Prediction review
Saturday, 19 September 2026
On 2026-09-19 we had published a prediction for 99 fixtures before kick-off. We called 42 of them correctly, 42%.
Our strongest call, and it missed
The most confident thing we published was Aris to beat Iraklis, at 85.2%. It finished 0–0. That is the day's boldest call and its largest error in one fixture. A model that is never surprised is not saying much; the question is whether it is surprised this often at this level, and that only answers itself over months rather than over one evening.
When we were confident
12 of the day's calls were at 60% or above. 7 of them landed. Over a single day that number means very little — it is the same figure across a month that says whether our stated confidence is honest.
The ones we called close
21 fixtures were split closely enough that no outcome reached 40%. 7 of them ended level. We publish these without pretending to a view: a fixture the model divides three ways is one it has little to tell you about, and saying so is more useful than manufacturing a lean.
Coverage
The day's predictions spanned 21 competitions: League Two (12), League One (11), Championship (9).
Only predictions published before the match day are counted. Results come from public fixture feeds. Every day we have called → How the model is measured →