Sleep tracking is the overnight measurement of when you sleep, how long you sleep and how often you wake, estimated by a wearable or bedside device from body movement, heart rate and other physiological signals. A consumer tracker cannot record the brain waves that define sleep, so it infers sleep from its effects on your body, and it measures total sleep time far more accurately than it measures sleep stages.
Your watch answers one question well, how long you slept. It answers a second question, when you were awake, only moderately well. The question most sleep reports put front and center, how many minutes of deep and REM sleep you got, is the one it answers worst. Knowing which number rests on which evidence lets you act on the data without chasing noise. The wearable data quality guide sets sleep data beside the other metrics a wearable reports.
01What a sleep tracker reports
Most trackers report the same handful of measures, borrowed from clinical sleep medicine. The right column summarizes the validation evidence covered later in this entry.
| Measure | What it means | How well wrist devices measure it |
|---|---|---|
| Time in bed | From settling in to getting up | Depends on detecting when you lay down |
| Sleep onset latency | Minutes from lights out to falling asleep | Fair, since lying still reads as sleep |
| Total sleep time | Minutes asleep | The strongest measure, off by about 17 min on average |
| Wake after sleep onset (WASO) | Minutes awake between first falling asleep and finally waking | Undercounted by most devices |
| Sleep efficiency | Total sleep time as a share of time in bed | Inflated when wake is missed |
| Sleep stages | Minutes of light (or core), deep and REM sleep | Fair to moderate agreement with a sleep lab |
| Regularity | How consistent your sleep and wake times are from night to night | Good, since timing depends on detecting sleep itself |
02What polysomnography measures
Polysomnography (PSG) is the overnight lab test that defines sleep. It records brain activity with electroencephalography (EEG), eye movements with electrooculography (EOG) and chin muscle tone with electromyography (EMG). Those three signals together determine whether you are asleep or awake (Ancoli-Israel et al., 2003). A clinical study adds an electrocardiogram, pulse oximetry, airflow and leg movement sensors to catch breathing and movement disorders (Apple, 2025).
A technologist labels each 30 second epoch under American Academy of Sleep Medicine rules as wake, REM or one of three non-REM stages. N1 is the brief transition into sleep, N2 makes up most of non-REM sleep, and N3 is deep, or slow-wave, sleep (Apple, 2025). The sleep stages entry covers what each stage does for you.
PSG is the reference, and it carries its own error. In the Academy's inter-scorer reliability program, more than 2,500 scorers labeled the same 1,800 epochs and agreed with the majority score 82.6% of the time. Agreement fell to 67.4% for N3 and 63.0% for N1, and nearly every scorer who disagreed on an N3 epoch called it N2 (Rosenberg and Van Hout, 2013). When a wearable struggles to separate deep sleep from N2, it is failing at the same boundary that trained humans reading brain waves find hardest.
03How a wearable infers sleep
A wearable sees the consequences of sleep in your body. Each sensor adds a different piece of evidence, and the evidence for stages is much thinner than the evidence for sleep itself.
Movement
Wrist actigraphy, the oldest method, counts arm movements and calls long still stretches sleep. Cole and colleagues (1992) built one of the standard scoring algorithms from 41 people wearing a wrist actigraph during PSG, and it separated sleep from wake about 88% of the time. Actigraphy records one kind of signal where PSG combines three, and a review of the method judged it doubtful that a motion record will ever carry the same information (Ancoli-Israel et al., 2003). If you lie awake and still, your wrist looks asleep.
Current devices sample the accelerometer fast enough to see more than arm movement. Apple's staging model reads high-frequency 3-axis accelerometer data, which captures the small movements of breathing, and assigns each 30 second epoch to Awake, REM, Core or Deep (Apple, 2025). Apple's Core corresponds to N1 plus N2, and Deep corresponds to N3. Apple's paper names accelerometer data as the model's input. Heart rate is recorded separately.
