What HRV actually is

The measurement, precisely

A healthy heart does not beat like a metronome. If your resting heart rate is 60, you are not getting one beat every 1,000 milliseconds — you are getting one at 1,020 ms, the next at 960, the next at 1,005. That scatter is the thing being measured. The average is your heart rate; the variation around it is your heart rate variability.

Those gaps are called R-R intervals, after the R wave — the tall spike in an ECG trace that marks the ventricles contracting. HRV is arithmetic performed on a list of R-R intervals, and every argument about HRV accuracy is ultimately an argument about how cleanly a device can produce that list.

RMSSD, and why it is the one you see

There are several ways to summarize the scatter. The one nearly every consumer wearable reports is RMSSD — the root mean square of successive differences. Mechanically: take each pair of neighboring intervals, subtract one from the other, square the result, average the squares, take the square root.

The reason RMSSD won is that it looks specifically at successive beats rather than at the spread of the whole recording. Beat-to-beat changes are dominated by the vagus nerve, which acts on the heart within a single beat. Slower influences — hormones, temperature drift, the sympathetic branch — cannot move the interval that fast. So RMSSD isolates parasympathetic activity better than the alternatives, and it stabilizes on shorter recordings, which matters when your measurement window is five minutes rather than twenty-four hours.

You will occasionally see SDNN, the standard deviation of all the intervals in the recording. It captures more of the total variability, including the slow components, and is much more sensitive to how long the recording ran. A five-minute SDNN and an overnight SDNN are not the same quantity. If your device changes which metric it reports, or which window it computes over, your number moves for reasons that have nothing to do with you.

Why the physiology makes it a stress signal rather than a fitness one

Two branches of the autonomic nervous system act on the heart continuously. The parasympathetic branch, acting through the vagus nerve, slows it. The sympathetic branch speeds it up. Vagal signaling is fast and is modulated by your breathing, which is why the interval between beats shortens slightly as you inhale and lengthens as you exhale.

When you are recovered and unstressed, vagal tone is high and that breathing-linked modulation is pronounced — the intervals scatter more, and RMSSD is higher. When you are under load — hard training, illness, alcohol, poor sleep, psychological stress — sympathetic activity rises, vagal tone is suppressed, the intervals become more uniform, and RMSSD drops.

That is the entire mechanism behind every recovery score you have ever seen. It is real physiology, and it is why a suppressed HRV genuinely is informative. What it is not is a measure of fitness, and a low number does not mean your heart is unhealthy.

The single most misunderstood thing about it

Absolute HRV values are not comparable between people. Not roughly comparable. Not comparable with an age adjustment. The between-person spread in this measurement is enormous — driven by age, genetics, resting heart rate, body position, breathing rate and where on the body the reading was taken — and it dwarfs the within-person changes anybody is actually trying to detect.

Two equally fit, equally recovered runners can sit forty milliseconds apart. A person at 30 ms who normally runs 28 is in better shape than a person at 70 ms who normally runs 95. Every legitimate use of HRV is a comparison of your number today against your own recent distribution of numbers, which is why every serious platform builds a rolling baseline before it will show you anything.

This is also why HRV leaderboards, forum comparisons and "is 45 good?" questions have no answer. There is no population norm you can usefully place yourself against.

Why the measurement is harder than it looks

To compute RMSSD you need beat timings accurate to a few milliseconds. That is a much harder requirement than heart rate, where an error of a beat or two per minute is invisible.

An electrical chest strap detects the R wave directly. The R wave is a sharp, high-amplitude electrical event with an unambiguous peak, which is precisely why it is the feature clinical instruments key on and why a strap is the reference researchers measure everything else against.

An optical sensor cannot see the R wave at all. It sees a pressure wave arriving at the skin some time after the heart contracted, and it estimates the peak of a smooth, rounded waveform rather than a spike. Two extra sources of error come with that: the rounded peak is harder to time precisely, and the delay between contraction and arrival — pulse transit time — is itself variable, changing with blood pressure and vessel tone. Some of the "variability" an optical device measures is the variability of the plumbing rather than of the heart.

That is the mechanism behind the spread in the measured evidence. The device-by-device numbers are on the evidence page, where five wearables were worn simultaneously against a single-lead ECG for 536 nights and the error ranged from roughly 6% to over 16%.

Overnight or morning? The two protocols are not interchangeable

Devices produce an HRV figure in one of two ways, and mixing them ruins a baseline.

Overnight, automatically. Rings and bands sample during sleep, usually across a defined window or the deepest portion of the night. The advantage is enormous compliance — it happens whether you remember or not — and a long, still recording with no motion artifact. The disadvantage is that you do not control what window it used, and the window definition is proprietary and can change with a firmware update.

Morning, deliberately. You lie still for two to five minutes after waking, breathing normally, with a strap and an app that computes RMSSD. This gives you a standardized, reproducible protocol you fully control. The disadvantage is that it is a routine, and routines fail.

Both are defensible. What is not defensible is comparing today's overnight figure to last month's morning figure, or switching devices mid-baseline and reading the step change as a physiological event. Pick one protocol and one device, and treat a change of either as starting the baseline over.

What genuinely moves it

Before you interpret a drop as overtraining, this is the realistic list of what else moves the number, roughly in descending order of how often it is the actual explanation:

• Alcohol. The most reliable HRV suppressor most people will ever encounter, and it commonly persists into a second night.
• Illness, including presymptomatic illness. Often the first thing to move, before you feel anything.
• Late or large meals. Digestion during the measurement window competes with the state you were trying to measure.
• Heat. A hot room, or heat acclimatization, raises heart rate and suppresses variability.
• Psychological stress and short sleep. Indistinguishable, in the data, from training stress.
• Genuine training load. Real, and last on this list rather than first — which is the practical point.

A single low morning is weak evidence of anything. A three-to-five-day depression below your normal band, with no obvious lifestyle explanation, is the pattern worth acting on.

What HRV cannot tell you

Which stressor caused the change. The autonomic nervous system does not label its inputs. A hard session, a bad argument, a head cold and two glasses of wine produce the same directional signal.

What to do about it. There is no measured mapping from a given HRV decrement to a specific training prescription. The recovery scores that appear to provide one are proprietary interpretations, and the authors of the best validation in this field noted directly that there is little transparency into what feeds them and that they are updated without notice.

Anything about your heart's health.These are consumer devices, not diagnostic instruments. HRV is studied clinically, but a wearable's figure is not a clinical measurement and should not be read as one.

Getting a number you can actually use

1. Pick one device and one protocol and do not change either for at least a month. A baseline assembled from two measurement methods is not a baseline.
2. Wear it every night, or measure every morning. Intermittent data produces a baseline made mostly of noise, and no hardware upgrade rescues that.
3. Improve the input if you can. An electrical strap times the R wave directly rather than inferring it — the upgrade path is here, and the cheapest option on it is also the most accurate.
4. Read the trend, not the night. Weekly means against your own rolling band. Ignore single-day movement almost entirely.
5. Log the obvious confounders. Alcohol, illness, late meals, room temperature. Most unexplained drops stop being unexplained.