Which Health Metrics Are Worth Tracking

Roughly three numbers on a modern watch are worth acting on. Heart rate, resting heart rate and sleep duration reward attention over weeks. Almost everything else is either a model dressed up as a measurement, or a measurement so noisy that a single day tells you nothing. Nothing here is medical advice, and persistent symptoms need a doctor rather than a wearable.

The useful question is not "is this accurate" but "is this measured, estimated or entertainment". A measured number comes from a sensor reading something real. An estimated number comes from a sensor reading plus a pile of assumptions. Entertainment is a number produced because a blank space would look bad.

The sorting table

Metric	Category	How to treat it
Heart rate (chest strap)	Measured	Trust it, including during intervals
Heart rate (wrist optical)	Estimated	Fine when steady, poor when effort changes fast
Resting heart rate	Measured, averaged	The single best trend on the watch
Heart rate variability	Measured, very noisy	Weekly trend, never a daily verdict
Step count	Estimated	A motivator, not a precise total
Calories burned	Estimated, often badly	Suspicion. Never budget food against it
Sleep duration	Estimated, usually close	Useful, and worth improving
Sleep stages	Estimated, an algorithm's guess	Interesting, not actionable
Blood oxygen	Wellness feature, not clinical	Curiosity only
Skin temperature	Measured as a deviation	Baseline shifts only, no explanation
Training load	Proprietary model	Useful within one brand, meaningless across brands
Readiness or body battery scores	Proprietary model	Entertainment with a sensible face

Heart rate

Heart rate is the foundation, and the method matters more than the brand. A chest strap reads the heart's electrical signal directly. A wrist watch shines light into your skin and infers the pulse from blood flow, a harder problem. Optical sensors are known to struggle with rapid changes in effort, cold hands and a loose strap, and the classic failure is a reading that locks onto your cadence instead of your pulse.

Good for: pacing steady efforts and keeping easy days easy. Cannot tell you: anything about your heart's health, and nothing reliable in the first minutes of a hard interval on the wrist. If your training is built on intervals, a strap fixes that for less than most watch upgrades cost. More detail in heart rate accuracy.

Resting heart rate

Resting heart rate is the quiet star of the category. It is the same optical estimate as above, but averaged over hours of stillness, which cancels most of the noise that ruins a single reading. Over weeks it drifts down as fitness improves and rises when you are under-slept, drinking or stressed.

Good for: a cheap fitness trend, and an early hint that you are asking too much of yourself. Cannot tell you: which cause is responsible, and nothing about any medical condition. If it changes and stays changed with no obvious explanation, that is a conversation with a doctor.

Heart rate variability

HRV is the variation in time between beats. It is a real signal with real physiology behind it, and the most over-interpreted number on any wearable. Day to day it swings on sleep position, alcohol, when the measurement window fell, hydration and chance. A single low morning means almost nothing.

Good for: a multi-week trend line, and only against your own history. Cannot tell you: whether to train today. HRV is not comparable between brands, since each device measures over different windows and processes the intervals differently. Take the shape of the line over a month and ignore the individual dots.

Step count

Steps are estimated from wrist motion, so pushing a stroller undercounts and chopping vegetables adds a few hundred. Nobody should care: the value of a step count is not precision, it is that it makes a sedentary day visible.

Good for: noticing the days you barely moved. Cannot tell you: an exact total, or anything about intensity. Ten thousand slow steps and an hour of hard running are not the same stimulus, and steps cannot tell them apart.

Calories burned

This is the number to treat with the most suspicion, because it looks precise and is not. A calorie figure is built on assumptions: your entered weight and age, an estimated resting metabolic rate, an efficiency assumption, and heart rate as a proxy for effort. Every one carries error and the errors compound. These estimates can be badly wrong in either direction, and wrong differently for different people.

Good for: comparing your own sessions on the same device. Cannot tell you: how much to eat. Eating back an estimated burn is the most common way this number does damage. If you are managing weight, treat it as a rough relative signal or ignore it.

Sleep duration versus sleep staging

These get lumped together and should not be. Total duration is usually close enough to be useful, because "moving and awake" versus "still and asleep" is an easy call from motion and heart rate. Stages are another matter. No wrist device measures brain activity, so light, deep and REM are an algorithm's guess. Change brands and the same night produces a different breakdown.

Good for: duration, consistency and bedtime drift, the sleep variables you can change. Cannot tell you: that you need more deep sleep, or that your staging is abnormal. Devices do not diagnose or detect anything about sleep, and if you feel exhausted despite adequate hours, see a doctor. More in sleep and recovery. A small band often beats a sports watch here, because you will wear it to bed.

Blood oxygen and skin temperature

Blood oxygen on a consumer watch is a wellness feature, not a clinical measurement. It reads through the back of a moving wrist in the dark, which is not what a medical device is designed for. It does not diagnose, detect or monitor any condition. Treat it as curiosity. If you feel unwell, a wrist reading should not decide what you do next.

Good for: very little on most devices. Cannot tell you: that you are fine, or that you are not. Skin temperature is more honest, because good implementations report a deviation from your own baseline rather than an absolute body temperature. A baseline shift is a real observation, but not an explanation: heat, bedding, alcohol and a hard session all move it, and the watch cannot say which.

Training load and readiness scores

Every brand has one, and they all present a model as though it were a measurement. Load, recovery time, body battery, readiness and training status are computed from the inputs above using an unpublished formula. They are not measurements and not comparable between brands. A readiness score of 68 on one watch and 68 on another are unrelated quantities.

Good for: a consistency check inside one ecosystem. If your load has climbed steeply for three weeks, that is worth noticing even if the units are arbitrary. Cannot tell you: whether to skip today's session. When the score and your body disagree, your body is the better instrument. Choosing a watch on its scoring model is a mistake, and how to choose a sports watch covers what actually differs.

Common mistakes

Reacting to single days is the big one: most metrics here are noisy enough that one reading is indistinguishable from chance, and people rearrange their training around it anyway. Comparing scores with other people is the second, since these numbers only mean something against your own baseline. The third is buying a pricier device hoping for better data, when the fix is usually a chest strap or a tighter band. The watch finder and the health tracking hub help narrow the choice.

Questions people ask

Which single metric should I watch?

Resting heart rate, tracked weekly. It is cheap to collect, needs no interpretation skill, and responds to the things that matter: training, sleep, alcohol and stress. Sleep duration is a close second, because it is one of the few numbers you can directly change.

My HRV dropped a lot this morning. Should I rest?

Not on that basis alone. One low reading is well within normal variation and can come from sleeping on your arm. Look at how you feel alongside the last two weeks. If the trend is falling and you feel flat, ease off. If the trend is fine and you feel fine, train.

Are the numbers comparable between brands?

Raw heart rate roughly yes, especially from a strap. Everything modelled, no. HRV, sleep stages, calories, load and readiness come from different algorithms and windows. Switching brands changes your numbers without changing your body, so start a fresh baseline.

A metric on my watch looks alarming. What should I do?

See a doctor, not a forum. Consumer wearables are not medical instruments, they do not diagnose or detect conditions, and a strange reading is as likely to be a sensor artefact as anything else. Nothing here is medical advice. If symptoms persist, get them looked at, and do not let a number on your wrist reassure you out of doing that.

The whole silo in one line: track trends over weeks, not single days, and if a number worries you, see a doctor rather than a forum. Still choosing a device? Start at the watches hub.