Heart rate and HRV
Your autonomic nervous system changes state as you sleep. During non-REM sleep the balance shifts toward parasympathetic activity, heart rate falls and heart rate variability rises. During REM, autonomic balance looks much like wakefulness (Trinder et al., 2001). The optical sensor on the back of a watch or inside a ring, called photoplethysmography, tracks those changes beat by beat. The HRV and resting heart rate entries cover what those signals mean on their own.
Cardiac data helps most with stages. In an Oura ring dataset of 106 people and 440 nights, accelerometer data alone classified sleep versus wake with 94% accuracy and four stages with 57% accuracy. Adding HRV raised four-stage accuracy to 76%, the largest gain from any sensor, and adding circadian timing features brought it to 79% (Altini and Kinnunen, 2021). Walch and colleagues (2019) combined raw Apple Watch motion and heart rate data in 31 adults and reached 90% overall accuracy for sleep versus wake, with 60% of true wake epochs correctly identified.
Temperature
Falling asleep and a drop in core body temperature happen together, and Harding and colleagues (2019) argue that non-REM sleep is most likely to begin when core temperature is falling fastest. In the Oura dataset, skin temperature added one percentage point to sleep versus wake accuracy and three points to four-stage accuracy (Altini and Kinnunen, 2021).
Temperature earns its place as a nightly reading. Apple Watch Series 8 and later sample wrist temperature every five seconds while you sleep and set a personal baseline after about five nights (Apple Support). The wrist temperature entry covers how to read it.
Breathing and blood oxygen
Breathing patterns change between sleep stages, and REM has a respiration pattern distinct from non-REM sleep (Apple, 2025). Many watches and rings also record respiratory rate and blood oxygen overnight as separate measurements, which help you spot a night that departs from your normal. None of them can diagnose sleep apnea. That diagnosis depends on airflow and breathing-effort channels a wrist device lacks, and actigraphy has long been judged unsuitable for it (Ancoli-Israel et al., 2003).
04How accurate consumer sleep tracking is
Sleep versus wake
The validation studies below all find the same asymmetry. Devices detect sleep well and wake poorly. Chinoy and colleagues (2021) tested seven consumer devices against PSG in 34 healthy young adults across three lab nights. Sensitivity for sleep was 0.93 or higher on every device, specificity for wake ranged from 0.18 to 0.54, and devices tended to perform worse on nights with poorer or disrupted sleep.
Schyvens and colleagues (2025) put six wrist devices on 62 adults for one lab night. All six detected more than 90% of sleep epochs, and wake detection ranged from 29.4% to 52.2%. The Apple Watch Series 8, worn by 20 of the participants, had the best wake detection and overestimated total sleep time by 19.6 minutes on average. Its limits of agreement ran from 38 minutes under to 78 minutes over, so a single night can miss by far more than the average.
A meta-analysis of 24 studies and 798 participants found that wrist devices as a class differed from PSG by about 17 minutes of total sleep time, 4.7 percentage points of sleep efficiency and 13 minutes of wake after sleep onset (Lee et al., 2025).
Sleep stages
Stage accuracy is usually reported as Cohen's kappa, which scores agreement beyond chance from 0 (chance) to 1 (perfect). The six devices in Schyvens (2025) ranged from 0.21 to 0.53, fair to moderate agreement. Apple reports a median kappa of 0.63 for its original staging model on its own held-out validation set, rising to 0.68 after the 2025 update, and 0.66 for the update in a separate group of patients undergoing clinical PSG (Apple, 2025). Those are manufacturer figures. The independent studies tested older firmware.
