Can I trust my ring or watch's overnight HRV?
As your own trend, mostly: Oura and WHOOP were off a chest strap by 6-8% on average, Garmin Fenix 6 and Polar Grit X Pro by 11-16%. Resting HR: about 1-2 bpm. 13 people.
Dial and colleagues asked how closely the overnight resting heart rate (RHR) and HRV (RMSSD) shown by five consumer wearables match a reference.
Thirteen healthy adults (7 men, 6 women) slept at home wearing a Polar H10 chest strap (recorded through a separate, unworn Polar watch) together with any of a Garmin Fenix 6, Oura Generation 3 and Generation 4 rings, a Polar Grit X Pro and a WHOOP 4.0 (one device per wrist), for 536 nights in all; participants with at least 10 nights were included.
The strap's RR intervals were processed in Kubios and averaged over the night (over the first 4 h for comparison with the Polar watch, which reports only that window). Resting heart rate agreed closely for the four devices compared: concordance 0.97 (Oura Gen 3), 0.98 (Gen 4), 0.91 (WHOOP) and 0.86 (Polar), MAPE 1.67-3.00%, mean absolute error 0.98-1.78 bpm.
Garmin was left out of the resting-heart-rate comparison because it reports the lowest 30-min average of the 24 h without saying when that was. HRV agreement differed more: Oura Gen 4 concordance 0.99 (MAPE 5.96%), Gen 3 0.97 (7.15%), WHOOP 0.94 (8.17%), Garmin 0.87 (10.52%) and Polar 0.82 (16.32%); Polar's HRV error was larger than every other device's.
Errors persisted when each night was expressed as a z-score against the person's own reference baseline, which the authors read as coming from the devices' processing rather than from differences between people.
Our reading: the most direct test the base has of the overnight numbers behind a readiness check-in — resting heart rate is trustworthy on all four devices compared, HRV is trustworthy to within about 6-8% on the Oura and WHOOP devices tested and less so on the Garmin and Polar watches tested — from 13 people, with reporting errors that limit what the Polar rows can be trusted for.
Protocol
Participants: n = 13 healthy adults (7 men, 6 women; 33.2 ± 8.6 y), over 18, without conditions that chronically impair sleep; recruited at Wright-Patterson Air Force Base and The Ohio State University; included if they completed at least 10 nights. Verbal consent; approved by the Air Force Research Laboratory IRB (FWR202200227N) as non-human-subjects research.
Wearables: Garmin Fenix 6 (Elevate V3 optical sensor), Oura Generation 3 and Generation 4 rings, Polar Grit X Pro and WHOOP 4.0, all green/red-LED photoplethysmography, all reporting HRV as RMSSD; one device per wrist; ring on for at least 30 min before bed and after waking; apps synced each morning. Firmware and app versions not reported; nights per person per device not reported.
How each device defines its numbers (manufacturer documentation, as reported): Garmin — resting heart rate is "the lowest 30 min average in a 24 h period", HRV the average of 5-min windows over detected sleep; Oura — heart rate in 10-min segments, HRV from 5-min samples averaged over the night (Gen 3 samples at 250 Hz); Polar Grit X Pro — heart rate at 1 Hz, resting heart rate and HRV from the first 4 h after sleep onset; WHOOP — 52 Hz, a "dynamic average during sleep… weighted towards your last slow wave sleep stage".
Criterion: Polar H10 single-lead ECG chest strap (1000 Hz), RR intervals exported per night and processed in Kubios HRV Scientific (automatic noise detection "medium", automatic beat correction); nights with 5% or more corrected beats excluded; sleep onset judged by a drop in heart rate, typically 10-15 min after the start; resting heart rate and HRV averaged over the whole night (first 4 h for Polar; the whole-night average for WHOOP despite its weighting).
