Shots tell the truth sooner

Goals are noisy. A side can play well and lose 1–0 to a deflection, and the goals-only model learns the wrong lesson. Shots on target happen ten times as often as goals, so a side's rate of creating them is measured far more precisely.

The platform fits a second Dixon–Coles model to shots on target, and blends its rates with the goals model:

The blend weight is another backtest choice. The shots model alone is too optimistic about sides that shoot from anywhere; the goals model alone is too slow. Together they closed 8.6% of the gap to the market in the Premier League.

Finishing

The difference between a side's goals and what its shots deserve is its finishing figure on the Strengths tab. It is real, but it fades: a hot finishing run is the most reliable thing in football to bet against. The blend does that automatically by leaning on the shots.