Library · Watch numbers · Explainer

What your watch actually measures

A sports watch measures a handful of things and works out the rest. Which number came from a sensor, which from a formula, and which from a guess about you — and why that order is the order in which to trust them.

Published 2 September 2026 15 sources · 9 figures · every number checked against the source
Sensor, model, or assumption
A runner's forearm and wrist mid-stride on an empty road at dawn, wearing a plain sports watch with a blank dark face; the background of tarmac and low sun is softly blurred.
Figure 1. Illustration. The screen shows a dozen numbers. The back of the watch sees three or four things. This page is about the difference.

A watch measures light bounced off the blood under your skin, where a satellite says you are, how your wrist moves, and air pressure. Everything else on the screen is worked out from those, and somewhere in the working-out sits a value about you that was never measured. That gives three kinds of number, and they deserve three levels of trust.

  • Measured numbers are close. Across a review of 158 validation papers, wrist heart rate landed within about three percent on average; on a 10 km run in the open, distance was within about five percent.
  • Modelled numbers are looser. In one head-to-head, the same seven devices that got heart rate within a few percent missed calories by 27 to 93 percent. VO2max, training load and recovery time live on this rung.
  • Assumed numbers are the quiet ones. A maximum heart rate of 220 minus your age was never measured on anyone in particular, and real maxima scatter seven to eleven beats around it. Every zone built on it moves with that scatter.
  • HRV and readiness get one rung here: what is measured, and what is computed from it. Whether the score is any good is a bigger question, and we will come back to it separately.
Prefer to watch? The whole explainer in 10:14, narrated — it has its own page too, with chapters and a full transcript. Subscribe on YouTube if you want to hear about the next one.

Somewhere on your wrist right now there is a number with a decimal point, a colour, and a word like "productive" or "unbalanced" next to it. It looks measured. Some of it was. The rest was worked out, and the working-out started from a value you typed in on the day you unboxed the watch and have not thought about since.

Watches deserve credit for this. A device that turns a few sensor readings into a whole dashboard is doing something clever, and most of the dashboard is useful. This page is about reading that dashboard well: knowing which numbers a sensor saw, which a formula produced, and which rest on a guess about you.

1 The ladder

Every number on the screen sits on one of three rungs, and the rung tells you how carefully to read it (figure 2).

Three columns side by side. Rung one, measured, in green: a sensor saw it — light bounced off the blood under the skin, where the satellites say you are, how the wrist moves, air pressure; examples heart rate, distance, cadence, elevation. Rung two, modelled, in blue: worked out from rung one by a formula; examples pace, calories, VO2max, training load, recovery. Rung three, assumed, in amber: typed in or taken from a table, a value about you that was never measured; examples max heart rate from age, weight, sex, activity level. Arrows point from left to right, labelled read with more care.
Figure 2. Three kinds of number. The first rung is what the hardware can physically sense. The second is what software works out from it. The third is what nobody measured but everything above depends on. This is the whole page in one picture; the rest is the detail per rung.

The rungs are not good and bad. A measured number can be wrong on a given second, and a modelled number can be a perfectly good guide to a trend. What changes as you climb is how many decisions somebody else made on the way to the number, and how much of it is about you rather than about the average person the formula was built on.

2 What can the watch actually sense?

Strip the software away and a sports watch has about four ways of knowing anything (figure 3).

Left, a cross-section drawing of a watch resting on the skin of a forearm: on its underside two green lights and one blue detector; green rays go down into the skin, reach two blood vessels drawn in rust, and one blue ray comes back up to the detector. Caption under it: green light in, some of it back — the pulse, not the heartbeat. Right, four green-tabbed entries: optical pulse sensor (light source plus detector on the back, measures blood-volume change under the skin, gives heart rate and HRV); satellite receiver (position every second or so, distance and speed follow from position over time); accelerometer (how the wrist moves, its rhythm and its shaking, also used to clean the pulse signal, gives cadence and steps); barometer (air pressure, falls as you climb, gives elevation gain).
Figure 3. The four sensors. The drawing is a schematic of the optical sensor, not a scale diagram. Everything a sports watch shows starts with one of these four, and most of the interesting numbers start with the first one.