Nights analysed: resting heart rate — Oura Gen 3 470, Gen 4 138, Polar 206, WHOOP 288; HRV — Garmin 150, Oura Gen 3 470, Gen 4 139, Polar 206, WHOOP 289.
Statistics: Pearson r; Lin's concordance (< 0.80 unacceptable, 0.80-0.89 poor, 0.90-0.95 moderate, 0.95-0.99 substantial, > 0.99 nearly perfect); MAPE (> 10% unacceptable); Bland–Altman bias and limits of agreement; mean absolute error; one-way ANOVA with Tukey tests; within-person z-scores against each participant's H10 mean and SD over 30 days.
Funding and interests: Air Force Research Laboratory (award S-194-11T-001); the authors declare no competing interests; how the devices were obtained is not stated and no manufacturer involvement is reported.
Key Results
Resting heart rate (bpm) against the chest strap (Table 1):
| Oura Gen 3 | Oura Gen 4 | Polar Grit X Pro | WHOOP 4.0 | |
|---|---|---|---|---|
| Nights | 470 | 138 | 206 | 288 |
| Mean bias ± SD | −0.88 ± 1.00 | −0.94 ± 1.43 | −0.01 ± 2.13 | −1.41 ± 1.69 |
| Limits of agreement | −2.84 to 1.08 | −3.75 to 1.87 | −5.36 to 2.06 | −4.72 to 1.90 |
| Mean absolute error | 0.98 ± 0.90 | 1.08 ± 1.33 | 1.72 ± 1.30 | 1.78 ± 1.31 |
| MAPE | 1.67 ± 1.54% | 1.94 ± 2.51% | 2.71 ± 2.75% | 3.00 ± 2.15% |
| Concordance (CCC) | 0.97 | 0.98 | 0.86 | 0.91 |
HRV, RMSSD in ms (Table 3):
| Garmin Fenix 6 | Oura Gen 3 | Oura Gen 4 | Polar Grit X Pro | WHOOP 4.0 | |
|---|---|---|---|---|---|
| Nights | 150 | 470 | 139 | 206 | 289 |
| Mean bias ± SD | −1.84 ± 6.86 | −2.50 ± 4.56 | −0.96 ± 5.52 | −4.65 ± 9.67 | −0.78 ± 5.98 |
| Limits of agreement | −15.218 to 11.60 | −11.43 to 6.43 | −11.78 to 9.85 | −14.30 to 23.60 | −12.50 to 10.94 |
| Mean absolute error | 5.29 ± 4.72 | 3.91 ± 3.40 | 3.93 ± 3.98 | 7.27 ± 7.88 | 4.17 ± 4.33 |
| MAPE | 10.52 ± 8.63% | 7.15 ± 5.48% | 5.96 ± 5.12% | 16.32 ± 24.39% | 8.17 ± 10.49% |
| Concordance (CCC) | 0.87 | 0.97 | 0.99 | 0.82 | 0.94 |
Within-person z-scores (Tables 2 and 4), mean absolute error / concordance: resting heart rate — Oura Gen 3 0.15 / 0.97, Gen 4 0.16 / 0.98, Polar 0.23 / 0.92, WHOOP 0.29 / 0.88; HRV — Garmin 0.51 / 0.77, Oura Gen 3 0.22 / 0.94, Gen 4 0.27 / 0.91, Polar 0.40 / 0.79, WHOOP 0.32 / 0.86.
Between devices: the two Oura rings did not differ from each other on any measure; for HRV, Polar's bias and MAPE were significantly larger than every other device's, and Garmin's MAPE larger than both Ouras' and WHOOP's.
Numbers that do not add up (our arithmetic):
Polar resting heart rate: a bias of −0.01 ± 2.13 bpm would give limits of about −4.2 to +4.2, not −5.36 to 2.06, whose midpoint is about −1.65; but the reported mean absolute error (1.72 ± 1.30 bpm) fits a bias near zero with SD 2.13 (1.72² + 1.30² ≈ 4.6 ≈ 0.01² + 2.13²), not a bias of −1.65 with SD 1.89 (≈ 6.3), so the printed limits, not the bias, are the likelier error; the Oura and WHOOP rows pass the same check (our arithmetic). The other three rows check out.
