Past Results — How Well Did the Predictions Hold Up?

Past model performance does not guarantee future accuracy. All results are for information only; not betting advice.

Matches evaluated

183

/ 183 total

Date range

240DAYS

Completed matches of the last 240 days

Matches skipped

Not enough form data

01 /

PROBABILITY ACCURACY

The lower the error score, the better. Reference columns: for accuracy, “always saying the result that happens most often”; for the error score, “writing the league's past result distribution for every match”. If we are ahead of the reference, judging every match separately is working.

PROBABILITY	OUR ACCURACY	REFERENCE ACCURACY	ERROR SCORE (US)	ERROR SCORE (REFERENCE)	GAP
Match Result	46.4%	42.1% (Home)	0.6190	0.6552	+0.0362
3+ Total Goals?	56.8%	55.2% (Over)	0.4905	0.4946	+0.0041
Both Teams to Score?	55.2%	53.6% (Yes)	0.4962	0.4975	+0.0013

02 /

ERROR SCORE: US AND THE REFERENCE

Match Result

MODEL

0.6190

REFERENCE

0.6552

3+ Total Goals?

MODEL

0.4905

REFERENCE

0.4946

Both Teams to Score?

MODEL

0.4962

REFERENCE

0.4975

03 /

WHEN WE SAY “70%”, IS IT REALLY 70%?

The horizontal axis is the percentage we gave, the vertical axis how often the result really happened in those matches. Ideally the points sit on the green diagonal: points below it show we gave too high a percentage (overconfidence), points above it that we were cautious.

Match Result

CALIBRATED

MODERATE

OFF

3+ Total Goals?

CALIBRATED

MODERATE

OFF

Both Teams to Score?

CALIBRATED

MODERATE

OFF

HONEST ASSESSMENT

“This page exists to show how well our predictions really held up. We show not only the good results, the places where we fall short as well.”

On some probabilities we may not beat the simple reference; in that case the “always the most popular choice” strategy gives a better result. We show this too so that users can make informed decisions.