The one that matters most is the optical sensor on the back. It is a small light source and a light detector pressed against your skin. The light goes in, some of it comes back, and how much comes back changes as the blood vessels underneath swell and shrink with each pulse. The technical name is photoplethysmography, PPG for short; the plain version is that the watch watches your blood pulse and counts.3 Most watches use green light for this, because it gives a clean signal and copes best with a moving wrist, though it does not travel far into the skin.4

Two things follow. First, the watch sees the pulse arrive at your wrist, which is a related but different signal from the heartbeat itself. That is why a chest strap, which reads the heart's own electrical activity through two electrodes, is still the reference in the lab.2,4 Second, anything that moves the sensor against the skin, or makes the blood under it slosh, gets into the signal. The accelerometer is partly there to help clean that up.3

The other three are simpler. A satellite receiver fixes your position roughly once a second; distance and speed follow from how that position changes. An accelerometer feels the wrist swing, which gives cadence and step counts. A barometer feels air pressure fall as you climb. None of these knows anything about your fitness, your fatigue, or your calories. That knowledge, where it appears, is on the next rung.

3 Rung one: the measured numbers

Heart rate and distance are the two numbers a sensor genuinely produced, and both have been tested against the truth many times.

Heart rate at the wrist

A systematic review from 2020 gathered 158 validation papers across nine brands. Its summary for heart rate: in controlled settings, devices from all the brands measured heart rate to within about three percent on average, with a slight lean towards reading low; the median error across comparisons was one percent under.1 On a heart rate of 150, three percent is four or five beats, which is a smaller difference than you would feel.

The catch is in the word "average". A guideline paper from the same year points out that wrist readings can be very accurate once you average them over a run, while any single reading in the middle of that run can be well off.4 A separate study of 53 people across the full range of skin tones found no effect of skin tone on accuracy. It did find that the error during activity was, on average, thirty percent higher than at rest. And it describes the failure mode that runners notice most: with a rhythmic movement like walking or jogging, the optical sensor can lock onto the rhythm of the arm swing and mistake it for the pulse.5 If your heart rate ever sat suspiciously close to your cadence for a few minutes, that was it.

A useful way to see what "a measured number" is worth is to compare it with a modelled one from the same device. Sixty volunteers wore seven wrist devices, a few at a time, while an ECG measured the true heart rate and a breathing mask measured true energy expenditure. Heart rate came out with a median error of 1.8 percent on the bike and 5.5 percent while walking, and between 2.0 and 6.8 percent across the seven devices overall. Calories, from exactly the same wrists, were off by between 27 and 93 percent depending on the device, and no device got below twenty percent (figure 4).6

Horizontal bars on a scale from zero to one hundred percent median error. Heart rate while cycling: 1.8 percent. Heart rate while walking: 5.5 percent. Heart rate, best to worst device: 2.0 to 6.8 percent, all in green as sensor numbers. Energy expenditure, in blue as a modelled number: 27.4 to 92.6 percent, no device below 20 percent error.
Figure 4. Same wrist, two rungs. The heart-rate bars are what a sensor produced. The calorie bar is what a formula produced from that sensor plus what the wearer typed in. From Shcherbina and colleagues (2017), sixty volunteers, seven devices; ranges are across devices, not the single best or worst.

Distance from the sky

Distance is the other measured number, and it has a weather of its own: the sky. Eight sport watches were walked, run and cycled over courses that had been measured to the metre, between 404.0 and 4296.9 metres long, in three kinds of terrain. Across all terrain the average error per watch was between 3.2 and 6.1 percent. On an open athletics track every watch was within five percent and tended to read slightly long. In forest and city streets every watch read short, by between 3.5 and 8.9 percent. The satellite signal bounces off buildings and trees before it reaches the receiver (figure 5).7 The authors' one-line summary: recorded distances might be underestimated by up to nine percent. On a 10 km run that is up to 900 metres, and five percent is 500 metres. That last step is our arithmetic on their number.

Horizontal bars on a scale from minus ten to plus six percent, recorded distance against the measured course. Athletics track: plus 0.9 to plus 4.1 percent, in green. Forest and city streets: minus 3.5 to minus 8.9 percent, in rust, labelled reads short. All areas together: 3.2 to 6.1 percent absolute error, in blue. A note reads: five percent short on a 10 km run is 500 metres, our arithmetic.
Figure 5. Where you run changes what the watch says you ran. From Gilgen-Ammann and colleagues (2020), eight watches from five brands over measured courses. A measured number, but measured through whatever is between you and the satellites.