Deep sleep errors run in both directions. Against PSG in Schyvens (2025), the Apple Watch undercounted deep sleep by 25.2 minutes, WHOOP 4.0 overcounted it by 31.5 minutes, and the Fitbit Sense and Charge 5 landed within 4 minutes. The first-generation Oura algorithm underestimated N3 by about 20 minutes and overestimated REM by about 17 minutes (de Zambotti et al., 2019). Deep sleep is a small share of the night, 12.8% of sleep in Apple's validation set, so a small number of misclassified epochs moves it by a large percentage.
| Device and algorithm | Study | People | True sleep detected | True wake detected | Stage kappa | Total sleep bias |
|---|---|---|---|---|---|---|
| Apple Watch, watchOS 26 model | Apple 2025, manufacturer | 166 | 96.8% | 78.9% | 0.68 | Not reported |
| Apple Watch Series 8 | Schyvens 2025 | 20 | 96.3% | 52.2% | 0.53 | +19.6 min |
| Fitbit Sense | Schyvens 2025 | 37 | 93.3% | 48.8% | 0.42 | +6.3 min |
| WHOOP 4.0 | Schyvens 2025 | 40 | 93.6% | 40.1% | 0.37 | +24.5 min |
| Garmin Vivosmart 4 | Schyvens 2025 | 25 | 95.9% | 29.4% | 0.21 | +38.4 min |
| Oura, first-generation | de Zambotti 2019 | 41 | 96% | 48% | Not reported | No significant difference |
Each row comes from its own sample and protocol. Apple's figures are per-night medians from its own validation set, and the Schyvens figures are group means from one lab night per person. Read the table for the pattern and avoid ranking devices from single cells.
Where trackers go wrong
The validation literature points to a short list of errors you will recognize on your own report.
- Quiet wake counted as sleep. Reading in bed or lying still after the alarm reads as light sleep. Apple built its 2025 model update around this error and reports correct wake classification rising from 70% to 79% (Apple, 2025).
- Short awakenings that disappear. All six devices in Schyvens (2025) undercounted wake after sleep onset, by 12 to 48 minutes on average, and devices do worse on broken nights (Chinoy et al., 2021). A night that felt broken can show as unbroken.
- Deep and REM minutes that move without you. Four of the six devices in Schyvens (2025) missed average deep sleep by 25 minutes or more, some over and some under.
- Naps and short sessions without stages. Apple Watch writes plain asleep and awake data for a detected session of one to three hours and writes full stages only past three hours (Apple, 2025).
- Firmware changes. A new algorithm can shift your numbers overnight. Apple's 2025 update improved classification for three of the four stages from the same wrist data (Apple, 2025), so your stage mix can change with no change in your sleep.
05Normal values and what moves them
The American Academy of Sleep Medicine and the Sleep Research Society recommend that adults sleep 7 or more hours per night on a regular basis. They add that more than 9 hours may be appropriate for young adults, people recovering from sleep debt and people who are ill (Watson et al., 2015). If you train hard, the sleep for athletes guide covers why your need sits toward the top of that range.
For a sense of a measured night, the 166 adults in Apple's validation set, average age 47, slept 378 minutes by PSG with 85.4% sleep efficiency and 35.5 minutes of wake after sleep onset. Their sleep split into 65.9% Core, 21.4% REM and 12.8% Deep (Apple, 2025). That group spent at least one night in a lab or with home PSG equipment, and some had sleep apnea, so treat those values as a description of real adults and avoid using them as targets.
Age changes the shape of the night. Across 65 studies and 3,577 healthy people, total sleep time, sleep efficiency, the share of slow-wave sleep and the share of REM all fell with age in adults, while sleep onset latency, the share of N1 and N2, and wake after sleep onset rose (Ohayon et al., 2004). If you are 55 and your deep sleep is lower than it was at 25, age explains at least part of the drop.
Timing matters alongside duration. In 60,977 UK Biobank participants with accelerometer data, people in the four more regular quintiles of sleep timing had a 20% to 48% lower risk of death from any cause than the least regular quintile, and regularity predicted mortality better than duration (Windred et al., 2024). The circadian rhythm entry covers why a stable wake time matters, and sleep hygiene covers the habits that protect it.
