Why you cannot rank a roster by recovery score

You open the app on a Tuesday morning and there are twenty-four names, each with a number beside it. The instinct is to sort them, so whoever sits lowest gets the easier session.

That sort is broken twice over, and the first break happens before biology is involved at all.

Device agreement

Dial and colleagues put five consumer wearables against a chest-strap ECG reference across 536 nights with 13 adults, and reported concordance for nightly RMSSD:

Their conclusion was that Oura showed the highest agreement, WHOOP was acceptable, and Garmin and Polar were lower. Read that as a coach and it says something plain: a reading off a Polar and a reading off an Oura are two different measurements of the same thing, each wrong by a different amount.

Most rosters are mixed. Nobody buys their clients matching hardware, so on any given morning you are looking at a column of numbers produced by four or five instruments that were never calibrated against each other.

Person to person

Suppose your whole roster wore the same ring. The ranking is still empty, and this is the part that catches people out.

From Esco and colleagues, reviewing HRV monitoring on mobile devices in 2025: “because inter-individual differences in HRV can be influenced by factors unrelated to fitness or health, a lower HRV compared to a peer does not necessarily indicate poorer physiological status.”

Resting HRV differs between healthy people for reasons that have nothing to do with how recovered any of them are. The client sitting at the bottom of your column may be entirely fine. The one at the top may be three weeks into digging a hole.

Person against themselves

The same review is direct about the alternative: “the focus in athletic monitoring should be on intra-individual changes, with frequent HRV assessments taken over time within an individual rather than single measurement comparisons.”

The method this points at has been standard in endurance monitoring since Plews and colleagues set it out in Sports Medicine in 2013. Work from a mean of the week's daily readings rather than any single morning, and watch the coefficient of variation next to it, which is the day-to-day spread relative to that mean.

Two shapes carry most of the signal. A stable or rising mean with a low coefficient of variation is someone absorbing the work. A mean sliding down while the spread widens is the overreaching pattern, and it is the one worth acting on.

A single reading of 70 tells you nothing on its own. A 70 from someone whose weekly mean has held near 90 since June, with the spread opening up over the last four days, tells you what to do with their afternoon.

How the reading is taken

Esco and colleagues are specific about how readings have to be taken to be worth comparing: “record HRV as close to awakening as possible” and “before consuming food or stimulants”, and “maintaining the same body position for every recording is crucial, in that HRV values differ significantly across supine, seated, and standing positions.”

A client who takes a reading sitting up on Monday and lying down on Thursday has produced two numbers that cannot be compared to each other, let alone to anyone else's. Devices that sample through the night avoid most of this, which is a real argument for the passive ones over a morning app reading.

What to do instead

Stop sorting the column. For each person the question is whether today sits inside or outside their own recent range, and what else moved at the same time. A client whose mean has slid across four days, with two short nights behind it and training load flat throughout, is a different situation from a client who is simply low this morning.

The uncomfortable part is that the device you would choose is rarely the device your client already owns. Agreement with a chest strap is something you can read off a table. The ring or the watch already on their wrist was bought for other reasons entirely, and the ceiling on what their data can tell you was set before they ever walked in.