Wearable devices are sensors you wear on the body, usually a watch, a ring or a chest strap, that record your heart activity and movement and sometimes your skin temperature and blood oxygen. Wearable metrics are the numbers built from those recordings, from a single heart rate reading to a nightly recovery score.
Every number on your wrist sits somewhere between a measurement and an estimate. Resting heart rate from a well-fitted device is close to a measurement. Sleep stages, calories and VO2 max come from models that infer one quantity from another. Recovery, readiness and strain scores are composites that combine several estimates with weights chosen by the company that built them. Knowing which kind of number you are looking at tells you how much to trust it and how to act on it.
The evidence base is thinner than the marketing suggests. Doherty and colleagues' 2024 living umbrella review in Sports Medicine gathered 24 systematic reviews covering 249 validation studies and 430,465 participants. About 11% of the consumer wearables released to date had been validated for even one measure, and the validation studies done so far represent 3.5% of the number needed to evaluate every device on every metric it reports.
01What the sensors record
Most wrist devices measure heart rate with photoplethysmography (PPG), an optical method. Apple describes its version plainly. Blood absorbs green light, so the watch flashes green LEDs "hundreds of times per second," and photodiodes measure how much light comes back as the blood volume in your wrist rises and falls with each beat. Apple Watch uses infrared LEDs during your sleep schedule. Apple Watch Series 4 and later also carry electrodes in the Digital Crown and the back of the case that record the heart's electrical signal when you touch the crown, which Apple says gives "a measurement every second instead of every 5 seconds." Chest straps record that electrical signal continuously through electrodes pressed against your skin.
Accelerometers and gyroscopes record movement, which devices use for steps, activity type and part of their sleep detection. Some devices add a skin temperature sensor, covered in wrist temperature, and red and infrared light for blood oxygen.
02Three kinds of metric
| Kind | Examples | How it is made | How to read it |
|---|---|---|---|
| Measured signals | Heart rate, beat-to-beat intervals, movement | Taken from the sensor after filtering | Trust at rest, check during jerky or arm-heavy exercise |
| Derived estimates | HRV, resting heart rate, sleep stages, steps, calories, VO2 max, respiratory rate | A model turns signals into a physiological value | Compare with your own history from the same device |
| Composite scores | Recovery, readiness, strain, sleep score, stress | Several estimates weighted into one number | Look at the inputs before acting on the score |
A composite score has no reference standard. A lab can hold a watch's heart rate against an ECG or its sleep staging against polysomnography, and no instrument measures "readiness" directly. You can judge a score by its inputs, by how clearly the company explains the weighting, and by whether it tracks the things you care about, such as how your training sessions actually go.
03How accurate the common metrics are
Heart rate
Across the reviews in Doherty's umbrella review, wearable heart rate showed a mean bias of about ±3%. Exercise type matters. Gillinov and colleagues (Medicine and Science in Sports and Exercise, 2017) tested 50 adults on a treadmill, a stationary bike and an elliptical trainer against ECG. A Polar H7 chest strap agreed best (concordance 0.996), followed by the Apple Watch (0.92), the TomTom Spark (0.83) and the Garmin Forerunner (0.81). Accuracy was worst on the elliptical with arm levers, and the authors advised using a chest strap "when accurate HR measurement is imperative." Bent and colleagues (npj Digital Medicine, 2020) found optical error during activity averaged 30% higher than at rest, with no statistically significant difference in accuracy across skin tones.
Overnight readings are much better. In a lab study of 53 healthy young adults, Miller and colleagues (Sensors, 2022) compared six devices with ECG during a night of sleep. Mean heart rate bias was 0.5 ± 2.1 bpm for the Apple Watch Series 6, −0.3 ± 1.0 bpm for WHOOP 3.0, −1.1 ± 2.2 bpm for the Polar Vantage V and 0.1 ± 4.5 bpm for the Oura Ring Gen 2, against 5.0 ± 12.8 bpm for the Garmin Forerunner 245. Sleeping heart rate from a well-fitted device is one of the most trustworthy numbers a wearable produces, which is why recovery models lean on it.