06How Titan detects your sleep
Titan reads sleep from Apple Health and builds one sleep session for each day. The rules below decide which night your Sleep and Recovery numbers describe. The how sleep is detected article covers them with examples.
Titan sorts every sleep sample by start time. A sample that starts within 2 hours of the latest end time joins the current session, and a longer gap starts a new session. Each session belongs to the day it ends, in your phone's local time, and it must end after midnight and no later than 12:00 PM. Sleep that ends after noon counts for no day, which affects daytime sleep after a night shift. When two sessions end on the same morning, Titan keeps the one with the most time asleep, counting Core, Deep, REM and unstaged Asleep time.
When more than one tracker writes sleep, Titan cuts the night into slices at every sample boundary and keeps one stage per slice. Awake outranks REM, REM outranks Deep, Deep outranks Core, and Core outranks unstaged Asleep. Since wrist devices miss wake more than anything else, that order keeps any wake-up either tracker caught.
Choosing which tracker Titan reads
The Sleep Tracker setting, under Settings > Recovery & Sleep, lists the apps and devices that have saved sleep to Apple Health. Titan builds each night from your chosen tracker first. On a day that tracker recorded no session, Titan falls back to your other sleep sources, so a night your ring spent on the charger can still use Apple Watch sleep. If you set allowed sources for Sleep under Settings > Health & Connected Apps, that setting wins, Sleep Tracker is ignored, and there is no fallback. The sleep source article covers each combination.
Pick the device you wear most consistently. Some sources write sleep without stages, including Titan's direct Oura connection, which saves each night as a single asleep block.
How sleep feeds Recovery and the Sleep score
Recovery reads the HRV and heart rate recorded inside the sleep session Titan picked and compares them with your own baseline of measured nights. HRV uses SDNN by default, and you can switch to RMSSD in Recovery & Sleep settings. If no session exists for the day, Titan reads HRV and heart rate between 10:00 PM and 10:00 AM and marks the score Partial night. A night that splits into two sessions or ends after noon therefore changes your Recovery window along with your sleep total.
The Sleep score gives its largest share, 44%, to time asleep against your sleep target, the measure trackers get most right. Restorative sleep, Deep plus REM as a share of time asleep, carries 35%, and awake time carries 21%. Restorative sleep comes from stage labels, so it needs a source that records Core, Deep and REM.
07How to read your own sleep data
Average across weeks. Night-to-night variation is large. In a study of 54 working adults, a reliable actigraphy estimate of total sleep time needed more than 7 nights, and sleep efficiency needed at least 5 (Aili et al., 2017). Judge a change in your sleep from a weekly or monthly average against your own baseline.
Compare a device with itself. Each device has its own bias. In the same lab, the Fitbit Sense overcounted total sleep by 6 minutes and the Garmin Vivosmart 4 by 38 (Schyvens et al., 2025). Switch between them and your recorded sleep moves by half an hour with no change in how you slept. Part of a device's error repeats every night, so a trend on one device is more trustworthy than a single night.
Trust duration first and stages last. Total sleep time and regularity rest on the strongest evidence. Wake time comes next, with the caveat that it is undercounted. Treat a single night's deep or REM minutes as a rough estimate, and pay attention only when a change holds for two weeks or more. The Apple Watch sleep accuracy article covers the stage errors in detail.
Check the night against how you feel. If you lay awake for an hour and the report shows solid sleep, believe your memory. Wrist devices are weakest on exactly those nights.
Keep the data in its place. Clinicians describe patients who chase perfect tracker numbers and lose sleep over it, a pattern named orthosomnia (Baron et al., 2017). If checking your score makes bedtime tense, check it weekly.
See a clinician for symptoms. Loud snoring, pauses in breathing that a partner notices, or daytime sleepiness despite 7 or more hours in bed call for a proper sleep evaluation. A tracker can prompt that conversation. The diagnosis needs a sleep study.
