounts the deep-sleep figure. Ring and wrist devices agreed with lab staging only 50 to 61% of the time, so a 17-minute change is within plausible error.
  • She checks how she feels. Legs fine, mood fine, last two sessions on target.
  • She keeps the ride but protects sleep. She goes to bed 30 minutes earlier on Thursday and Friday and stops checking the score first thing in the morning.

If she had felt flat, sore and irritable as well, the same sleep data would have added weight to an easier day. The data supports a decision; it doesn't make it.

Common mistakes

  • Treating deep and REM figures as measurements. They are estimates with large error on wrist, ring and strap devices.
  • Reacting to one night. Night-to-night error is large; trends over a week or more are more useful.
  • Trusting the device more than your own experience. If you lay awake for an hour, you lay awake for an hour, whatever the app says.
  • Comparing numbers between different devices or with a friend's. Algorithms differ, so the same night can produce different stage figures on different devices.
  • Assuming the device spots sleep problems. A consumer tracker isn't a diagnostic test for insomnia or sleep apnoea (see the safety notes).
  • Chasing a better score with pills or supplements. None of the studies here support that, and it isn't a safe way to manage sleep.

What we don't know

  • Whether wearing a sleep tracker improves training or racing. No trial has tested this in triathletes or other endurance athletes.
  • How accurate current devices are in triathletes. Only one study here was in athletes, and it used a 2017-era wrist device.
  • How firmware updates change accuracy. Validation studies test one software version; manufacturers update far more often than studies are published.
  • Older athletes and women across the menstrual cycle. Most validation samples are in their 20s, and cycle-related changes in temperature and heart rate haven't been studied in this context.
  • Whether composite sleep or recovery scores mean anything for performance. They haven't been independently validated against training outcomes.
  • How often athletes experience "orthosomnia". The evidence is a case series and one experiment in people with insomnia.

References

  • de Zambotti M, Cellini N, Goldstone A, Colrain IM, Baker FC. Wearable sleep technology in clinical and research settings. Medicine & Science in Sports & Exercise, 2019;51(7):1538–1557. doi:10.1249/MSS.0000000000001947
  • Chinoy ED, Cuellar JA, Huwa KE, Jameson JT, Watson CH, Bessman SC, Hirsch DA, Cooper AD, Drummond SPA, Markwald RR. Performance of seven consumer sleep-tracking devices compared with polysomnography. Sleep, 2021;44(5):zsaa291. doi:10.1093/sleep/zsaa291
  • Miller DJ, Sargent C, Roach GD. A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults. Sensors, 2022;22(16):6317. doi:10.3390/s22166317
  • Sargent C, Lastella M, Romyn G, Versey N, Miller DJ, Roach GD. How well does a commercially available wearable device measure sleep in young athletes? Chronobiology International, 2018;35(6):754–758. doi:10.1080/07420528.2018.1466800
  • Lee T, Cho Y, Cha KS, Jung J, Cho J, Kim H, Kim D, Hong J, Lee D, Keum M, Kushida CA, Yoon IY, Kim JW. Accuracy of 11 wearable, nearable, and airable consumer sleep trackers: prospective multicenter validation study. JMIR mHealth and uHealth, 2023;11:e50983. doi:10.2196/50983
  • Gavriloff D, Sheaves B, Juss A, Espie CA, Miller CB, Kyle SD. Sham sleep feedback delivered via actigraphy biases daytime symptom reports in people with insomnia: implications for insomnia disorder and wearable devices. Journal of Sleep Research, 2018;27(6):e12726. doi:10.1111/jsr.12726
  • Baron KG, Abbott S, Jao N, Manalo N, Mullen R. Orthosomnia: are some patients taking the quantified self too far? Journal of Clinical Sleep Medicine, 2017;13(2):351–354. doi:10.5664/jcsm.6472

Frequently asked questions

Is my watch accurate for how long I slept?

Reasonably, on a normal night. In lab studies with sleep-lab recording as the reference, consumer devices agreed on whether each 30-second slice of the night was sleep or wake 86 to 89% of the time. They are much weaker at spotting time awake in bed, so they tend to overestimate sleep, especially on broken nights. One older wrist tracker overestimated night-time sleep in young athletes by 52 minutes on average.

Can I trust the deep sleep and REM numbers?

Not as precise measurements. In a 2022 study of 53 adults, five wrist, ring and strap devices matched sleep-lab staging only 50 to 61% of the time. Stage numbers can drift with firmware updates too. Treat them as a rough indication at most and don't change your training on one night's deep-sleep figure.

Should I train less when my sleep score is low?

Not on the score alone. Use it alongside how you actually feel, how the previous days went and how a warm-up feels. A low score after a night you felt fine about may be a measurement problem, and a randomised study found that being told you slept badly can itself make you feel worse.

What's the most useful thing a sleep tracker can do for a triathlete?

Show patterns over weeks: what time you actually go to bed, how consistent that is, and whether hard training blocks, late sessions, travel or alcohol shorten your nights. Those are the measures the devices get most right, and they point to habits you can change.

The evidence behind this guide

1 graded principle, 7 sources. Open one to see the sources.