HRV
Devices compute HRV with different formulas over different recording windows. Apple Watch stores SDNN from short background readings, and many training studies use RMSSD. Even on the same formula, devices disagree. When Miller and colleagues compared overnight RMSSD with ECG, mean bias ran from −4.5 ± 3.9 ms for WHOOP 3.0 to −22.4 ± 46.9 ms for the Garmin Forerunner 245. A 40 ms reading on one device says little about a 40 ms reading on another. The HRV entry covers accuracy studies and the reasons nightly HRV is the most useful version.
Sleep
Apple's validation paper for Apple Watch sleep staging compares the watch with polysomnography. The original four-stage algorithm, released in 2022, labeled 97.9% of true sleep epochs as sleep and 75.0% of true wake epochs as wake, with a four-stage kappa of 0.63. Updates in 2024 and 2025 raised correct classification of wake from 70% to 79% and lifted kappa to 0.68. Across brands, Miller and colleagues found the six devices agreed with polysomnography on sleep versus wake 86% to 89% of the time and on specific stages 50% to 65% of the time. Chinoy and colleagues (Sleep, 2021) tested seven consumer devices and found that most matched or beat research actigraphy for sleep and wake, while their sleep stage assessments "were inconsistent." Doherty's review found that wearables tend to overestimate total sleep time, with errors typically above 10%.
Your total sleep time and timing are usable. Stage minutes are rough, and brief awakenings are the thing wrist devices miss most. The sleep stages and sleep tracking entries and the Apple Watch sleep accuracy article go further.
Calories and steps
Energy expenditure is the least reliable common metric. Doherty's review found a mean bias of −3% with individual errors ranging from −21.27% to 14.76%, step counts mostly underestimated with mean absolute percentage errors of −9% to 12%, and activity-intensity estimates off by 29% to 80% depending on intensity. O'Driscoll and colleagues' 2020 meta-analysis in the British Journal of Sports Medicine, covering 60 studies, found accuracy varied with the type of activity and improved when heart rate was combined with accelerometry. Use active calories to compare days and weeks, and treat them cautiously for precise food planning.
VO2 max
Watch estimates of VO2 max are models built on heart rate and pace. Lambe and colleagues (PLOS ONE, 2025) compared Apple Watch estimates with laboratory gas analysis in 30 adults. The watch underestimated VO2 max by 6.07 mL/kg/min on average, with a mean absolute percentage error of 13.31%. Doherty's pooled data put wearable VO2 max errors at about 15% for estimates from resting data and about 10% for estimates from exercise. The trend over months is more useful than any single value.
Blood oxygen, breathing and temperature
For blood oxygen, Doherty's review found a mean absolute difference from reference methods of up to 2.0 percentage points. Respiratory rate and wrist temperature have their own entries. Both are most useful as changes from your own normal.
Health alerts
Some wearables screen for heart rhythm problems. In the Apple Heart Study (New England Journal of Medicine, 2019), 419,297 participants enrolled over 8 months, and 2,161 of them (0.52%) received an irregular pulse notification. Among the 450 who returned analyzable ECG patches, 34% had atrial fibrillation on the patch. When a notification arrived while the patch was being worn, the positive predictive value for atrial fibrillation was 0.84. Doherty's review reports pooled arrhythmia detection sensitivity of 100% and specificity of 95%. A notification is a reason to see a clinician. An absence of alerts does not rule out a heart problem, and symptoms such as chest pain, fainting or a racing heart need medical attention whatever your device shows.
04Getting reliable numbers from your device
Fit comes first. Apple's heart rate guidance begins with "a good fit," so that the sensor stays against your skin. A loose band during sleep produces gaps and noisy readings.
Keep the recording conditions the same. Wear the same device on the same wrist, and rely on overnight values for anything you compare day to day. During sleep you hold still for hours, and meals, coffee and conversations stay out of the signal.
Stick with one device per metric. The Miller study shows each device carries its own bias, so switching from one brand to another moves your numbers even when your body has not changed. If you do switch, give your baseline a few weeks to rebuild before you read much into changes.
Watch for algorithm updates. Apple's staging updates in 2024 and 2025 raised correct wake classification from 70% to 79%, so a night recorded after the update can show more wake than an identical night recorded before it. Firmware and app updates from any brand can shift your numbers in the same way.
Judge changes against your own history. A single low HRV night or a short sleep is common and usually means little. Several nights in a row outside your normal range, especially with resting heart rate rising at the same time, deserve attention. The HRV and training readiness guide covers how to turn that into training decisions.
