The spike gets blamed for what the quiet months did

Attributing an injury to the last visible jump in load credits the wrong event and lets long undertraining escape notice.

The hamstring goes on Saturday, and by Monday somebody has found the cause: a hard session on Wednesday, clearly visible as a spike on the chart. Case closed, and the conclusion is almost always the same, which is that the spike should not have happened.

Maybe. There is a competing explanation with the same evidence, and it never gets equal billing. The Wednesday session might have been a perfectly ordinary session for a player who had been prepared for it, and the reason it was not ordinary for him is four quiet months in which he did very little fast running. Under that reading the cause is the trough, not the peak, and the treatment is more high-speed exposure earlier rather than less of it later.

The two stories fit the same chart. What separates them is what happened before the window you are looking at, and the standard four-week rolling view is exactly the wrong length to see it. Anything that took a season to develop is outside the frame, so it is systematically invisible to the analysis and the visible thing takes the blame by default.

There is a bias in what gets recorded, too. Sessions are logged; absences are not. A player who did nothing much for six weeks generates no rows in the dataset. The spike generates a row with a big number in it, so the analysis sees an event where there was an event and sees nothing where there was six weeks of nothing. Missing rows do not argue back.

Assigning blame to the most recent visible change is a habit, not an inference, and it has a cost beyond the individual case. Every time a spike takes the blame, the institutional response is to cap peaks. Cap enough peaks and you produce a squad that is never exposed to the speeds it will meet on Saturday, which raises the risk you were trying to reduce and generates more injuries to blame on more spikes.

I am not claiming spikes are harmless. Some are.

What I am claiming is that a monitoring system which can only see the last four weeks will always find one, and will therefore never in its life return the answer that a player was undercooked since August.