What training load actually measures

The chain, one link at a time

It helps to see how many steps sit between your heartbeat and the number in the app.

1. A sensor detects your heart rate. If it is a wrist optical sensor, the published evidence says this step is reliable during steady work and much less reliable during rapid intensity change.
2. The watch compares that against thresholds — maximum heart rate, lactate threshold, or zones — which are themselves usually estimated rather than tested.
3. Time in each zone is weighted, typically exponentially so that hard minutes count for far more than easy ones.
4. The weighted total becomes a session score, which accumulates into rolling short-term and long-term averages.
5. The ratio between those two averages produces the "productive", "maintaining" or "overreaching" label you actually read.

Every link multiplies the uncertainty of the one before it. A load score is a fifth-order derivative of a measurement that, on a wrist, has documented failure modes at exactly the intensities that dominate the calculation.

What the acute-to-chronic ratio is trying to say

The idea underneath most load models is straightforward and sensible: compare what you have done recently against what you are used to doing. Train far above your habitual load and you accumulate fatigue faster than you adapt; train far below it and you lose fitness.

The labels brands attach to that ratio are interpretations of a real concept. Where they differ — and they differ a lot — is in the weighting, the time windows and the thresholds, none of which is published in enough detail to reproduce.

Why the numbers are not comparable across brands

Garmin, COROS, Polar, Suunto and Apple all compute something they call training load or training status, from proprietary models, on different scales. A COROS number and a Garmin number are not two measurements of the same quantity — they are two different opinions expressed in different units.

This matters practically when you switch watches. Your load history does not transfer in any meaningful sense, and a lower number on the new device is not evidence that you trained less.

The evidence position, stated plainly

We could find no independent validation of any consumer training-load model. Not peer-reviewed, not a measured enthusiast test. Nothing.

The closest relevant finding is from the 2025 nocturnal HRV validation, whose authors noted there is "little transparency into what metrics affect each device's own Readiness or Recovery Score" and that these algorithms are periodically updated without notice. That was written about recovery scores, and it applies with equal force here.

So the honest evidence tier for every training-load score on the market is no validation study exists. That is not the same as saying the numbers are useless — it is saying nobody has checked, and any page that tells you one brand's load model is more accurate than another's is making it up.

How to use it anyway

As a relative signal on one device. Consistent arithmetic applied to consistent inputs will still show you when this week was much harder than last. That is genuinely useful and it does not require the model to be validated.

Fix the input before you trust the output. If your heart-rate data comes from a wrist sensor during interval sessions, the load score is built on the measurement with the largest documented error in this category. A chest strap improves everything downstream of it, which is the single highest-leverage change available — the upgrade path is here.

Get your thresholds right. A load model calibrated against an estimated maximum heart rate that is 15 beats wrong will misweight every session you do. Most watches let you set these manually; a field test is more useful than a formula.

Do not let it override how you feel. A number computed from an unvalidated model on top of an inferred measurement should not outrank the fact that you are exhausted.