Editorial analysis · Vendor documentation checked November 16, 2026
Two screens can both say “HRV,” both use milliseconds and still be reporting different quantities. Before deciding that one watch is wrong, ask a more basic question: did the apps tell you enough to compare them?
An HRV number needs a statistic, a time window and an averaging rule. The brand name is not an adequate substitute. A single sample, a nighttime average, the highest five minutes and a seven-day trend can all sit inside the same product.
We reviewed public documentation from Apple, Oura, WHOOP and Garmin and built the field reference below. This is an explanation of definitions, not a device accuracy contest, a diagnosis or a recommendation to change training because of one reading.
Compare the documented fields · See why the formulas differ · Check your own comparison
The name on the screen is only part of the record
| Source or field | Statistic | Window | Aggregation |
|---|---|---|---|
| Apple HealthKit SDNN field | SDNN | Discrete samples; inspect record timestamps and source | A sample is not automatically a nightly or daily average |
| Apple HealthKit RMSSD field | RMSSD identifier | Not specified by the checked symbol page | Not specified by the checked symbol page |
| Oura average nighttime HRV | rMSSD (manufacturer cardiovascular white paper) | Five-minute samples during sleep | Mean of the five-minute samples |
| WHOOP HRV | rMSSD | During sleep; checked support descriptions differ in specificity | Recovery page describes weighted whole-night average emphasizing slow-wave sleep and later night; HRV page says deepest sleep period |
| Garmin HRV Status | Exact statistic not specified in the checked HRV Status overview | Overnight; separate highest five-minute value | Overnight average and seven-day average are separate; status compares seven-day average with personal baseline |
Oura's 2026 cardiovascular white paper identifies its HRV measure as rMSSD. Its member documentation describes the nighttime average and distinguishes it from the maximum. Garmin's checked HRV Status overview explains its different windows and status comparison, but does not name the exact statistic there. We leave that field unresolved instead of borrowing a definition from a different Garmin feature.
Manufacturer documentation tells us how the manufacturer describes a field. It does not independently establish measurement accuracy. We did not wear these devices together, collect member records or reproduce their internal processing.
Apple makes the case for naming the field, not just the brand
Apple's established SDNN quantity is explicitly a standard-deviation measure. Its documentation says Apple Watch automatically records samples. That does not make every daily chart a standardized nighttime RMSSD reading.
There is also a newer detail that older comparison tables can miss: Apple's documentation now lists a separate RMSSD identifier, with iOS 27.0+ and watchOS 27.0+ availability. The symbol page does not explain automatic measurement, the sampling window or how every app displays it.
Consequently, “Apple uses SDNN” is too broad if it is meant to describe every supported HealthKit field, while “Apple HRV is RMSSD” is also inadequate. Ask which field the record uses, which software version produced it and which app is presenting it. An API's availability does not prove that a particular watch collected that quantity or that a receiving app uses it.
This is why a data integration should preserve the original field identity. Replacing a specific name with a generic “HRV” can remove the detail needed to interpret a change later.
Sometimes the documentation leaves a question open
WHOOP's checked Recovery page describes a weighted whole-night average that emphasizes slow-wave sleep and later parts of the night. Its HRV page names rMSSD but describes measurement during the deepest sleep period.
Those descriptions do not give us a single fully specified window and weighting algorithm. We retain both in the reference. We have not established whether the difference reflects shorthand, a documentation update or a product change. Repeating a confident “last five minutes” rule would add precision that these checked sources do not provide.
There is value in publishing that uncertainty. A reader comparing exports needs to know which details remain unavailable. A table with an honest unresolved field is more useful than one that fills every cell by assumption.
The same intervals can produce two different answers
SDNN describes the spread of normal-to-normal intervals. RMSSD describes the size of successive interval changes through a root-mean-square calculation. They are different summaries, even before a device selects its samples.
Consider a deliberately tiny synthetic sequence: 1,000, 1,020, 1,000, 1,020 milliseconds. This is a formula illustration, not a physiological recording or a valid device test.
The mean interval is 1,010ms. Using the sample standard deviation, the four deviations of 10ms give SDNN = √(400 ÷ 3), or about 11.55ms. Consecutive differences are +20, -20 and +20ms. Their squared mean is 400, so RMSSD = √400 = 20ms.
Neither answer is a failed attempt to calculate the other. The formulas ask different questions. The SDNN example explicitly uses the N-1 denominator; we are not claiming that this miniature calculation reproduces any manufacturer's implementation.
Now reorder the same intervals as 1,000, 1,000, 1,020, 1,020. The sample standard deviation stays about 11.55ms because the values and mean have not changed. RMSSD falls to √(400 ÷ 3), about 11.55ms, because successive changes are now 0, 20 and 0. That alone shows why there is no universal multiplier from one statistic to the other.
Averaging adds another layer
Even when two products name the same statistic, the displayed summaries can differ. An average of five-minute calculations is not automatically the same as calculating the statistic once over a combined night. Selecting different valid segments or giving segments different weights can also change the summary.
Garmin makes another potential mismatch visible in its own interface definitions: the previous night's average, the highest five-minute period and the seven-day average are separate views. Its status classification compares the seven-day average with a personal baseline. A status word and a millisecond value are not competing measurements of the same quantity.
The practical implication is modest but important: a disagreement is a question to investigate, not proof that a device is broken or that your health changed between opening two apps. Definition checks do not prove either device is accurate, but they prevent an invalid comparison from masquerading as an accuracy test.
What to keep beside a comparison
- Exact field and unit. Record SDNN, RMSSD, a transformed value or a proprietary score as applicable. Do not infer it from “ms” alone.
- Measurement window. Keep the start and end, sleep session or daily boundary. A daytime sample should not silently become a night summary.
- Aggregation and missing data. Note whether the display is a maximum, mean, weighted result or rolling average, and whether gaps are visible.
- Source and version. Keep the device, app and available algorithm version. A source change deserves an annotation in a trend.
If a value changes after switching devices, start by checking those details. Even a consistent personal trend is not a diagnosis. This article does not supply thresholds for deciding whether to train, rest or seek medical care.
Download the reference, including its unknowns
Download the field reference CSV, the sources, limits and arithmetic JSON, or a blank comparison log. No personal health data needs to be uploaded to use them.
Our earlier sleep-tracker comparison examines a related issue with sleep definitions. Here the contribution is a dated HRV field dictionary, explicit documentation gaps and reproducible arithmetic. We have not established a cross-device conversion, ranked current hardware or measured clinical outcomes. Original compilation and calculations may be reused with attribution and these limits. Corrections: support@askvora.com.
Publisher disclosure: we make Vora. Vora's proactive AI is built around connected wearable context alongside nutrition and workouts. Bringing information together creates a responsibility to preserve its meaning. See Vora's integrations for the product context; this reference does not claim that Vora imports every field listed here.