Which health tracker gives the best overnight recovery metrics

Short answer

Oura, Garmin and Google all report overnight HRV as RMSSD. WHOOP does too. Apple reports SDNN. But even among the RMSSD group the measurement window differs — whole night, main sleep only, or just the last deep-sleep cycle — so the numbers are not interchangeable. In the one study measuring several devices against ECG across 536 nights, Oura's newest ring was closest.

If your HRV has looked different on two devices on the same night, you were not imagining it — they are measuring different things.

Note: this is a wellness comparison, not medical advice. None of these devices is a diagnostic tool. Facts about other companies come from their own documentation and APIs, read on 22 September 2026, and from named peer-reviewed studies.

Do you need a wearable for overnight metrics?

You do not need to buy one. A phone cannot measure you while you sleep, so health trackers for overnight recovery metrics are the only option for that window — but if you already own any of the devices below, you already have one. Welltory reads Apple Watch, Samsung Watch and Bluetooth heart-rate monitors directly, and pulls overnight data through Apple Health, Health Connect, Samsung Health, Oura and Garmin.

What is worth knowing is that the overnight window is not automatically better. It is more standardised — you are lying still, at a consistent time, hours after your last coffee — and that is a real advantage for comparing one night with another. But a morning spot measurement taken the same way each day gets most of the same standardisation, and you control it.

Welltory's camera measurement against simultaneous ECG was published in Computer Methods and Programs in Biomedicine in 2022, with correlations of r = 0.77–0.94 in 26 professional cyclists. If you already own a wearable, Welltory reads it rather than replacing it — Apple Watch, Samsung Watch and Bluetooth heart-rate monitors directly, and Apple Health, Health Connect, Samsung Health, Oura and Garmin through their integrations.

The window problem, which is bigger than the accuracy problem

Oura vs WHOOP is the comparison people look for most often here, and it makes the point neatly: both report the same statistic, and they still do not produce the same number. Oura vs Garmin is the closest pair on this list, because both average across the whole night.

Every company reports a single overnight HRV number. Here is what that number is made of.

HRV rises across the night and differs by sleep stage. So a value taken from the last slow-wave period is structurally higher than an average across the whole night — not because one device is more accurate, but because they are measuring different things and calling them the same thing.

This is not a theoretical objection. It is an argument that played out in print. When a 2025 study in Physiological Reports compared several devices' overnight RMSSD against ECG, two WHOOP employees published a rebuttal arguing that WHOOP's deep-sleep-weighted value had been compared against an all-night ECG average, which is not a like-for-like test. The exchange itself shows that the industry has not settled what "overnight HRV" means.

And SDNN is a different quantity again. SDNN captures total variability across both branches of the autonomic nervous system; RMSSD is beat-to-beat and weighted towards parasympathetic activity. An Apple number and an Oura number from the same night are not one measurement taken twice.

How accurate is the overnight HRV itself?

One study measured this properly. Dial and colleagues, in Physiological Reports in 2025, compared several devices against a Polar H10 chest strap with ECG-grade input across 536 free-living nights in 13 participants — real nights at home, not one lab night.

Two things to take from this. Overnight HRV is measured with lower error than the stage estimates these same devices produce. And the spread between the lowest and highest error here is large enough to matter if you switch devices.

Note also that this is the only peer-reviewed test of Oura Ring 4 hardware that exists, and that 13 participants is a small sample even across 536 nights — wearable HRV accuracy has been studied far less than the marketing implies.

What else each one produces overnight

Oura gives a Sleep Score and a Readiness Score, both 0–100, with published contributor lists. Overnight values include lowest resting heart rate and when it occurred, average and maximum HRV, respiratory rate derived from the beat-interval stream, skin temperature deviation and blood oxygen. Its stage estimate uses movement, skin temperature, resting heart rate, HRV and respiratory rate together — the most inputs of anyone here.

Garmin gives a Sleep Score 0–100 with published bands, Body Battery 5–100, HRV Status classified as Balanced, Unbalanced, Low or Poor against a personal baseline, and Training Readiness 1–100. It reports both an all-night average and a 5-minute chart, and explicitly contrasts itself with methods that sample only part of the night. Garmin also states that this is really pulse-rate variability rather than heart-rate variability — an optical measurement, not an electrical one.

Google Health — the renamed Fitbit app since May 2026 — gives a Sleep Score with published bands and six contributors whose weights are not published, a Readiness score, SpO2, breathing rate, RMSSD, resting heart rate and skin temperature variation. The old duration-quality-restoration split no longer exists. Notably, its consumer accuracy page publishes no validation numbers at all, despite the company's own research existing.

WHOOP gives Recovery 0–100% from HRV, resting heart rate, respiratory rate and sleep duration, with the weighting unpublished.