Polar HRV: a bias of −4.65 ± 9.67 ms would give limits of about −23.6 to +14.3; the table prints −14.30 to 23.60, whose midpoint is +4.65 — one sign is flipped. The other four rows check out.
The resting-heart-rate ANOVA is reported with "Fenix 6: highest bias", although the Fenix 6 was excluded from that analysis.
Within-person concordances in the text (resting heart rate: Gen 4 0.96, Gen 3 and Polar 0.93, WHOOP 0.87; HRV: Gen 3 0.84, WHOOP 0.76) do not match Tables 2 and 4 (0.98, 0.97, 0.92, 0.88; 0.94, 0.86).
The Conclusion gives the Polar Grit X Pro "intermediate" HRV accuracy and the Garmin "notably lower concordance", the reverse of the Results and abstract (Polar concordance 0.82 and MAPE 16.32%, the largest error of all devices).
The Methods' MAPE formula is signed; the reported MAPEs are all positive, so absolute values were presumably taken.
Mechanisms
Measured: device error persisted after each night was standardised against the person's own baseline, most for the Garmin and Polar watches.
Authors' interpretation: the devices differ in sampling rate, the window of the night used, the weighting of sleep stages, and undisclosed artefact filtering, signal-quality rules and interpolation; error that survives z-scoring points to processing, sensor design and segmentation rather than to physiology.
Inference: part of the disagreement is definitional rather than sensor error — WHOOP weights HRV to the last slow-wave sleep but was compared with a whole-night strap average, while the authors matched Polar's first-4-h window on the strap, so Polar's larger error is not explained this way; a device can be faithful to its own definition and still differ from another brand's number.
Inference: overnight resting heart rate is the steadier number (mean absolute error 0.98-1.78 bpm), well under the 5-7 bpm the authors cite as clinically meaningful, but of the order of the night-to-night shifts athletes read — REF-1193 found nocturnal heart rate 3.9 bpm higher and HRV 14.1% lower after a late maximal session than a light one, averaged over many nights; for a single night that is about twice the WHOOP 4.0's resting-heart-rate error here (mean absolute error 1.78 bpm) and less than twice its HRV error (MAPE 8.17%) (our arithmetic).
Practical Application
Can I trust my ring or watch's overnight HRV? As your own trend, mostly: Oura and WHOOP were off a chest strap by 6-8% on average, Garmin Fenix 6 and Polar Grit X Pro by 11-16%. Resting HR: about 1-2 bpm. 13 people.
Compare a night with the athlete's own baseline on the same device, never across brands or after a device change (Not from this paper; the base's own practice — supported here by the different windows each brand measures, and by REF-364); a new device starts a new baseline.
A one-night HRV dip smaller than the device's error is not a signal: 95% of single nights fell within about −11.8 to +9.9 ms of the strap on the Oura Gen 4, but within about −15.2 to +11.6 ms on the Garmin Fenix 6 and −14.3 to +23.6 ms on the Polar Grit X Pro (as printed); read the 3-day direction against a 28-day baseline as SYNTH-30 does (the window lengths are the base's practice; averaging rather than single nights, REF-361).
Lean on resting heart rate when the two disagree (Not from this paper; the base's practice, from its smaller error here): it agreed within limits of about ±5 bpm on all four devices compared (B-, 13 non-athletes; Polar's printed limits conflict with its own bias and SD).
Garmin's resting heart rate is not a whole-night value: it is the lowest 30-min average in 24 h, so it is not comparable with another device's night average; its HRV (Fenix 6; newer Garmin watches untested here) agreed less well than the rings'.
Brand and model matter: these results are for the named models; accuracy of one model is not the brand's, and firmware updates can change it.
