We track everything. But do we really understand our fitness?

More fitness data won't close the gap between tracking and understanding. A single run can now produce a step count, a heart rate trace, a pace graph, a calorie total and a VO2max estimate. VO2 is the volume of oxygen your body takes up and uses. VO2max is the highest rate it can reach during severe exercise, as Bassett and Howley define it.

The harder part is turning those numbers into a decision. Each metric has its own error and blind spots, which can be easy to miss when the numbers sit side by side. From that research, we drew four questions for deciding which changes deserve a response.

The case for tracking is real

An umbrella review of 39 systematic reviews and meta-analyses by Ferguson and colleagues (2022) says activity trackers "appear to be effective" at increasing physical activity. Its pooled results suggested about 1,800 extra steps per day and, separately, about 40 more minutes of walking per day, across a wide range of people rather than endurance athletes specifically.

Fuller and colleagues' 2020 systematic review of 158 publications concluded that commercial wearables are accurate for steps and heart rate in laboratory settings, though this varies by manufacturer and device type. Its results were more qualified: some brands "appeared to measure steps accurately", and heart rate measurement was "more variable".

But moving more and understanding your fitness are different outcomes. If you already track everything, the useful question is what last week's numbers mean for next week.

More fitness data doesn't automatically become understanding

Adding metrics doesn't address these three problems.

Estimates can look like measurements. A calorie total can sit beside a heart rate reading, yet the evidence behind them differs widely. We suspect that makes it easy to trust both equally.

Normal variation looks like change. Hopkins defines typical error as the standard deviation of one person's repeated measurements. Even a lab VO2max test has it. In a small 1982 study of five trained people, repeat tests varied by about ±5.6% in combined biological and technical variation, mostly biological. If lab tests vary that much, a small shift in a wearable estimate deserves caution.

Numbers can arrive without context. A pace figure alone doesn't show that it was hot, and a heart rate trace alone doesn't show that you were an hour into a long effort.

Measuring may also carry a cost. Etkin's six experiments found that measuring an activity such as walking or reading can increase how much people do while reducing how much they enjoy it. The paper attributes this to measurement drawing attention to output and making an activity feel more like work. It didn't study endurance athletes or compare more metrics with fewer.

What each number can and can't tell you

Steps

Steps count movement, but a count alone says little about intensity. The 10,000 target has modest origins: Tudor-Locke and Bassett trace it to Japanese walking clubs and a business slogan, and they note the promise of goals based on improvements from a person's own baseline. In our view, that baseline is a more useful comparison than a round number.

Heart rate

Heart rate is among the better-supported wrist signals. In Shcherbina and colleagues' 2017 test of seven wrist-worn devices, median error met the authors' 5% standard in most settings, though those were 2017 models and current devices may differ.

Interpretation is harder, because heart rate responds to more than fitness:

• Duration: during prolonged exercise, heart rate tends to rise while stroke volume (the blood pumped per beat) falls, a pattern called cardiovascular drift. Researchers still debate its exact mechanism.
• Conditions: Achten and Jeukendrup report that dehydration and ambient temperature can strongly affect the relationship between heart rate and oxygen uptake.
• Normal variation: Buchheit's 2014 review lists a typical error of about 3% for submaximal exercise heart rate and a smallest worthwhile change of about 1%. The second figure comes from group averages, and the table doesn't state the sensor or population behind the first.

At 150 beats per minute, 3% is about 4.5 beats, so a few beats between two similar sessions may be normal variation, especially since each session carries its own error.

Pace

Pace blends fitness with terrain and weather. We cover interpretation only here, as we didn't review GPS accuracy. In Minetti and colleagues' treadmill study of ten runners, the highest tested speed on the level was 3.13 m/s at an oxygen uptake of 35.5 ml/kg/min. At a +10% gradient, it was slower at 2.90 m/s, yet oxygen uptake was 53.0 ml/kg/min. A lower speed took considerably more oxygen.

In Ely and colleagues' analysis of seven marathons, top men finished 1.7% off the course record in the coolest conditions and 4.5% off in the warmest, and slower runners were affected more. Differences among top women followed a similar trend but weren't statistically significant. This is marathon race data, so it shows a pattern rather than measuring your training runs. Still, a slower run on a hot, hilly day may say little about fitness.

Calories

In our reading, calories have the weakest support of the wrist-sensor metrics reviewed here. Fuller's review found no brand accurate for energy expenditure. In Shcherbina's study, none of the seven devices reached the authors' acceptable range in any setting.

