Simulated league tables are entertainment dressed as forecasting

Running a season ten thousand times produces a beautifully precise answer to a question nobody should be asking with that much confidence.

By October somebody has published a table saying a club has a four point two per cent chance of being relegated and another has an eleven per cent chance of finishing in the top four. It gets shared widely. The decimal place is doing a lot of persuading.

The method behind it is sound enough in outline. Estimate each team's strength, simulate every remaining fixture using those estimates, do it many thousands of times, count how often each outcome appears. There is nothing wrong with the arithmetic and the code is not hard to write.

The trouble is everything the simulation assumes without saying so. It assumes the team strengths it starts with are correct. It assumes those strengths hold for eight months. It assumes fixtures are independent of each other, so that a January collapse does not lead to a sacking that changes February, and that a squad's quality is unaffected by a transfer window sitting right in the middle of the period being forecast.

None of those hold. Managers get sacked, and the change is not random with respect to results. Injuries arrive in clusters. Teams safe by April behave differently from teams fighting, and a side with nothing to play for is a genuinely different opponent, which every gambler knows and very few models encode.

Ten thousand runs quantifies the randomness inside the model's own assumptions. It says nothing whatever about whether those assumptions are right, and the second kind of uncertainty is much larger than the first. Presenting the output to a decimal place hides that completely, and the presentation is a choice made by the publisher, not a property of the mathematics.

I like these things. I read them. They are a fine way of showing that a six point gap in November is smaller than it feels, and they are a useful corrective to a press cycle that treats every weekend as final.

Just round them. Say roughly one in twenty rather than four point two per cent, because the honest precision of a football forecast in October is about that coarse.

The decimal is not accuracy. It is the interface pretending to be sure.