How Match Forecast works

The model. Match Forecast uses a Dixon-Coles statistical model trained on completed league results from the current season. It estimates each side’s attacking and defensive rates from how often they score and concede, adjusts for whether they’re playing at home, and converts that into a probability for every plausible scoreline. The output is what you see on each match: a home / draw / away split, plus over-2.5 goals and both-teams-to-score lines. There is no betting market input and no tipping language anywhere in the product — only football probabilities.

What it ignores. The model deliberately does not weight player availability, manager changes, fixture congestion, derbies, or stakes. Those are surfaced separately as context chips and folded into the editorial Caveat, because the model can’t reason about them with statistical confidence. It is also intentionally conservative on cup matches and two-legged ties, where league-trained statistics over-predict outcomes. We track every forecast against the eventual result, which is why the accuracy line you see on each forecast updates as the season runs.