The point of both stories is the same. A measured number is close on average, wanders on any single reading, and has a known list of things that push it around: arm swing for heart rate, tree cover for distance. That list is short and physical. On the next rung it gets longer and less visible.

4 Rung two: the modelled numbers

Pace, calories, VO2max, training load, recovery time. None of these was measured. Each was worked out from rung one by a formula, and the formula needed something about you to start from.

Pace is the gentlest case: distance divided by time, so it inherits the distance error and nothing else. Calories are the harshest, as figure 4 showed, because the body's energy use depends on things the watch cannot see. The review of 158 papers found fewer than ten percent of energy estimates within its acceptable band, and no brand that got it right.1

The most interesting case, because a manufacturer has published how it works, is the fitness number. One widely licensed method estimates VO2max from a normal run, and its white paper lists the steps. Log the person's background information, at least their age. Record heart rate and speed. Cut the run into segments and keep only the reliable ones. Then read fitness off the relationship between heart rate and speed. Because you never ran to your limit, the method scales that relationship to an age-based estimate of your maximum heart rate.9 The same family of methods turns a session into a training effect on a one-to-five scale. That comes from an estimated oxygen debt after the session and an activity class you report yourself on a scale from zero to ten (figure 6).10

A flow diagram. At the bottom, four inputs: heart rate measured at the wrist and speed from the satellites, both green; your age typed in once and activity level self-reported on a zero to ten scale, both amber. Age feeds an amber box, estimated maximum heart rate, from age unless you measured it, an assumption. Heart rate, speed and that estimated maximum feed a blue box, VO2max estimate, heart rate against speed scaled by that maximum. That box and the activity level feed a second blue box, training effect or load, from estimated oxygen debt plus your activity class. A note gives the manufacturer's own figure: a maximum heart rate guessed 15 beats too low makes the VO2max about 9 percent off, 15 beats too high about 7 percent, and a known real maximum about 5 percent.
Figure 6. How a modelled number is built. One published method, read as an example of the category; other makers publish less. The green inputs were measured. The amber ones were typed in or taken from a table. The blue numbers are what you see on the screen. From the manufacturer's own white papers, which describe the method and are read here as documentation, not as proof that it works.

Manufacturer documentation, read as documentation. The white papers behind figure 6 are the maker's account of its own method. We use them for what they are good for, which is the list of inputs and steps, and not for how accurate the result is. The maker's own sensitivity chart does earn a place here, because it is about the assumption and not about the product. If the age-based maximum heart rate is fifteen beats too low, the VO2max estimate is about nine percent off. Fifteen beats too high, about seven percent. With the real maximum known, about five percent.9

How the fitness number performs against a lab test is a question we answered in the VO2max explainer. Across fourteen validation studies, the exercise-based estimates were close on average, and could be off by nearly ten units either way on one person.15 The reading rule that follows holds for the whole rung: a modelled number is a decent guide to your own trend, and a poor substitute for a measurement of your level.

Training load is the same idea with a longer history. The classic version, the training impulse, multiplies how long you trained by how hard, where "how hard" is your heart rate placed between your resting and maximum values. A simpler cousin multiplies your own effort rating on a one-to-ten scale by the minutes. A power-based cousin defines one hour at threshold power as a hundred points.14 Notice what every version needs: a resting or maximum heart rate, a threshold, or your own opinion of the effort. That is the third rung leaking into the second.

5 Rung three: the assumptions

The quietest numbers on the watch are the ones you entered once, or that the watch filled in from a table, and that every model above leans on. The oldest of them is your maximum heart rate.

Most watches start you off with a maximum of 220 minus your age. Two exercise physiologists went looking for where that formula came from and published the answer in 2002. There is no study behind it. It traces back to a 1971 review in which the authors drew a chart of about 35 data points from roughly eleven earlier sources, and did no regression. Their caption said that no single line would adequately represent the data, but that 220 minus age "defines a line not far from many of the data points".8 When the 2002 authors refitted those same points properly, they got 215.4 minus 0.9147 times age, with individual maxima scattered 21 beats around the line.

The spread is the useful part. Across the research since, real maxima scatter seven to eleven beats either side of the age formulas. The review's conclusion on the whole family of age-only formulas: they carry prediction errors of more than ten beats a minute.8 The better-known alternative that pooled thirty of them, 208 minus 0.7 times age, lands almost on top of 220 minus age for a forty-year-old, and has the same scatter. Figure 7 shows what that does to a heart-rate zone.