05How Titan uses wearable data
Titan reads your data from Apple Health, so any device or app that writes to Health can feed it, including Apple Watch and other wearables that sync there. Titan's direct Oura connection pulls Oura data and writes it into Apple Health. Under Settings > Health & Connected Apps, you can choose a primary source and allowed sources for sleep, workouts, HRV, resting heart rate, heart rate, steps, active energy, exercise minutes, stand hours and mindful minutes. Duplicate Handling sets how Titan resolves overlapping data, with Prefer Primary, Most Recent or Highest Quality. The data sources article covers setup.
Recovery scores the night you just slept. Titan takes the mean of ln(HRV) for readings from one source during your main sleep session, or from 22:00 to 10:00 when no session was saved, and your sleeping heart rate, the lowest 30-minute rolling average of heart rate in the same window. It compares each with the median of your prior measured nights within your Recovery window, 60 nights by default, in units of your own night-to-night spread. A night exactly at both baselines scores 56. HRV carries twice the weight of sleeping heart rate. A baseline needs at least 7 measured nights covering at least 25% of the window, so 15 nights on the default window. Only measured nights, with both HRV and sleeping heart rate, enter your baseline. A night without HRV gets an estimated score and stays out of it. The HRV entry and the Recovery score article cover the details.
Stress runs through the day in 5-minute bins. For each bin, Titan averages your heart rate and any HRV readings, then compares them with fixed 60-day baselines of your resting heart rate and HRV. The stress value rises as heart rate moves above its baseline and as HRV falls below its own, on a scale of 0 to 100. Bins that fall inside a workout are drawn on the chart and left out of the day score, and bins during sleep count toward sleep stress. Values up to 25 read as Low, up to 50 as Moderate, up to 75 as Elevated and above that as High. The Stress score article covers the screen.
Sleep is scored from your main sleep session. Duration against your Sleep Goal carries 44% of the score, restorative sleep 35% and time awake 21%. Restorative sleep is Deep plus REM as a share of time asleep. Respiratory rate and blood oxygen do not enter the Sleep score.
Training load counts the minutes your workouts spend in each of five heart rate zones and weights them 1 through 5, a TRIMP-style calculation. A workout without heart rate data still counts, with its duration weighted by activity type.
Because every Titan score is built from readings you can inspect in Apple Health, you can check a surprising score against its inputs. A low Recovery score after a night when your watch recorded HRV from a single brief window, or a Stress spike during a hot commute, tells you something about the data before it tells you something about your body.
06References
- Apple (n.d.). Monitor your heart rate with Apple Watch. Apple Support. https://support.apple.com/en-us/HT204666
- Apple (2025). Estimating sleep stages from Apple Watch, updated October 2025. Apple. https://www.apple.com/health/pdf/EstimatingSleepStagesfromAppleWatchOct_2025.pdf
- Bent B et al. (2020). Investigating sources of inaccuracy in wearable optical heart rate sensors. npj Digital Medicine 3:18. https://doi.org/10.1038/s41746-020-0226-6
- 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
- Doherty C et al. (2024). Keeping pace with wearables, a living umbrella review of systematic reviews evaluating the accuracy of consumer wearable technologies in health measurement. Sports Medicine 54(11):2907-2926. https://doi.org/10.1007/s40279-024-02077-2
- Gillinov S et al. (2017). Variable accuracy of wearable heart rate monitors during aerobic exercise. Medicine and Science in Sports and Exercise 49(8):1697-1703. https://doi.org/10.1249/MSS.0000000000001284
- Lambe R et al. (2025). Investigating the accuracy of Apple Watch VO2 max measurements, a validation study. PLOS ONE 20(5):e0323741. https://doi.org/10.1371/journal.pone.0323741
- Miller DJ et al. (2022). A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults. Sensors 22(16):6317. https://doi.org/10.3390/s22166317
- O'Driscoll R et al. (2020). How well do activity monitors estimate energy expenditure? A systematic review and meta-analysis of the validity of current technologies. British Journal of Sports Medicine 54(6):332-340. https://doi.org/10.1136/bjsports-2018-099643
- Perez MV et al. (2019). Large-scale assessment of a smartwatch to identify atrial fibrillation. New England Journal of Medicine 381(20):1909-1917. https://doi.org/10.1056/NEJMoa1901183