Achten and Jeukendrup's 2003 review describes a related heart-rate-based method as satisfactory at group level but not very accurate for individuals. That review predates wrist wearables, so this is our inference, but we'd read a wrist calorie estimate the same way: as a rough signal, not a precise count.

VO2max estimates

Wearables estimate VO2max with algorithms rather than measuring oxygen directly. Our guide to VO2 max measurement explains the lab method. A 2022 meta-analysis of 14 validation studies found low average bias for exercise-based algorithms, but concluded that individual-level error is large. The INTERLIVE network behind it includes the technology company Huawei, which also partly funded the work. The authors declared no conflicts of interest. A pooled average can hide differences between devices: the two single-device studies below read lower than lab values on average.

Limits of agreement are the range expected to cover about 95% of differences from the lab. As a rough illustration, a lab value of 50 could correspond to an estimate of around 40 to 60, although error may differ with fitness level: the Engel study found larger errors in its highly trained group (9.4% to 10.4%) than its moderately trained group (2.8% to 4.1%).

Even a perfect number would be only part of the picture. Bassett and Howley describe VO2max as setting the upper limit for endurance performance, while running economy (the oxygen uptake needed to run at a given speed) and the share of VO2max you can sustain also matter. They name speed at lactate threshold as the best physiological predictor of distance running performance.

A better approach: start with the decision

Choose metrics by the decision they could change. Before responding to any change, ask four questions:

1. What is this number? Is it closer to a sensor reading, like steps or heart rate, or a modeled estimate, like calories or VO2max?
2. How much does it wobble by itself? Is the change larger than the normal variation for that metric?
3. What else moved it? Consider heat, hills, hydration and session length before crediting or blaming your fitness.
4. What would I do differently? If no answer would change your next choice, the number is interesting but not actionable.

A worked example

Take a hypothetical case. Your average heart rate on a regular 60-minute route was 155 beats per minute, five above last month's, and the day was warm.

• What is it? A better-supported wrist signal.
• Does it exceed normal variation? This evidence can't separate it from normal variation. By our arithmetic, a 3% typical error at 155 is about 4.7 beats, and the error in a difference between two sessions is larger, at about 6.6 beats.
• What else moved it? Temperature can shift how heart rate relates to oxygen uptake, so the warm day may have contributed.
• What would you do differently? One session probably doesn't support a conclusion about fitness. Note the conditions and see whether the pattern persists in similar ones.

Individual responses differ too. In the HERITAGE Family Study, 481 sedentary adults followed the same 20-week program. VO2max rose by about 0.4 L/min on average, yet some people gained little or nothing and others gained more than 1 L/min. These were sedentary adults, so the spread may not map directly onto trained athletes, and some of it may reflect test variation even with repeat testing. A population average can't tell you which response is yours.

How to apply this

These steps are general education rather than a training plan, so take individual decisions to a qualified coach or health professional.

1. Write down one question. "Is my easy effort getting easier?" is workable. "Am I fit?" is too broad.
2. Choose the metric that answers it.
3. Standardize the comparison. Use similar routes, durations and conditions where you can.
4. Note the context. Record temperature, terrain and how the session felt.
5. Review monthly. Look for changes that persist beyond normal variation, not single-session swings.

Caveats

We know of no study testing this four-question approach, so read it as a reasoning aid rather than a proven method. The validation studies cited here also compare a device with a lab test at one point in time. A device with a consistent bias could still track direction of change better than its absolute error suggests, and as far as their abstracts show, these studies didn't test that question.

Fuller's review itself notes that constant model updates call for newer research, so some findings here may understate what current hardware can do. And simple tracking may be all some people need, given the apparent activity gains in the review evidence.

Reader questions

Should I stop tracking?

We wouldn't, on this evidence. Trackers "appear to be effective" at increasing activity, though that evidence isn't specific to endurance athletes. If tracking starts to feel like work, Etkin's findings make it reasonable to review which numbers you really use.

Is my watch's VO2max estimate useless?

Not on the evidence we found, though it depends on the use. Engel and colleagues, studying one Garmin model, describe smartwatches as useful for general monitoring, with lab gas analysis preferred when precision is required. Our guide also covers what a lab VO2 max test measures. The Apple Watch study above reported 13.31% error, and neither covers every device. Read an estimate as broad context and be cautious with small changes.

Understanding comes from asking better questions

For most people who own a wearable, collecting data is no longer the hard part, in our view. We'd like more tools to show uncertainty as clearly as the number itself. Until then, ask what would change your next choice, then look only at the data that can answer it.

What's the one number you check most, and when did it last change what you did?