A chart with beats per minute from 110 to 200 on the vertical axis, for a 40-year-old. Left, the formula says: a thin amber bar at 180. Middle, where real maxima sit: a rust block from 169 to 191, plus or minus 11 beats. Right, a 70 to 80 percent of max zone three ways: blue blocks at 118 to 135 for a maximum of 169, 126 to 144 for 180, and 134 to 153 for 191.
Figure 7. The assumption, and the zone built on it. For a forty-year-old the formula gives 180. Real maxima at that age sit up to eleven beats either side, per Robergs and Landwehr (2002). The three zones on the right are the same "70 to 80 percent of maximum" band computed for 169, 180 and 191; that arithmetic is ours. The lowest and the highest barely overlap.

The consequence runs upward through the ladder. Every zone that is a percentage of that maximum moves with it. Every training-effect number that scales effort against it moves with it. And, as the maker's own chart in figure 6 showed, the fitness estimate moves with it too. For you, the most useful value in the whole settings screen is a maximum heart rate that was actually observed, on a day when you went to your limit.

The other assumptions are the profile: weight, sex, age, and the activity level you reported. They feed the calorie estimate, the fitness estimate and the training-effect scale, and they are only as current as the last time you looked at that screen. A weight that is three kilos out of date is a small thing. But it is also a modelled number quietly drifting for a reason that has nothing to do with your training.

6 What about HRV and readiness?

Heart-rate variability and the recovery or readiness scores built on it get one rung on this page: what is measured, and what is worked out. Whether the score is any good is a bigger question, and we will come back to it separately.

What is measured is the gap between beats, in milliseconds, usually at night when the wrist is still. What is computed starts immediately after that. The most common single number, RMSSD, takes each successive difference between those gaps, squares it, averages the squares and takes the root. It is the standard short-term measure of how much the beat-to-beat rhythm varies.13 Where a lab would use an electrocardiogram, the watch uses the same optical pulse as everything else. So the motion caveats from rung one apply here too, softened by the fact that you are asleep.

From one number, a manufacturer's method builds several more. One published approach detects stress when heart rate is elevated, variability is reduced and the breathing pattern in the signal is irregular. It detects recovery when heart rate sits near your resting level and variability is large and regular with your breathing.11 Its night-time recovery index is computed from heart rate, two frequency bands of variability and a breathing rate derived from the same signal. The window is four hours, starting thirty minutes after you go to bed. The result is then scaled against your own measurement history, so what you see is a comparison of you with you.12 A "body resources" or readiness score is the running balance of those stress and recovery periods, with sleep and training folded in (figure 8).11

Six boxes left to right joined by arrows. Pulses, optical sensor usually at night, and gaps between beats in milliseconds, both green under the heading measured. One HRV number such as RMSSD, your own baseline scaled to your measurement history, and stress or recovery from heart rate up, HRV down and breathing pattern, all blue under the heading modelled. Readiness score, plus sleep, plus training load, plus a scale, in amber under the heading modelled plus assumed. Notes below: each step to the right adds a decision somebody else made; one manufacturer's night-time window is four hours starting thirty minutes after going to bed, compared with your own past readings.
Figure 8. From pulses to a score. The sensor's contribution is the two boxes on the left. Everything to the right is a choice: which statistic, which window, which baseline, which weights. The method shown is one maker's published description; the weights that turn it into a score on the screen are not published by anyone.

The part this page leaves alone. None of the above says whether a readiness score predicts anything about how your next session will go. That is a separate question with its own literature, and it belongs to a check on recovery wearables that is in the queue on our library page. Here the reading rule is only this. The score is several rungs above the sensor. It is calibrated to your own past, not to any external truth. And a single morning's value inherits every wobble from the pulse signal underneath it.

7 How do you read the number in front of you?

Put the rungs back together and the watch face sorts itself (figure 9).

A table with seven rows and five columns: the number, measured, modelled, assumed, read it as. Heart rate: pulse under the skin; beats per minute; nothing assumed; close, worst in rhythmic movement. Distance and pace: satellite position; distance divided by time; nothing assumed; within about five percent in the open, short under cover. Calories: heart rate and motion; energy from a formula; weight, sex, age; a rough order of magnitude. VO2max: heart rate and speed; fitness from heart rate against speed; max heart rate from age; a trend, not a lab value. Training load: heart rate and time; oxygen debt then a score; activity level you reported; a relative scale, yours to yours. Max HR and zones: nothing measured; a percentage of a maximum; 220 minus age unless measured; plus or minus eleven beats wide until you test it. HRV and readiness: gaps between beats; a statistic, a baseline, a score; weights nobody published; your own trend, validity to be checked.
Figure 9. Which rung each number is on. A colour per rung, as in figure 2. The right-hand column is our reading of the sources in this piece, not a manufacturer's specification.