Apple Watch gives overnight heart rate, respiratory rate, wrist temperature, blood oxygen and, on the newest hardware, an overnight HRV it calls recovery HRV, with the algorithm undisclosed. Its Vitals feature flags when several metrics fall outside your typical range at once.

Samsung gives a Sleep Score 1–100 whose band cut-offs are not published, five sleep score factors whose definitions and weights are not published, and an Energy Score. Samsung does not name its HRV statistic anywhere in its own documentation.

The most useful overnight number is the least discussed

Resting heart rate overnight gets a fraction of the attention that HRV and sleep stages get, and it is measured with the lowest error of anything on the list.

It is measured more accurately than anything else on the list. In the 536-night comparison against ECG, resting heart rate errors were around 3% — roughly half the error of HRV on the same devices and the same nights. It is also stable: it does not swing by thirty percent because you turned over, and a change of four or five beats sustained across a week is unambiguous in a way an HRV change of the same proportion is not.

It is the one number a clinician will recognise without translation. Nobody in a consulting room knows what a Readiness of 62 means, and everybody knows what a resting heart rate that has climbed from 54 to 61 over two months means — or at least knows that it is worth asking about.

And it needs no baseline period to be interpretable, because it has a physiological scale of its own rather than a manufacturer's.

Oura additionally reports when your lowest heart rate occurred during the night. That timing is more informative than it looks: on an ordinary night the low point tends to arrive in the first half of sleep, and a low point that keeps arriving unusually late is a pattern worth noticing — commonly after alcohol or a late heavy meal, both of which keep the cardiovascular system busy into the night.

If you only ever look at one overnight metric, this is the one to pick.

Why does my overnight HRV change so much night to night?

Before reading anything into a bad night, it is worth knowing what reliably shifts overnight measurements. Alcohol is the most dramatic — it raises heart rate and suppresses HRV for most of the night, and the effect is often visible at quantities people consider small. A late large meal does something similar and smaller. Hard training within a few hours of bed does it too, which is why an evening session frequently produces a poor recovery score the next morning without meaning the training was a mistake.

Illness, including the day or two before you notice symptoms, raises resting heart rate. So does a warm room, dehydration and, for many people, the second half of the menstrual cycle. Altitude does it. Some medications do it, including common ones.

None of this is exotic, and none of it is a fault in the device. It is a reminder that a single overnight value has several plausible explanations, and that the interesting signal is the one that persists across a week when none of the obvious causes apply.

Which is also the argument for looking at these numbers weekly rather than daily. Nightly variation is large enough that a single reading tells you mostly about last night's circumstances, while a weekly average tells you something about you.

Temperature is always a deviation, never a temperature

Every device here reports skin temperature as a deviation from your own baseline, not an absolute value, and skin temperature is not body temperature. What differs is how long the baseline takes to form: three nights on Garmin, three nights refined over thirty days on Google, roughly the first fortnight on Oura, not published on Samsung.

This matters more than it sounds. "Your temperature was elevated last night" means something different when the comparison is three nights old versus thirty. Two devices can disagree about the same night purely because of baseline length.

Whether you can see the data underneath

This is where the devices separate most sharply, and it has nothing to do with accuracy.

Apple exposes per-beat timestamps through HealthKit, so any app can recompute RMSSD, SDNN or anything else from the intervals themselves.

Samsung exposes raw beat intervals in milliseconds — but only through its Privileged Health SDK, which requires partner approval, and Samsung states that this SDK does not share data with Samsung Health. So the intervals a developer can read are not the ones Samsung's own scores are built from.

Garmin exposes 5-minute RMSSD values to consumers. Per-beat intervals overnight exist only through a licensed partner channel with a commercial fee.

Oura computes beat intervals internally but exposes nothing finer than five minutes. There is no raw-interval endpoint.

Google Health exposes 5-minute HRV aggregates and no intervals. Its legacy API is being retired this month in favour of a new one.

WHOOP exposes one HRV value per night and nothing underneath it.

So on one end you can audit the calculation yourself, and on the other you can accept the number or ignore it. For anyone who wants to check whether a score is behaving sensibly on their own body, that is the most consequential difference in this comparison.

How long before the HRV baseline means anything?

Recovery metrics are all relative to your own history, so the warm-up period is part of the product.

Garmin's HRV Status needs three weeks with at least four nights a week. Google's Readiness needs seven nights, and its typical ranges take up to thirty days. Oura's Readiness settles over about two weeks, with some features — cardiovascular age, blood pressure signals — needing two to four weeks of nights. Samsung's coaching needs seven days. WHOOP declares itself calibrated after four nights, though its marketing describes thirty.

Before that, you are being compared against population estimates for your age and sex rather than against yourself.

What the validation actually shows, and what it does not cover

Two things are worth knowing before trusting any published accuracy figure.

