Form, head-to-head and the home edge

Before any probabilities appear, the model boils each team down to a few honest numbers drawn from recent matches. These are the raw materials everything else is built from.

Last-10 form

For each side, the model looks at its last 10 matches (any venue) and computes its average goals for and average goals the opponent scored against it. That pair — what a team tends to score and what it tends to concede — is the heart of its current quality. It also records the recent W/D/L run and points-per-game, used later in the decision tree.

Ten games is a deliberate balance: long enough to smooth out one fluky scoreline, short enough to reflect the team as it is now rather than a year ago.

Head-to-head

Separately, the model pulls the last 10 head-to-head meetings between these two specific teams, with each side's average goals in those games. Some matchups simply play out differently from each team's general form — a rivalry that's always tight, a side that always struggles at a particular ground. H2H captures that, and it's only trusted when there are enough meetings (about three or more) to mean something.

Home advantage

Finally, playing at home is worth a real, repeatable edge in football. The model applies a fixed home-advantage factor that nudges the home side's expected goals up and the away side's down. It's modest and constant — not a guess per match, just the structural fact that home teams score a little more.

Three plain inputs — form, head-to-head, home edge — do most of the work. The cleverness is in how they're combined, not in any single secret number.