Two habits fall out of the table. Read a measured number as a value and a modelled number as a trend. Your heart rate at this moment is a fact with a small error bar; your VO2max this week is a direction of travel. And when a modelled number surprises you, look down the ladder before you look at your training. A changed weight in the profile, a maximum heart rate that was never measured on you, a run under tree cover, a loose strap. Most surprises live on a lower rung than the number that reported them.

8 Seven things to remember

If you close this tab and look at your wrist, these are the things that were probably not on the same screen as the number.

  1. The watch senses about four things. Light through the skin, satellite position, wrist motion, air pressure. Everything else is worked out.
  2. Wrist heart rate is close on average, within about three percent across 158 validation papers, and least reliable when your arm swings in a steady rhythm.
  3. Distance is a sensor number with a sky problem. Within five percent in the open; up to nine percent short under trees and between buildings.
  4. Calories are the loosest number on the screen. Twenty-seven to ninety-three percent off in one head-to-head, from the same wrists that got heart rate within a few percent.
  5. VO2max, training load and recovery are models, and a good model of you is still scaled to something you typed in. Read them as trends.
  6. 220 minus age was never measured on you. Real maxima scatter seven to eleven beats around it, and every zone and every fitness estimate built on it moves with that.
  7. Readiness is several rungs up. Measured gaps between beats, then a statistic, a baseline, and a score. What it is worth is a question for another day, and it is on our list.

A watch is a very good sensor wearing a very confident dashboard. Read the sensor for what it saw, read the dashboard for where it thinks you are going, and check the settings screen before you check your training.

Sources

This is an explainer rather than a review: the fifteen sources below were picked to cover each rung of the ladder — how the sensors work, how well the measured numbers hold up, how the modelled numbers are built, and where the assumptions come from. We read every one in full at its own address and keep a copy. T1 is research; T3 is manufacturer documentation, which tells you how a method works and nothing about how well. After each entry: what we took from it.

