Measuring a body

A wearable is a measuring instrument worn on the body around the clock. Like any instrument, it is accurate for some quantities, poor for others, and noisy for all of them. The number on the screen each morning hides that. It arrives as a recovery score or a readiness percentage, built from a population model the wearer was never fitted to, with no statement of how far it can be trusted. The more useful question is narrower and harder: is the thing I changed working, for me, by more than my own noise?

What the 2025 validation studies show

Independent validation in 2025 gives a clear picture of where consumer devices stand.

On overnight heart signals, they can be very good. Dial and colleagues, in Physiological Reports in 2025, compared overnight resting heart rate and heart rate variability from five wearables against an ECG reference across 536 nights in 13 healthy adults. Oura's rings agreed best, with a concordance of 0.99 for HRV on the Gen 4. Agreement was lower elsewhere: 0.94 for HRV on Whoop, 0.87 for HRV on a Garmin Fenix 6, and 0.86 for resting heart rate on Polar. Garmin was left out of the resting heart rate analysis for methodological inconsistencies.

On aerobic capacity, the error is larger. Lambe and colleagues, in PLOS ONE in 2025, tested Apple Watch VO2max estimates in 30 participants against indirect calorimetry. The mean absolute percentage error was 13.31%, with a mean absolute error of 6.92 mL/kg/min, and the watch underestimated on average by 6.07 mL/kg/min. The authors concluded the estimates need refinement before clinical use.

On energy expenditure, the error is largest. A meta-analysis of 56 studies by Choe and Kang at the University of Mississippi in 2025 found Apple Watch errors of 4.43% for heart rate and 8.17% for steps, rising to 27.96% for energy expenditure.

The pattern is consistent. Signals the sensor measures directly, such as heart rate during sleep, come out close to the reference. Quantities the device estimates from a model of the average person, such as VO2max and calories, carry far wider error. The headline score sits on top of both.

A score answers a different question

Every wearable hands its wearer a number. Livgens Health's site puts the gap precisely: a recovery of 64% is a number about you, built from a population model you were never fitted to. It cannot separate your signal from your noise, so it cannot tell you whether the late session, the alcohol or the travel is the thing that matters.

That is a question of individual effect, and it has a different shape. It needs your own normal range, your own measurement error, and a change made deliberately enough that its effect can be told apart from the night-to-night wobble.

Answering it on your own measurements

Livgens Health answers that question and shows its working. The answer it gives reads like this: evening training is costing you 3.9 bpm, between 2.9 and 5.0, on your own nights, at your own measurement error. That is a decision someone can act on tonight. When the evidence does not support a decision, Health says so instead of estimating anyway.

The route is short. On day one, connect the ring or watch already worn, with no new hardware and nothing to log. By day eight, eight nights of sleep give your own normal range; everyone in Livgens's testing had theirs by then. From day twelve, change one thing, some days on and some off, and get the size of the effect on you and how sure the estimate is, or be told it is too small to bother with.

Measuring alone, an answer takes the best part of a year. Measuring with a squad alongside, it takes about twenty-eight nights.

Tested on real people

The site reports the results on real measurement. Across 1,623 nights, all 13 people tested ended up with a threshold around five times tighter than the device figure: 2.5 beats against 13. And on a trial where the drug genuinely did nothing, Health told all 38 patients it did nothing.

That second result is the one that earns trust. A system that finds an effect in everyone is reporting its own optimism. A system that reports no effect where none exists is measuring.

Health also looks ahead: resting heart rate, HRV and jump height projected four weeks from a person's own history, with the odds of moving outside their own normal range, week by week. Tested on the same 233 people from four public research datasets, none of whom were used to build it, it was more accurate than a leading AI forecasting model. Readings are opened with the user's passkey and forecast on their own phone or laptop, and nothing is sent to Livgens.

The instruments with a known floor

Wearables are one instrument among several, and the better ones have published error floors. Health holds each to its own standard. A countermovement jump on a force plate resolves 2.79 cm, with an ICC of 0.93 and a CV of 2.5%, where a wearable sensor needs 8.63 cm on the identical jump. VO2max by true cardiopulmonary exercise testing resolves 166 to 415 mL/min, against plus or minus 12 mL/kg/min for a wearable estimate. Lumbar spine bone density uses the ISCD least significant change of 5.3%, set by the measuring facility. Spirometry uses the ATS/ERS within-session repeatability criterion of 150 mL for FEV1 and FVC. Eccentric hamstring peak torque is held to the published floor of 15% or more.

A change smaller than the instrument's floor is noise. A change larger than it is worth a decision.

What it refuses

An exact representation of a body includes what cannot be said about it. Health's refusals are part of the product.

Your DNA does not predict how you respond to a diet: a randomised trial at the extremes of genetic predisposition found P = 0.94. Limb asymmetry is refused as an individual verdict, because bilateral strength ratios reach ICC below 0.60. Ranking teammates against each other is refused. Per-food verdicts from a single continuous glucose monitor pass are refused, because around 90% of individual food effects fall inside the within-person noise of a single pass. Gait speed and heart rate variability are shown as a trend, never as a verdict.

Locked before looking

The analysis is pre-registered. Every threshold and decision rule was fixed in advance and rehearsed end to end, so the cohort cannot be read into saying what was hoped. Validation figures come from openly licensed data, PMData and a Raynaud n-of-1 series, alongside measured results on Livgens's own system and the published literature.

Where Livgens fits

Livgens represents the thing exactly, in sound, light, matter and motion, and Living Worlds runs it forward through its lifecycle. Bodies are one of the things Livgens represents: your own measurements, and bodies fitted to any size. Health is live today at app.livgens.com and needs no request. Connect what you already wear, and find out what works for you, measured on you, with the evidence shown.