Consumer sleep trackers are fair for sleep timing and duration but poor at detecting wake and classifying sleep stagesModerate evidence7 sources

In practice: Use a sleep tracker for bedtime, wake time, consistency and weekly average sleep, and treat nightly deep-sleep, REM and composite scores as rough estimates that shouldn't decide training on their own. Check your own device against a two-week sleep diary and after firmware updates. If the score is driving worry about sleep, take a break from it. No study shows that sleep tracking improves triathlon performance, and transfer to athletes and masters age groups is inference.

Peer-reviewed studyNarrative reviewIndependently verified
Wearable sleep technology in clinical and research settings ↗
de Zambotti M, Cellini N, Goldstone A, Colrain IM, Baker FC · 2019 · Medicine & Science in Sports & Exercise

State-of-the-art review of consumer wearable sleep trackers in research and clinical use. Highlights that proprietary algorithms, device malfunction and firmware updates must be considered before relying on device output, and calls for standardised validation. Narrative review, no pooled accuracy estimate.

Peer-reviewed studyCross-sectionalIndependently verified
Performance of seven consumer sleep-tracking devices compared with polysomnography ↗
Chinoy ED, Cuellar JA, Huwa KE, Jameson JT, Watson CH, Bessman SC, Hirsch DA, Cooper AD, Drummond SPA, Markwald RR · 2021 · Sleep
34 healthy young adults (22 F), 28.1 ± 3.9 y, three laboratory nights incl. one disrupted night

Four wearables (two wrist sports watches, 2017-era) and three bedside devices tested with research actigraphy against polysomnography. Epoch sensitivity for sleep was high (all ≥0.93) but specificity for wake was low to medium (0.18-0.54); sleep-stage comparisons were mixed and devices performed worse on disrupted nights. Most devices matched or beat actigraphy for sleep/wake; the two sports watches did worse. Non-athletes; specific device generations.

Peer-reviewed studyCross-sectionalIndependently verified
A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults ↗
Miller DJ, Sargent C, Roach GD · 2022 · Sensors
53 healthy adults (26 F, 27 M), 25.4 ± 5.9 y, one laboratory night

Six devices worn simultaneously against polysomnography. Sleep/wake agreement 86-89% (kappa 0.30-0.51) across devices, judged valid for sleep timing and duration. Sleep-stage agreement only 50-61% (kappa 0.20-0.44) for the five wrist, ring and strap devices and 65% (kappa 0.52) for a forehead EEG device; all needed improvement for sleep stages. Single night, young non-athlete adults.

Peer-reviewed studyCross-sectionalIndependently verified
How well does a commercially available wearable device measure sleep in young athletes? ↗
Sargent C, Lastella M, Romyn G, Versey N, Miller DJ, Roach GD · 2018 · Chronobiology International
Well-trained young athletes (number and sex split not in abstract); 30 night-time sleep periods and 20 daytime naps

A wrist-worn consumer tracker (2017-era model) auto-detected 24 of 30 night-time sleep periods and 6 of 20 daytime naps, and overestimated night-time total sleep by 52 ± 152 min (naps by 4 ± 8 min) compared with polysomnography. Old device generation; small sample.

Peer-reviewed studyCross-sectionalIndependently verified
Accuracy of 11 wearable, nearable, and airable consumer sleep trackers: prospective multicenter validation study ↗
Lee T, Cho Y, Cha KS, Jung J, Cho J, Kim H, Kim D, Hong J, Lee D, Keum M, Kushida CA, Yoon IY, Kim JW · 2023 · JMIR mHealth and uHealth
75 participants from a tertiary hospital and a sleep clinic in Korea (sex split not in abstract)

Eleven consumer trackers (5 wearables, 3 bedside, 3 smartphone-based) against polysomnography. Sleep-stage classification varied widely (macro F1 0.26-0.69); wearables showed high proportional bias in sleep efficiency; accuracy varied with BMI, sleep efficiency and apnoea severity, with minimal male/female difference. Clinical sleep population, not athletes.

Peer-reviewed studyRCTIndependently verified
Sham sleep feedback delivered via actigraphy biases daytime symptom reports in people with insomnia: implications for insomnia disorder and wearable devices ↗
Gavriloff D, Sheaves B, Juss A, Espie CA, Miller CB, Kyle SD · 2018 · Journal of Sleep Research
63 community adults with DSM-5 insomnia disorder (sex split not in abstract), randomised to negative (n = 32) or positive (n = 31) sham feedback

Sham next-morning feedback from a wrist device: the negative-feedback group reported lower alert cognition (d = 0.79) and more sleepiness/fatigue (d = 0.55) by evening than the positive-feedback group, with greater morning gains in mood and alertness after positive feedback. No differences in sleep-related attentional bias (d = 0.20) or psychomotor vigilance (d = 0.12). People with insomnia, not athletes; feedback deliberately false.

Peer-reviewed studyCase studyIndependently verified
Orthosomnia: are some patients taking the quantified self too far? ↗
Baron KG, Abbott S, Jao N, Manalo N, Mullen R · 2017 · Journal of Clinical Sleep Medicine

Case series of sleep-clinic patients seeking treatment for self-diagnosed sleep problems based on tracker data; coins 'orthosomnia' for a perfectionistic quest for ideal sleep data. Patients often trusted the tracker more than polysomnography or actigraphy. Descriptive only; cannot show how common the problem is.

Full write-up of this principle

Get The Forward

Research-led triathlon guidance, straight to your inbox. No spam, no filler — unsubscribe any time.