Expected goals is fine and the arguing about it is not

The metric does exactly what it was built to do, and almost every public fight about it comes from using it over ninety minutes.

Somebody loses two-nil, posts the shot map, and points out that they generated more expected goals than the winners. Somebody else replies that the goals are on the scoreboard. Both of them are having the wrong argument, because the number was never designed to settle a single match.

What the model does is assign each attempt a probability of becoming a goal, based on how similar attempts have finished historically: distance, angle, body part, whether the ball was crossed, how the move began. That is all. It is a way of describing chance quality without asking a human to eyeball it.

The reason it goes wrong over one match is sample size, and the maths is unforgiving. A team taking twelve shots worth an average of one in ten each will finish with a wide spread of outcomes, and the difference between zero goals and three is comfortably within normal variation. Quoting a ninety-minute total to two decimal places implies a precision that a dozen coin flips cannot support.

Over a season it stabilises and becomes the most useful summary of attacking process available, which is exactly why recruitment departments use it and why the argument in public bears no relation to the argument inside the sport. Nobody in a club is claiming a team deserved to win on Saturday. They are looking at thirty matches and asking whether a striker's finishing has been unusually cold, because a cold striker is often cheaper than a hot one and just as good.

There are honest limits and they are worth stating. The models mostly do not know where the defenders were. They do not know if the goalkeeper was out of position. They treat a shot taken under no pressure from twelve yards the same as one taken with a boot arriving, and better versions have started to fix this but the public ones lag.

The metric is not a verdict on a match. It is a description of process across many of them.

Use it on one game and you deserve every argument you get.