Why the level moves

Game patches, roster changes, which operators are on shift and how the league sets up its rounds all shift the baseline. None of it is visible in a fixture list. It shows up only as the whole league quietly scoring more or less than it did last month — 0.85 goals between 2025-11 and 2026-03.

What it means for a prediction

A model fitted on a year of history and left alone would sit at the year’s average and be wrong in one direction for months at a stretch. This one weights recent matches far more heavily — a three-week half-life — and re-fits every night, so it tracks the level instead of averaging over it.

Does it matter more than who is playing?

They are comparable, and it is worth being precise rather than dramatic about it. The level moves 0.85 goals across the year and about 0.25 in a typical month. Swapping the two operators moves an individual prediction by a similar order. Neither dominates — which is exactly why the model carries both a per-operator rating and a fast-moving level, and why leaving either out is expensive.