Where the numbers come from
Every figure this app derives rather than measures is computed from a published source, and all of them are listed here. If a number on your screen cannot be traced to something on this page, that is a bug — tell the assistant and it goes straight to us.
Investigating the accuracy of Apple Watch VO₂ max measurements: A validation studyverified
Measured: 28 healthy adults (30 recruited, 2 excluded), mean age 31.9, 50% female, Dublin; Apple Watch worn 5-10 days, then a maximal treadmill test with indirect calorimetry
Used for: The error range shown beside every fitness age.
Mean absolute percentage error 13.3% (95% CI 10.0-16.6); mean absolute error 6.92 ml/kg/min. The watch read LOW by 6.07 ml/kg/min on average (43.27 against a measured 49.35), with limits of agreement from -6.11 to +18.26. That bias is described but deliberately NOT corrected for — see the INTERLIVE meta-analysis and the note below. Twenty-one of the 28 participants were rated Superior or Excellent and only one Fair, which the authors name as the study's most significant limitation. Data: github.com/rorylambe/applewatch-validation
https://doi.org/10.1371/journal.pone.0323741Last checked 2026-08-23
Validity of Estimating the Maximal Oxygen Consumption by Consumer Wearables: A Systematic Review with Meta-analysis and Expert Statement of the INTERLIVE Networkverified
Measured: Meta-analysis of 14 studies across multiple wearable brands
Used for: Why no bias correction is applied to a watch's VO₂ max.
Pooled bias across 14 studies is -0.09 ml/kg/min — essentially zero — but with limits of agreement from -9.92 to +9.74. Large errors in BOTH directions that average out. Together with the finding that wearables tend to underestimate for fit people and OVERESTIMATE for unfit ones (Caserman et al. 2024), this is the reason a single correction is not applied here: the direction of the error depends on the fitness of the person being measured, so subtracting a mean taken from a very fit sample would make the number worse for exactly the users least likely to be fit.
https://doi.org/10.1007/s40279-021-01639-yLast checked 2026-08-23
Aerobics Center Longitudinal Study reference standards, reproduced in ACSM's Guidelines for Exercise Testing and Prescription, 11th ed., Table 4.7verified
Measured: 80,000+ adults, maximal treadmill tests with direct gas analysis
Used for: Fitness age — the default reference population.
The reference table clinicians use. Broadest sample of the three, and its median sits closest to the one implied by Apple's own cardio-fitness bands — which matters because almost every VO₂ max entered here is read off a watch.
https://vo2maxcalculators.com/chart/Last checked 2026-08-23
The HUNT3 Fitness Study reference valuesverified
Measured: Several thousand Norwegians, directly measured on a treadmill
Used for: Fitness age — selectable alternative reference.
Europe's largest directly-measured fitness database, and the dataset the 'fitness age' idea comes from. Its participants are notably fit — it puts the average 44-year-old man at 47.2 — so it returns older ages than the others for the same measurement. That is the population, not an error.
https://www.ntnu.edu/cerg/fitness-numbersLast checked 2026-08-23
A Review of the Science of Colorful, Plant-Based Food and Practical Strategies for “Eating the Rainbow”cited
Measured: Narrative review of the phytonutrient literature by colour family (a corrigendum was later published; the review stands)
Used for: The colour-variety cue on logged meals: why the dots count colour families at all.
Different pigments travel with different phytonutrient families — carotenoids with orange and red, anthocyanins with purple and blue, chlorophyll and folate with the greens, allium flavonoids with the whites — so colour variety is a visible, low-effort proxy for nutrient variety. Used as a design rationale only: no number in the app is computed from this paper, and the dots are deliberately a marker, never a score.
https://doi.org/10.1155/2019/2125070Last checked 2026-08-30
Graded Exercise Testing Protocols for the Determination of VO₂max: Historical Perspectives, Progress, and Future Considerationscited
Measured: Review; meta-analysis of 39 studies for measurement error
Used for: Context for how exact the 'gold standard' itself is.
Laboratory indirect calorimetry — the criterion every watch is judged against — carries its own error of about ±5%, with a mean standard error of 2.58 ml/kg/min across 39 studies. Worth stating plainly: there is no number here that is exact, only numbers whose error is known.
https://doi.org/10.1155/2016/3968393Last checked 2026-08-23
Cardiorespiratory fitness is the strongest modifiable predictor of all-cause mortality (background)cited
Measured: —
Used for: Why fitness age is shown at all, rather than a body-composition figure with no validated derivation.
Informs the decision in app/nutrition/fitness_age.py to derive an age from VO₂ max only, and to refuse to derive one from muscle and fat.
https://pubmed.ncbi.nlm.nih.gov/27881567/Last checked 2026-08-23
The HUNT4 Fitness Studyunverified
Measured: 1,500+ people re-tested ten years after HUNT3
Used for: Nothing yet.
HUNT4 is largely a repeated-measures study of the same people, which makes it valuable for how fitness CHANGES rather than for population norms. No published VO₂ max reference table was located. Worth re-checking.
https://www.ntnu.edu/cerg/fitness-numbersLast checked 2026-08-23
Reference Standards for Cardiorespiratory Fitness Measured With Cardiopulmonary Exercise Testing (FRIEND)unverified
Measured: 32,329 maximal treadmill tests (FRIEND-I update: 11,678, three countries)
Used for: Not used for any number. Listed because it is the other standard reference and the comparison is informative.
The full 50th-percentile table is behind a paywall and the figures reproduced by secondary sources could not be separated from the Cooper table with confidence. It is therefore NOT offered as a selectable population: a source we could not read is a gap to record, not a number to ship. Two anchors are published — men 20-29 at 49.5, men 70-79 at 30.8, declining about 9% per decade.
https://pubmed.ncbi.nlm.nih.gov/26455884/Last checked 2026-08-23
Verified means the values in the code were read from that source. Cited means it informs a design decision but no table was taken from it. Unverified means we wanted it and could not get it — listed anyway, because a gap you can see is worth more than a tidy list that hides one. Nothing marked unverified is used to produce a number.