Consumer wearables surface a single HRV number and call it your "recovery" or "stress" score. That number is a compressed, filtered, approximated version of a much richer signal — and for a lot of applications, the approximation hides the thing you actually care about.
Start with what's under the hood
HRV is not a number. HRV is a category.
When your heart beats, you can measure the interval between consecutive beats — the R-R interval, so named for the R peaks on the ECG trace that mark each ventricular contraction. A typical resting adult at 60 BPM does not produce 60 beats evenly spaced at 1.000 seconds. Instead you get a sequence like: 987 ms, 1052 ms, 998 ms, 1021 ms, 1088 ms, 945 ms, 1012 ms...
That's your HRV. Not any single number — the sequence of intervals, and all the statistical patterns you can extract from it. A five-minute recording at rest might contain 300 of these intervals. Analysts have spent the better part of fifty years inventing ways to summarize them.
The resulting zoo of metrics breaks into three families.
Time-domain metrics
These treat the sequence of R-R intervals as a time series and compute descriptive statistics on it directly.
- SDNN: standard deviation of the whole series. Captures total variability, including slow rhythms that take many seconds to play out. Sensitive to both branches of the autonomic nervous system.
- RMSSD: root-mean-square of the differences between successive intervals. Captures beat-to-beat variability, which is almost entirely vagal.
- pNN50: the percentage of successive intervals that differ by more than 50 ms. Another vagal indicator, highly correlated with RMSSD.
SDNN and RMSSD answer different questions. If a person's heart rate swings slowly from 60 to 75 BPM and back over a two-minute period but individual successive beats are nearly identical, their SDNN will be high and their RMSSD will be low. That's a meaningful physiological distinction. Compressing it to one number throws it away.
Frequency-domain metrics
These take the R-R sequence, interpolate it into a continuous signal, run a Fourier transform, and report the power in various frequency bands.
- VLF (very-low-frequency, 0 – 0.04 Hz): power in oscillations slower than 25 seconds. Mechanism poorly understood; correlated with thermoregulation, renin-angiotensin activity, and possibly mortality risk.
- LF (low-frequency, 0.04 – 0.15 Hz): power in 6 – 25-second oscillations. Contains the baroreflex resonance frequency. Mixed sympathetic/parasympathetic.
- HF (high-frequency, 0.15 – 0.4 Hz): power in 2.5 – 7-second oscillations. Contains respiratory sinus arrhythmia at normal breathing rates. Almost pure vagal.
Frequency-domain analysis is where resonance frequency work lives. When you breathe at 6 BPM, your RSA oscillation has a period of 10 seconds, which puts its frequency at 0.1 Hz — smack in the middle of the LF band. Coherence breathing is, in a sense, the deliberate project of shoveling variability from HF into LF by slowing your breath.
Non-linear metrics
Newer metrics trying to capture patterns that linear statistics miss: entropy measures (SampEn, ApEn), Poincaré-plot geometry (SD1, SD2), detrended fluctuation analysis (DFA α1, α2). These are increasingly popular in sports science for their sensitivity to autonomic state during exercise. For coherence breathing work they're less directly relevant; we'll leave them here.
What wearables actually give you
Now the hard part. Most consumer wearables report a single number called "HRV" on their home screen. What is that number?
For most — Oura, Whoop, Apple Watch's HRV feature, Garmin's HRV Status, Fitbit's HRV — it is RMSSD, computed over some overnight or morning window, often with the specific method not clearly documented.
This is a reasonable choice. RMSSD is a decent single-number summary of vagal activity, it's robust to small amounts of missing data, and it tracks interesting lifestyle factors (sleep quality, training load, alcohol). It's also what most consumer HRV research literature uses.
But there are four problems with leaning on it as the HRV metric:
Problem 1: It can't see the frequency-domain picture
An all-RMSSD view of HRV is blind to the thing coherence breathing is trying to change. During a coherence-breathing session, RSA power moves from HF to LF and the LF peak becomes sharp and narrow-band. RMSSD picks up some of this but not cleanly. You can have an excellent resonance-frequency session and see only a modest RMSSD bump afterward, because RMSSD weighs all beat-to-beat changes equally whether they're organized into a tall sine wave or scattered as noise.