  1. T1 Fuller D, Colwell E, Low J, et al. Reliability and Validity of Commercially Available Wearable Devices for Measuring Steps, Energy Expenditure, and Heart Rate: Systematic Review. JMIR mHealth and uHealth. 2020;8(9):e18694. doi:10.2196/18694 — 158 publications, nine brands; heart rate "within ±3% on average in controlled settings", median error −1%; energy expenditure: 9.2% of comparisons within ±3%, no brand accurate.
  2. T1 Mühlen JM, Stang J, Lykke Skovgaard E, et al. Recommendations for determining the validity of consumer wearable heart rate devices: expert statement and checklist of the INTERLIVE Network. British Journal of Sports Medicine. 2021;55:767–779. doi:10.1136/bjsports-2020-103148 — the description of PPG as absorption and reflection of emitted light by the blood, and of chest straps as two-electrode ECG.
  3. T1 Castaneda D, Esparza A, Ghamari M, Soltanpur C, Nazeran H. A review on wearable photoplethysmography sensors and their potential future applications in health care. International Journal of Biosensors & Bioelectronics. 2018;4(4):195–202. doi:10.15406/ijbsbe.2018.04.00125 — light source plus photodetector, reflected light proportional to blood-volume variation, green LED most common, motion artefacts and accelerometer correction.
  4. T1 Nelson BW, Low CA, Jacobson N, Areán P, Torous J, Allen NB. Guidelines for wrist-worn consumer wearable assessment of heart rate in biobehavioral research. npj Digital Medicine. 2020;3:90. doi:10.1038/s41746-020-0297-4 — pulse rate versus heart rate as distinct signals, green versus red light, and the finding that aggregated readings can be accurate while any single observation deviates.
  5. T1 Bent B, Goldstein BA, Kibbe WA, Dunn JP. Investigating sources of inaccuracy in wearable optical heart rate sensors. npj Digital Medicine. 2020;3:18. doi:10.1038/s41746-020-0226-6 — 53 participants across all skin tones; no skin-tone effect; absolute error during activity on average 30% higher than at rest; the "signal crossover" effect.
  6. T1 Shcherbina A, Mattsson CM, Waggott D, et al. Accuracy in Wrist-Worn, Sensor-Based Measurements of Heart Rate and Energy Expenditure in a Diverse Cohort. Journal of Personalized Medicine. 2017;7(2):3. doi:10.3390/jpm7020003 — 60 volunteers, seven devices; median heart-rate error 1.8% cycling, 5.5% walking, 2.0% to 6.8% across devices; energy-expenditure error 27.4% to 92.6%, no device below 20%.
  7. T1 Gilgen-Ammann R, Schweizer T, Wyss T. Accuracy of Distance Recordings in Eight Positioning-Enabled Sport Watches: Instrument Validation Study. JMIR mHealth and uHealth. 2020;8(6):e17118. doi:10.2196/17118 — reference distances 404.0 to 4296.9 m; mean absolute percentage error 3.2% to 6.1%; track +0.9% to +4.1%; forest and urban −3.5% to −8.9%; "underestimated by up to 9%".
  8. T1 Robergs RA, Landwehr R. The surprising history of the "HRmax=220-age" equation. Journal of Exercise Physiology online. 2002;5(2):1–10 — Fox et al. 1971 as origin, about 35 data points from approximately 11 references, no regression; standard error of estimate 7–11 beats per minute; refit 215.4 − 0.9147 × age with Sxy 21; pooled regression over 30 equations 208.754 − 0.734 × age; all age-based univariate equations with errors above 10 beats per minute.
  9. T3 Firstbeat Technologies. Automated Fitness Level (VO2max) Estimation with Heart Rate and Speed Data. White paper, published 7 November 2014, updated 30 June 2017 — the calculation steps (background information, at least age; heart rate and speed; reliable segments) and figure 5, the sensitivity of the estimate to the age-based maximum heart rate (15 beats too low, about 9%; 15 beats too high, about 7%; real maximum known, about 5%).
  10. T3 Firstbeat Technologies. EPOC Based Training Effect Assessment. White paper, published September 2005, last updated March 2012 — training effect on a 1–5 scale from peak EPOC and an activity class on a 0–10 scale.
  11. T3 Firstbeat Technologies. Stress and Recovery Analysis Method Based on 24-hour Heart Rate Variability. White paper, published 16 September 2014, updated 4 November 2014 — definitions of stress, recovery and body resources, and how the states are detected from heart rate, variability and respiratory pattern.
  12. T3 Firstbeat Technologies. Recovery Analysis for Athletic Training Based on Heart Rate Variability. White paper, published 15 June 2015 — the Recovery Index from heart rate, low- and high-frequency HRV power and HRV-derived respiration rate, scaled to the user's history, over a 4-hour window starting 30 minutes after going to bed.
  13. T1 Shaffer F, Ginsberg JP. An Overview of Heart Rate Variability Metrics and Norms. Frontiers in Public Health. 2017;5:258. doi:10.3389/fpubh.2017.00258 — the definition of RMSSD and its role as the primary time-domain measure of vagally mediated variability.
  14. T1 Halson SL. Monitoring Training Load to Understand Fatigue in Athletes. Sports Medicine. 2014;44(Suppl 2):S139–S147. doi:10.1007/s40279-014-0253-z — the training impulse (duration with maximal, resting and average heart rate), the session-RPE method (rating 1–10 times minutes) and the power-based training stress score (one hour at functional threshold power = 100 points).
  15. T1 Molina-Garcia P, Notbohm HL, Schumann M, et al. Validity of Estimating VO2max by Consumer Wearables: A Systematic Review with Meta-analysis and Expert Statement of the INTERLIVE Network. Sports Medicine. 2022;52(7):1577–1597. doi:10.1007/s40279-021-01639-y — used in our VO2max explainer; fourteen validation studies, exercise-based algorithms with a pooled bias of −0.09 and limits of agreement from −9.92 to +9.74 mL/kg/min.

Found a mistake here? Tell us at info@enduranceproof.com — you do not need to be a scientist, and "this number looks off" is a perfectly good message. Anything we correct, and when, goes on our corrections page.

This is educational material about how sports watches produce their numbers, not training, medical or nutrition advice, and not a recommendation for or against any device. Nothing here is a recommendation about what any individual should do.

One email when the next piece lands.

A short note when a new check or explainer goes up. No spam, never shared, one click to leave.

No spam, never shared, one click to leave.