Oura, WHOOP or Garmin: How Accurate Is Your HRV?

Same body, different verdicts

You wake up. The ring says you are fully recovered. The watch says take it easy. The app on your phone, which reads both, says something in between.

It is tempting to ask which one is right. That is usually the wrong question. A recovery score is the end of a chain, and each link adds its own wobble. Let's walk the chain.

Link one: how well does the device measure?

Recovery scores lean heavily on heart rate variability, or HRV: how much the gaps between your heartbeats vary. Wearables estimate it overnight from the pulse at your wrist or finger.

A 2025 study in Physiological Reports (Dial and colleagues) compared five consumer wearables with a chest-strap ECG (a Polar H10) over 536 nights in 13 healthy adults. On the main HRV measure, the average error was:

All of them rose and fell with the reference to some degree, but the size of the miss differed, and the Garmin and Polar readings agreed least with the reference. These are specific models; a newer watch from the same brand may do better or worse.

Two honest notes. This is one small study, so treat the ranking as a first look, not a final league table. And the numbers above come from the paper's abstract, so check the full paper for the exact method before quoting one.

A separate study of 114 people wearing one ring found strong agreement overall, but more than half of the older participants had HRV errors above 10%. Averaging over the whole night, rather than a short window, improved accuracy.

So link one alone can make two devices disagree by several percent before any scoring happens.

Link two: your body is not a constant

Suppose you had a perfect sensor. Your HRV would still move. A review by Buchheit (2014) put the normal day-to-day variation of the common HRV measure at 10 to 20% in healthy people, and the resting heart rate at around 10%. Nothing is wrong when it moves. That is what healthy variation looks like.

Put this next to the device error. A 15% drop sounds dramatic. Next to 10 to 20% of natural wobble plus 6 to 10% of measurement wobble, it is within the ordinary range.

Link three: where the lines are drawn

Now each app turns your reading into a colour. The usual approach is a rule such as "if today is more than 15% below your normal, go easy."

We worked out what that rule does to a perfectly healthy athlete. The answer depends on how noisy they are:

(These come from simple arithmetic assuming bell-curve noise, in our lab's notes. Real data may differ.)

Different apps use different cut-offs, different baselines (7 days, 30 days, 60 days) and mix in sleep, resting heart rate and sometimes your own answers. Two apps can look at the identical heartbeat and honestly reach different colours.

So which one should you believe?

Probably neither, on one morning alone. Here is what the research and our simulations suggest.

Look for runs, not blips. In our simulations of made-up athletes, requiring two low mornings in a row cut false alarms to about 1.3 per 28 days, compared with about 4 for a one-day rule at typical noise. The price is that it notices a real dip a day late and catches fewer of them (about 58% in our test).

Judge against your own spread. A rule built on your own usual range fires at a steady rate. A fixed percentage fires far more often for noisy sleepers than for steady ones.

Ask for a second sign. A low HRV plus a bad night plus a raised resting heart rate is a stronger story than any one of them. In our simulations, asking for that kind of agreement cut the share of healthy days forced to easy-only from about a third to about a tenth.

Count disagreement as information. If your ring and watch disagree, the honest summary is often "the signal is weak today." That is a reason to lean on how you feel and on the plan, not a reason to pick the more dramatic number.

Does steering by HRV even help?

Two pooled analyses of trials that guided endurance training by HRV (Manresa-Rocamora 2021, 199 people; Düking 2021, 198 people) found no statistically clear gain in performance. These trials are small, so that is not proof of no benefit. But it is a good reason not to hand your whole week to a single morning number.

What we cannot tell you

We have not opened the black box of any vendor's score, and each one is proprietary. We have no data on real athletes. Our figures about false alarms come from computer-made athletes and describe rules, not outcomes.

The short version

Recovery scores disagree because they stack three uncertain things: a sensor with a few percent of error, a body that varies by 10 to 20%, and a threshold somebody picked. Treat a score as a hint. Trust a pattern.

We are currently improving how the pace24 engine treats noisy recovery signals, using the same simulations described above, and the early results are promising: in our tests, a design that asks for agreement between signals forces far fewer unnecessary easy days.

Interested in how pace24 treats noisy recovery signals? Get early access.