Training load and readiness, and how much to believe

Every version of this — Garmin’s training readiness, Polar’s recovery, Coros’s base fitness — is assembled from a short list: heart rate during sessions, heart rate variability overnight, sleep duration, and the accumulated load of recent weeks.

• Training load: intensity multiplied by duration, decayed over time. The best-founded of the four.
• Heart rate variability: the spacing between beats at rest, which falls when you are stressed — by training, illness, alcohol or a bad night.
• Sleep: measured by movement and heart rate, so good at duration and rough at stages.
• Recent history: the model needs weeks of data before it means anything at all.

Load is arithmetic on things that were measured, and the principle behind it — that injury risk rises when this week is much harder than the last few — is one of the better-supported ideas in the field. Used as a brake on enthusiasm, it earns its place.

The catch is that it inherits whatever your heart rate sensor reported. A session of intervals recorded through a lagging wrist sensor lands in the week as easier than it was.

A single readiness number compresses sleep, HRV, load and recent stress into one figure between 1 and 100. It is useful as a prompt to ask yourself a question. It is not useful as an instruction.

The practical rule: when the score and your legs agree, act on it. When they disagree, your legs are the better sensor — and a score that is consistently wrong for you usually means it has not seen enough of your baseline yet.

1. 1

   Give it three weeks before reading anything

   Every one of these models is relative to your own baseline. Week one tells you nothing.

2. 2

   Watch the trend, not the day

   A single low morning is noise. Five in a row is a signal, and usually a correct one.

3. 3

   Wear it to bed, or ignore the recovery half

   HRV and sleep come from overnight data. Without them, readiness is load and guesswork.

4. 4

   Override it when you know better

   It cannot see a stressful week at work. You can.

There are four situations where these scores mislead reliably, and recognising them saves a lot of second-guessing.

• The first three weeks, before the model has your baseline. Everything it says in that window is about an average person.
• After a race, where the load from one enormous effort dominates the model for days and tells you to rest longer than you need.
• During heat, where an elevated heart rate at the same pace reads as a harder session than it was.
• When you are ill, where a poor HRV correctly signals something wrong but the watch attributes it to training.

Notice that three of those four are the model doing exactly what it was designed to do with the data it has. The failure is not in the arithmetic — it is that the watch cannot see the cause. It is measuring stress, not training.

Which is also the argument for setting it up properly at the start: a model fed good data from the first week is wrong far less often than one retrofitted onto three months of patchy history.