Limitations
Thirteen people. Many nights (536) but few participants; how many people contributed to each device is not reported, so the Oura Gen 4 (139 nights) and Garmin (150) rows may rest on a handful of sleepers.
Criterion: a chest strap, not a clinical ECG, though one that agreed with a 12-lead ECG within a few milliseconds (REF-1250); sleep onset was judged by eye from the heart-rate drop.
Different definitions: devices report different parts of the night; WHOOP was compared with a whole-night average despite weighting to late slow-wave sleep, which inflates its apparent error by an unknown amount.
Garmin resting heart rate untested; only one Garmin model.
Who: healthy adults, training status not reported; no exercise, illness, alcohol or arrhythmia nights were studied deliberately.
Reporting: contradictory Polar rows, text-versus-table concordances and a Conclusion inconsistent with the Results (see Key Results); firmware and app versions absent.
Interests: Air Force Research Laboratory funding; the authors declare no competing interests; no manufacturer role is reported.
How this entry was checked and changed
Why this level
a test validation study (rubric B) with many nights (536) but few people (n = 13) and a chest-strap rather than clinical ECG criterion, read from the full text (open access, CC BY 4.0). Held at B- for a serious limitation beyond the design: several reported numbers contradict each other (the Polar rows of both agreement tables, within-person concordances in the text against the tables, an ANOVA statement about a device excluded from that analysis, and a Conclusion that ranks the Polar watch's HRV above the Garmin's against its own Results), and neither firmware nor nights per person are reported. The Oura, WHOOP and Garmin HRV rows check out arithmetically
Full text (checked 2026-10-27)
Thirteen healthy adults (6 females) wore an ECG reference and multiple wearables simultaneously during sleep, totaling 536 nights (full text)
Thirteen subjects (male: n = 7; female: n = 6, age = 33.2 ± 8.6 years); participants that completed at least 10 nights were included (full text)
The Polar H10 was used as the criterion measurement for heart rate: single-lead ECG with a sampling frequency of 1000 Hz (full text)
Resting heart rate: Oura Gen 3 bias −0.88 ± 1.00 bpm, limits −2.84, 1.08; Oura Gen 4 −0.94 ± 1.43, limits −3.75, 1.87; Polar Grit X Pro −0.01 ± 2.13, limits −5.36, 2.06; WHOOP 4.0 −1.41 ± 1.69, limits −4.72, 1.90 (full text)
Resting heart rate MAPE 1.67 ± 1.54 (Oura Gen 3), 1.94 ± 2.51 (Gen 4), 2.71 ± 2.75 (Polar), 3.00 ± 2.15 (WHOOP); CCC 0.97, 0.98, 0.86 and 0.91 (full text)
HRV: Garmin Fenix 6 bias −1.84 ± 6.86 ms, limits −15.218, 11.60; Oura Gen 3 −2.50 ± 4.56, limits −11.43, 6.43; Oura Gen 4 −0.96 ± 5.52, limits −11.78, 9.85; Polar −4.65 ± 9.67, limits −14.30, 23.60; WHOOP −0.78 ± 5.98, limits −12.50, 10.94 (full text)
HRV MAPE 10.52 ± 8.63 (Garmin), 7.15 ± 5.48 (Oura Gen 3), 5.96 ± 5.12 (Gen 4), 16.32 ± 24.39 (Polar), 8.17 ± 10.49 (WHOOP); CCC 0.87, 0.97, 0.99, 0.82 and 0.94 (full text)
The Garmin Fenix 6 calculates RHR as "the lowest 30 min average in a 24 h period" and was omitted from the RHR analysis (full text)
Polar reports RHR and HRV as an average value only for the first 4 h of sleep; for Whoop 4.0 the all night HRV average was utilized (full text)
This study was financially supported by the Air Force Research Laboratory (AFRL); the authors declare that they have no competing interests (full text)