08References
- Aili K et al. (2017). Reliability of actigraphy and subjective sleep measurements in adults, the design of sleep assessments. Journal of Clinical Sleep Medicine 13(1):39-47. https://doi.org/10.5664/jcsm.6384
- Altini M, Kinnunen H (2021). The promise of sleep, a multi-sensor approach for accurate sleep stage detection using the Oura ring. Sensors 21(13):4302. https://doi.org/10.3390/s21134302
- Ancoli-Israel S et al. (2003). The role of actigraphy in the study of sleep and circadian rhythms. Sleep 26(3):342-392. https://doi.org/10.1093/sleep/26.3.342
- Apple Inc. (2025). Estimating sleep stages from Apple Watch, updated October 2025. Apple technical paper. https://www.apple.com/health/pdf/EstimatingSleepStagesfromAppleWatchOct_2025.pdf
- Apple Support. Track your nightly wrist temperature changes with Apple Watch. https://support.apple.com/en-us/102674
- Baron KG et al. (2017). Orthosomnia, are some patients taking the quantified self too far? Journal of Clinical Sleep Medicine 13(2):351-354. https://doi.org/10.5664/jcsm.6472
- Chinoy ED et al. (2021). Performance of seven consumer sleep-tracking devices compared with polysomnography. Sleep 44(5):zsaa291. https://doi.org/10.1093/sleep/zsaa291
- Cole RJ et al. (1992). Automatic sleep/wake identification from wrist activity. Sleep 15(5):461-469. https://doi.org/10.1093/sleep/15.5.461
- de Zambotti M et al. (2019). The sleep of the ring, comparison of the ŌURA sleep tracker against polysomnography. Behavioral Sleep Medicine 17(2):124-136. https://doi.org/10.1080/15402002.2017.1300587
- Harding EC et al. (2019). The temperature dependence of sleep. Frontiers in Neuroscience 13:336. https://doi.org/10.3389/fnins.2019.00336
- Lee YJ et al. (2025). Performance of consumer wrist-worn sleep tracking devices compared to polysomnography, a meta-analysis. Journal of Clinical Sleep Medicine 21(3):573-582. https://doi.org/10.5664/jcsm.11460
- Ohayon MM et al. (2004). Meta-analysis of quantitative sleep parameters from childhood to old age in healthy individuals, developing normative sleep values across the human lifespan. Sleep 27(7):1255-1273. https://doi.org/10.1093/sleep/27.7.1255
- Rosenberg RS, Van Hout S (2013). The American Academy of Sleep Medicine inter-scorer reliability program, sleep stage scoring. Journal of Clinical Sleep Medicine 9(1):81-87. https://doi.org/10.5664/jcsm.2350
- Schyvens AM et al. (2025). A performance validation of six commercial wrist-worn wearable sleep-tracking devices for sleep stage scoring compared to polysomnography. Sleep Advances 6(2):zpaf021. https://doi.org/10.1093/sleepadvances/zpaf021
- Trinder J et al. (2001). Autonomic activity during human sleep as a function of time and sleep stage. Journal of Sleep Research 10(4):253-264. https://doi.org/10.1046/j.1365-2869.2001.00263.x
- Walch O et al. (2019). Sleep stage prediction with raw acceleration and photoplethysmography heart rate data derived from a consumer wearable device. Sleep 42(12):zsz180. https://doi.org/10.1093/sleep/zsz180
- Watson NF et al. (2015). Recommended amount of sleep for a healthy adult, a joint consensus statement of the American Academy of Sleep Medicine and Sleep Research Society. Sleep 38(6):843-844. https://doi.org/10.5665/sleep.4716
- Windred DP et al. (2024). Sleep regularity is a stronger predictor of mortality risk than sleep duration, a prospective cohort study. Sleep 47(1):zsad253. https://doi.org/10.1093/sleep/zsad253
