Expected goals is a model of chance quality, not an alternative scoreline

Expected goals assigns a probability to a shot based on how similar attempts have performed in historical data. A chance valued at 0.20 xG is not a prediction that one fifth of a goal will appear on the scoreboard. It means that, under that model, comparable shots have been scored about twenty percent of the time. Adding shot values describes the quality and volume of chances a player or team accumulated.

The scoreboard remains governed by goals, not expected goals. The IFAB Laws define when a goal is scored and how the winner is determined. xG asks a separate analytical question: how difficult were the recorded shooting opportunities? A team can lose while producing higher xG because football matches contain few goals and finishing outcomes vary. Calling that the 'wrong' score confuses a model with the competition rule.

Models can legitimately disagree. Stats Perform describes inputs including shot angle, distance, assist type and whether an attempt was a header or a big chance; its explainer discusses additional context and historical training shots. Another provider can use different variables, definitions and samples. An xG value therefore needs a named provider and should not be combined casually with values from another model.

Goals above xG can indicate strong realised finishing over the observed shots, but interpretation depends on sample size. A short hot streak can create a gap that later narrows. Sustained over-performance is more interesting, yet it does not isolate technique from shot selection, defensive pressure, goalkeeper performance or model blind spots. The responsible language is evidence of over-performance in the model, not proof of an immutable trait.

Post-shot models ask a different question. Stats Perform describes expected goals on target as incorporating where an on-target attempt reaches the goalmouth, building on the original pre-shot chance quality. This helps separate chance creation from shot execution and goalkeeper outcomes. It also shows why labels matter: xG and xGOT are related but not interchangeable.

For comparisons, xG is most useful beside shots, non-penalty goals, minutes and competition filters. A forward receiving many close-range chances may have high xG because of excellent movement and team creation. A wide attacker may create value through carries and passes not represented in shooting xG. No single number can settle those different jobs.

FootIndex will publish model metrics only when provider and scope are explicit. Until consistent data is available, we prefer verified goals, minutes and appearances with clear limitations. Refusing to display an unsupported xG estimate is better than creating one from goals or shot counts and giving it a scientific-looking label.