Back to Aria →
Aria is trying to measure what happens next in a football match. Goals decide bets, but they do not always tell you whether the read itself was good. These two blind-archive games show the difference.
Aria was 0.002 xG away from what the second half actually produced. As a read of the attacking output, that's about as close as we could ask for.
And there wasn't a single goal.
If we judged the model from the scoreboard alone, we'd throw away an excellent read. The chances arrived. They just didn't become goals.
This time the scoreboard flatters Aria. Four second-half goals looks fantastic until you look underneath it.
The actual chance quality was 1.29 xG, more than two xG below Aria's projection. That's a poor read, regardless of how many of those chances happened to go in.
A winning outcome doesn't get to turn a bad measurement into a good one.
Over one match, finishing can make a good read look terrible and a bad read look brilliant. That's why Aria's blind results are judged primarily against the xG subsequently generated, with goals recorded alongside it rather than used to rewrite the story afterwards.