The funding pattern is visible in the results. Oura's four-stage agreement with laboratory sleep scoring was 0.65 on Cohen's kappa in a study Oura funded, and 0.31 in an independent unfunded study on a comparable task. Both are real studies. The gap is a reason to weight independent work more heavily, not a reason to dismiss either.

Almost nothing current has been validated. There is no peer-reviewed comparison against laboratory sleep measurement for Oura Ring 4 or 5, the Galaxy Ring, current Galaxy Watches, current Pixel Watches, WHOOP 5.0 or MG, Apple Watch Series 9 through 12, Fitbit Charge 6, or any current Garmin. Every accuracy number published anywhere — including in this article — comes from earlier hardware.

There is also a finding that rarely appears in comparisons. A 2026 study with home sleep measurement found that every device tested degraded substantially in older adults: total sleep time errors moved from roughly 15 to 23 minutes in younger participants to around 75 minutes in older ones. Accuracy figures collected on young healthy volunteers do not transfer.

And across independent studies, specificity for detecting wake runs between roughly 18% and 57%. Because wake gets misallocated into sleep at device-specific rates, every percentage downstream — deep, REM, efficiency — inherits a device-specific error in its denominator.

What the documented differences come down to

Stated as facts, with the limits of each attached.

Measured HRV error. In the only multi-device overnight ECG comparison, the newest Oura ring had the lowest error — on 13 participants, with a spread narrow enough that a different sample could reorder the middle of the table.

Access to the underlying data. Apple exposes per-beat intervals to any app. Samsung exposes them only through a partner-gated channel that does not feed its own app. Garmin, Oura and Google expose five-minute aggregates. WHOOP exposes one value per night.

Measurement window. Garmin and Oura average across the night. Google uses the main sleep period. WHOOP uses the last deep-sleep cycle. Apple reports a different statistic altogether.

Cost structure. WHOOP has no hardware price and a membership the device requires to function. The others are bought once, with health features carrying no subscription footnote in their specifications.

How much method is published. Oura publishes contributor lists, sampling intervals and a named HRV statistic. Samsung publishes none of those three.

What no device offers is a validated claim that its recovery number corresponds to anything about your day. That is uniform across the category.

What none of this can do

No overnight metric diagnoses anything. A low HRV night is consistent with a hard session, alcohol, a late meal, a coming infection, poor sleep or nothing identifiable.

No overnight metric predicts your day. No company here claims a validated predictive function, and the evidence does not support one.

No overnight metric transfers between devices. If you switch, your history starts again, because the windows and statistics differ.

How to bring this up with your doctor

Bring resting heart rate over weeks rather than any composite score. It is the most interpretable output a consumer device produces and the one a clinician is most likely to engage with.

Name the statistic and the device. "My overnight RMSSD has fallen from about 45 to about 30 over six weeks on the same ring" is a usable sentence. "My recovery is red" is not.

A persistent unexplained change in resting heart rate or HRV is a reason for a conversation, not a conclusion.

How Welltory fits

Welltory reads the devices above rather than competing with them, pulling overnight data through Apple Health, Health Connect, Samsung Health, Oura and Garmin, and reading Apple Watch, Samsung Watch and Bluetooth heart-rate monitors directly. It names the statistics it computes — SDNN, RMSSD, pNN50 and others — rather than presenting a composite whose inputs are undisclosed.

Like every product above, Welltory holds no regulatory clearance and no validated claim that any output corresponds to your day. Its published validation, 26 participants against simultaneous ECG, covers the measurement rather than a score.

What that leaves is the problem this article opened with. Each company computes a differently-defined number and calls it overnight HRV, so switching devices restarts your history and comparing two of them tells you nothing. Reading your existing device and computing named statistics from it is how you get a series that is comparable with itself — plus a controlled morning reading to set beside the overnight one.

How we made it

Facts about other companies come from their own support pages, developer documentation and API references, retrieved on 22 September 2026, and from named peer-reviewed studies. Where a company publishes inconsistent descriptions of its own method, that is noted rather than resolved. Where a figure is not published, the article says so instead of estimating. Study funding is disclosed where it is relevant to the result.

Made with AI tools, then edited and fact-checked by people.

Data analysis by Jane Smorodnikova — the founder and CEO of Welltory. A recognized tech leader with two Master's degrees and experience at MIT, she has scaled Welltory to over 17 million users.

Written by Tatsiana Yashyna — Deputy COO at Welltory. With a background in medicine and years of working with health data, she translates research and real physiological signals — sleep, stress, heart rate, and hormones — into clear, evidence-based explanations that help people understand what their bodies are telling them.

Discounts for blog readers: up to 36% off

See what affects your energy, stress, sleep, and daily state with Welltory

Explore your data