Research note 01
What HRV Actually Tells You About Readiness
Heart-rate variability is one of the few practical windows onto autonomic state that a club can measure daily. It is also the metric most often presented with a confidence it has not earned. Here is what the measurement is, what it responds to, and where the interpretation goes wrong.
Ask five performance departments what their readiness score means and you will get five answers, at least three of which involve heart-rate variability in some form. HRV has become the default physiological input to daily monitoring, partly because it is genuinely informative and partly because it is easy to collect: a chest strap, ninety seconds, an app. The ease of collection has run ahead of the discipline of interpretation.
This note sets out how we use HRV in the load and readiness module, and — more usefully — the specific ways we have seen it misread.
What is actually being measured
The heart does not beat metronomically. The interval between successive R-waves varies from beat to beat, and that variation is modulated by the autonomic nervous system: parasympathetic (vagal) activity increases beat-to-beat variability, sympathetic activity reduces it. Measuring the variation therefore gives an indirect, noisy view of autonomic balance.
Of the many ways to quantify that variation, the one worth using in a field setting is RMSSD — the root mean square of successive differences between adjacent RR intervals. RMSSD is dominated by high-frequency, vagally mediated variation, it is comparatively robust in short recordings, and it is less sensitive to breathing rate than frequency-domain measures like HF power. For a sixty- to ninety-second field measurement it is the defensible choice. Frequency-domain metrics need longer, more controlled recordings than a squad will ever reliably produce.
Two processing points matter more than most people realise. First, RMSSD needs beat-level RR intervals, not an averaged heart rate; if your device only exposes BPM, you cannot compute it, whatever the app claims. Second, raw RMSSD is strongly right-skewed, so a natural-log transform should be applied before any averaging or comparison against a baseline. Comparing raw RMSSD against a raw mean produces day-to-day swings that look like physiology and are arithmetic.
What it responds to
In a well-controlled measurement, log-RMSSD reliably falls with acute fatigue, sleep restriction, alcohol, illness onset, heat stress, dehydration and psychological stress. It rises with rest and with aerobic adaptation over months. That list is the source of both its usefulness and its ambiguity: a fall tells you the autonomic system is under load, but it does not tell you which of those seven causes is responsible.
The single most common misreading
A drop in HRV is treated as evidence of training fatigue, and the session is reduced. The athlete had four hours of sleep because of a new baby, drank two glasses of wine at a family dinner, and is entirely capable of training normally. The metric was correct; the causal attribution was invented.
Protocol beats precision
Almost every anomaly we are asked to explain turns out to be a protocol violation rather than a physiological event. HRV is sensitive to posture (supine and standing values differ substantially), to time since waking, to time since eating, to caffeine, to having just walked up two flights of stairs to the medical room, and to breathing rate during the measurement.
A club that measures at an inconsistent time, in an inconsistent posture, with athletes arriving at a run, is generating noise and interpreting it as signal. A club that measures every athlete standing, within twenty minutes of waking, before caffeine, with a consistent breathing instruction, gets a usable series from cheaper hardware. Protocol discipline is worth more than sensor quality here, and it costs nothing.
Reading the series, not the day
A single day's value should almost never drive a decision. What carries information is the relationship between a rolling seven-day mean of log-RMSSD and the athlete's own longer-term baseline, plus the variability within that window.
- Stable mean, low variability. The athlete is coping with the current load. This is the state you are trying to preserve.
- Falling mean, rising variability. The most interesting pattern, and the one most worth a conversation. Something is accumulating.
- Falling mean, low variability. Often seen in a heavy but planned block. Not automatically a problem; it is the expected response to intensification, and the question is whether it recovers on the planned taper.
- Rising mean. Usually a taper working, occasionally a sign of parasympathetic overreaching in extreme cases. Rare enough in team sport that it is worth checking rather than assuming.
Note that all four descriptions are about an individual against their own history. Between-athlete comparison of RMSSD is close to meaningless: baseline values differ by a factor of three or more across a normal squad for reasons that have nothing to do with fitness or readiness.
What we build, and what we refuse to
The module presents log-RMSSD as a seven-day rolling mean with the individual's baseline band, the within-window coefficient of variation, and the measurement conditions recorded per reading so a protocol violation is visible rather than buried. It exports raw RR series for departments running their own analysis.
What it does not do is collapse HRV into a single readiness percentage. We have been asked for that repeatedly and have declined, for a straightforward reason: the moment a number becomes a score out of a hundred, it starts being used as a decision rather than an input, and the uncertainty that should have travelled with it does not survive the conversion.
Where this leaves you
HRV is worth collecting if — and only if — you can run a consistent protocol, you interpret it against the individual rather than the squad, you read the series rather than the day, and you treat a change as a prompt to ask a question rather than an answer to one. Under those conditions it is one of the more useful things a performance department can measure daily. Outside them, it is an expensive way to generate arguments.
