Results
Most recent 60 matches, newest first
What xG diff tells you
50 of 59 most-likely calls landed — 85%. Always backing the home side over the same matches would have called 50%, so the model is +35 points ahead of that. 1 match had no forecast on record and is not counted.
Picking the most likely of three outcomes is a coarse test: it ignores how confident the model was, and a model can be right more often while forecasting worse. The measured versions are on Model, Calibration and Receipts.
The model pick is the out-of-sample forecast: made by a model fitted only on matches played before that one, so it is what the model actually said in advance. A dash means no forecast was on record before kick-off.