The expected-goals blend

The model's first real output is an expected goals figure for each team in this specific match — its best estimate of how many goals each side should score. Everything downstream flows from these two numbers.

Attack meets defence

A team's goals don't depend only on how good its attack is — they depend on the defence it's facing. So the model builds each side's expected goals by averaging two things:

• the home team's average goals scored, and
• the away team's average goals conceded.

The same is done in reverse for the away side. A strong attack against a leaky defence pushes the number up; the same attack against a miserly defence pulls it down. This symmetric blend is what makes the estimate a matchup number, not just a team rating.

Folding in head-to-head

When there's a meaningful head-to-head history (roughly three meetings or more), the model blends it in — weighting the general-form estimate more heavily but letting the actual goals these teams have scored against each other shift the figure. Recent form leads; the specific rivalry adjusts.

Then home advantage

Last, the home-advantage factor is applied — lifting the home side's expected goals slightly and trimming the away side's. The result is two numbers: expected goals for the home team and for the away team. There's also a small floor so neither figure can collapse to zero — even a heavy underdog has some scoring chance.

Expected goals is the pivot of the whole model. Get these two numbers right and every market — result, totals, both-teams-to-score — follows from them.