How to track your health with the smartwatch you already own

What your watch actually measures well

Most of the basic numbers on your wrist are real. Steps, resting heart rate, and whether you were awake or asleep have all been tested against proper instruments and held up.

Steps are the most reliable metric on the device. A systematic review across 158 publications and nine commercial brands found Fitbit, Apple Watch and Samsung measured steps accurately in the lab (PMID 32897239). If your watch says you walked 10,000 steps, you probably did.

Heart rate is solid too, as long as you are not near your limit. At rest and during steady aerobic work the error is negligible: a mean difference of -0.01 bpm at rest and -0.51 bpm on a treadmill across 44 studies and 15 brands (PMID 32552580). For most of your day the number is a real reflection of your physiology.

Sleep is the nuanced one. Your watch is very good at telling you that you are asleep. Sensitivity for detecting sleep was 0.93 or higher on every device in a three-night lab study of 34 adults (PMID 33378539). It is much worse at telling you when you are awake. Specificity for wake ranged from 0.18 to 0.54, so an hour of lying there can get scored as sleep.

One caveat worth naming: the Garmin devices in that particular test performed worse than the rest of the field (PMID 33378539). If you own one, discount the sleep timing a little harder. That is a statement about two specific models in 2021, not about the brand forever.

The calorie number is the one to ignore

I would not spend a second of attention on calorie burn. It is the most visually prominent metric in most health apps and it is close to fiction.

A study of seven wrist devices and 60 volunteers, measured against indirect calorimetry while sitting, walking, running and cycling, found that no device kept energy expenditure error under 20 percent (PMID 28538708). The 158-study review reached the same place: no brand was accurate (PMID 32897239). The authors of the first study explicitly recommend caution in using these numbers at all.

A watch cannot know your muscle mass or your metabolic rate. It estimates from heart rate and movement, and those inputs are already noisy. When it tells you that you burned 600 calories, the number carries a margin of error wide enough to swallow your entire deficit.

Track minutes instead. Minutes are a hard number: you know how long you spent in a given heart rate zone. Calories are derived from that plus guesswork, and the deriving is where it falls apart.

Why your heart rate is wrong when you lift

If you lift or do intervals you have noticed this. Heart pounding, watch reading low. It is not your brand. It is the sensor class.

Wrist optical sensors read blood flow under the skin. That works when your arm swings steadily on a run and fails when you grip a bar. Wrist devices systematically underestimate heart rate during resistance training: mean difference -7.26 bpm, against -4.55 bpm on a bike, which is also an underestimate (PMID 32552580). The error grows about 3 bpm for every 10 bpm your actual heart rate climbs.

Grip compresses the tissue over the sensor and muscle tension adds noise the algorithm cannot separate from a pulse. Nothing in the firmware fixes that.

Switching brands will not fix it. If you need a true number while lifting, the sensor has to leave your wrist.

Where the error gets personal

The sensors do not perform equally on every wrist. In the same seven-device lab study, error was higher for men, for people with a higher BMI, and for people with darker skin tone (PMID 28538708).

Be precise about what that means. It was a laboratory result from 2017 hardware, and it does not establish that the watch on your wrist today is biased. It does mean the absolute number deserves less confidence, and it is a good reason the industry should publish validation across skin tones rather than assuming one sensor fits everyone.

If that is you, read the direction of the line rather than the point on the graph. A resting heart rate drifting down over a month is a real signal. A 3 bpm jump in a single day can be sensor noise.

HRV only means something against your own baseline

Heart rate variability gets treated like a daily grade for your nervous system. Without a fixed protocol it is closer to a random number generator.

HRV is a defined measurement with published standards (PMID 8598068). The value depends on the conditions it was recorded under. Read sitting up and then read lying down and you have measured two different things.

A methodology review in psychophysiology research makes the practical point: the methodological differences are large enough that results are difficult to compare across laboratories (PMID 28265249). Which means your number is not comparable to a friend's, to a different device, or to an average an app published.

The only comparison that holds is you against you. Same time, same posture, same sleep window, then look at the weekly trend instead of today's verdict. Change the conditions and you are measuring your morning routine, not your recovery.

If you want the long version of what that number means, which habits actually move it, and why the six-breaths-a-minute rule is oversold, I wrote it separately in HRV explained.

Deep sleep minutes are a guess

The morning sleep breakdown looks authoritative: light, deep, REM, each with a number. Your watch does not know what happened in your brain.

Devices are good at detecting sleep and mediocre at staging it. In the three-night lab study, sleep stage comparisons came out mixed across devices, and every device performed worse on nights with disrupted sleep (PMID 33378539). Disrupted sleep is the night you actually want the data.

The app renders deep sleep minutes with the same confidence as your step count. Steps are graded Strong here. Sleep staging is graded Limited. Trust your own rested-or-not read of the morning over a stage chart, and use total sleep time as the number that means something.

If your watch says irregular rhythm

This is the one place a wrist device can produce a genuinely serious signal, and it is still not a diagnosis.

The Apple Heart Study followed 419,297 participants over eight months. Half a percent, 2,161 people, received an irregular pulse notification. Of the 450 who returned ECG patches with analysable data, 34 percent had atrial fibrillation on the patch (PMID 31722151). The study was funded by Apple, which is worth knowing when a company studies its own product.

Read that percentage both ways. About one in three notified people had AF confirmed. Roughly two in three did not, on a patch applied an average of 13 days later. Positive predictive value was 0.84 when the notification recurred alongside an irregular tachogram on the ECG. Paroxysmal AF comes and goes, which is why a later patch can miss it.

Do not panic and do not dismiss it. The alert did its job if it gets you to a clinician and a real ECG. It is not a reason to diagnose yourself.

Blood oxygen is the thinnest claim on the watch

SpO2 gets marketed for altitude and sleep apnea screening. The validation record behind the wrist version is thin.

Pulse oximetry as a method carries documented measurement bias (PMID 33326721). I looked for a primary study comparing a wrist wearable against arterial blood gas, the reference method, and did not find one I could verify. The absence of that study is not proof your watch is wrong. It does mean nobody has shown the number is a clinical reading.

So treat it as a trend with a shrug, and if you have a condition that requires oxygen monitoring, use a device built for that. A wrist reading is not a substitute for it.

Readiness scores are a black box

Readiness, recovery and stress scores are the most confident-looking numbers on the device and the least evidence-backed. I found no published validation of a commercial readiness score against any physiological criterion.

What exists is a proprietary formula over inputs you cannot inspect, returning a single score with a color. It may be useful to a company and still be unvalidated science. Those are not the same thing.

What to look at instead: your resting heart rate against your own baseline, whether last night was short, and how you feel walking to the kitchen. That is not sophisticated, and it is at least measurable.

What to do this week

• Wear the watch snug, one finger above the wrist bone, on your non-dominant side when you lift.
• Stop reading calorie burn. Track minutes of activity instead.
• Take HRV at the same time, in the same posture, every morning, and watch the weekly trend only.
• Read resting heart rate as a seven-day average, not a daily number.
• Look at total sleep time and how you feel, and skip the deep sleep chart.
• Treat any irregular rhythm alert as a doctor visit, not a diagnosis.

Bottom line

Your watch is a good tool for steps, resting heart rate and sleep timing, and a poor one for calories, deep sleep staging, lifting heart rate and readiness scores. The difference between those two groups is documentation, not marketing.

Almost everything that improves the data is free: fit, protocol, and knowing which number to ignore. If you lift, the chest strap at $104.95 is the only purchase on this page, and it is a fix for a specific documented failure.

Trust trends. Ignore the black boxes. Believe the steps.