If your practice is coherence-oriented, watching an RMSSD tracker for the benefit is like watching the scale after a weights workout — it's the wrong instrument for that question.
Problem 2: The sensor isn't a real ECG
Most wrist-worn devices use photoplethysmography (PPG): shine a green LED at the skin, measure how reflected light changes as blood pulses through the capillaries, and infer each heartbeat from the pulse waveform. Under optimal conditions — you're still, warm, well- hydrated, the device is snug against your wrist — PPG can be quite accurate for average heart rate.
For beat-to-beat timing precision, which is what HRV analysis requires, it's soft. The peak of a PPG pulse wave arrives at your wrist 200 ms or so after the corresponding R peak at the heart, and that delay varies slightly with every beat depending on blood pressure, arterial tone, limb position, and skin perfusion. So even when the sensor correctly detects a pulse, it may report a timing that's 10 – 30 ms off from the true R-R interval.
Ten milliseconds sounds trivial. It isn't, for HRV. Normal resting RMSSD is 20 – 50 ms. A random timing error of 10 ms on every beat is large relative to the signal you're trying to measure. Nelson et al. (2020) found that consumer wrist devices can reproduce overnight RMSSD trends acceptably but cannot replicate short-window HRV metrics at the accuracy required for frequency-domain or biofeedback work.
Problem 3: The value is smoothed
Wearable manufacturers know their sensor data is noisy. Their response is to smooth heavily. Your Whoop "HRV" is not today's RMSSD — it's a rolling average over several days, often trimmed to exclude outliers, adjusted against your personal baseline. This makes the number stable and trendable, at the cost of throwing away within-day resolution.
If you run a 20-minute coherence session at 3 pm, your Whoop HRV will not move in response. It can't. The metric is constructed not to.
Problem 4: The interpretation is proprietary
Your Oura "recovery" score combines HRV with resting heart rate, body temperature, respiratory rate, and sleep stages into a single 1 – 100 value using a proprietary algorithm. Whoop's is similar. Garmin's "HRV Status" categorizes your overnight HRV as low / balanced / unbalanced / high based on a seven-day baseline against an unspecified population norm.
Each of these is defensible in its own terms. None of them is HRV. They are "a vendor-specific wellness score derived partly from HRV." Treating them as HRV, and then trying to use them to evaluate an intervention like coherence breathing, is a common and frustrating mistake.
What a chest strap actually gives you
A chest-strap heart-rate sensor — Polar H10 is the standard — is effectively a miniature two-lead ECG. It detects the R peak directly from the electrical signal of the heart, reports each R-R interval at millisecond precision over Bluetooth, and makes no attempt to smooth or cook the data.
The validation literature on these devices is extensive. The Polar H10 in particular has been benchmarked against clinical 12-lead ECG and found to reproduce HRV metrics with errors typically under 2% across time-domain and frequency-domain measures. For research purposes, it's generally accepted as a reference sensor.
The cost of this is that you have to wear a strap across your chest for the duration of the recording. For overnight tracking this is uncomfortable enough that most people won't do it, which is exactly why wrist sensors dominate the market. For a 20-minute coherence breathing session, it's trivial.
Chest strap for the session. Wearable for the lifestyle. They're measuring overlapping but different things, with different accuracies, at different time scales. You can have both.
What to actually do
- Keep tracking overnight HRV on your wearable if you already do it. It's a reasonable lifestyle signal, sensitive to alcohol, poor sleep, illness, overtraining. Just don't expect it to move cleanly in response to a meditation practice.
- For coherence-breathing work, use a chest strap and an app that shows you the frequency-domain signal. You need to see the LF peak rise and narrow as you lock on to your resonance frequency. Without that feedback you're flying blind, which defeats the point of doing a measurement-based practice.
- Don't average the two together. The RMSSD on your wrist and the in-session coherence on your chest strap answer different questions. Trying to reconcile them into a single "HRV score" is what started this confusion in the first place.
Further reading
- The science of resonance frequency breathing — the full background
- Heart coherence in plain English — what coherence actually measures
- The research bibliography — Nelson 2020, Shaffer 2017, Laborde 2017 all worth reading
- The glossary — if any term here was unfamiliar