Test forty players on ten markers and somebody flags every week

Hundreds of weekly comparisons guarantee alerts by arithmetic alone, long before any athlete has anything wrong with them.

Forty athletes. Ten markers each. That is four hundred comparisons made every Monday morning against whatever thresholds the system holds.

Set those thresholds so that a healthy athlete trips one about five times in a hundred, which is a normal and defensible choice, and the expected number of alerts on an entirely healthy squad is around twenty. Every week. Before anybody has strained anything. The dashboard lights up because you asked it four hundred questions, not because the answers mean anything, and this is arithmetic rather than a fault in the sensors.

What happens next is the interesting part, and it is a human process rather than a statistical one. Staff cannot chase twenty flags, so they start filtering informally. They look at the flags on players they were already worried about and dismiss the rest as noise. That filtering is often good clinical judgement, and it also means the monitoring system has stopped contributing anything, since the decisions are being made by the prior suspicion rather than by the data. A flag on a player nobody was watching gets ignored, and that flag was the only one capable of telling you something you did not already know.

The volume also degrades everyone's response over a season. Twenty alerts in week one get attention. By November the same twenty are wallpaper. Alarm fatigue is a well-described failure in hospitals, where it has killed people, and there is no reason to think sports staff are more resistant to it than nurses.

There are honest fixes. Reduce the number of markers, which is unpopular because each one had a champion when it was purchased. Require two markers to move together before anything is raised, which cuts false alerts sharply at some cost in sensitivity. Or set the threshold by how many alerts you can actually act on in a week, and let that operational capacity define the cut-off rather than a statistical convention imported from somewhere else.

That last option offends people because it sounds like choosing the answer first.

It is not. It is admitting that a flag nobody has time to investigate is not a measurement at all, it is a light on a panel, and a panel of lights that nobody looks at has stopped being a monitoring system some time ago.