As accurate as
the bookmakers.
Checked in public.

An expected-goals model prices win, draw, away, over/under 2.5 and both-teams-to-score for every fixture it covers. We graded it against 1,014 finished matches and published what came back — including where it falls short.

No betting tips and no claim to beat the market — we tested that, and it does not. Here is what we found.

Published now

All fixtures

Live from the model. Bars show win / draw / away; the pair of numbers is expected goals.

Graded against the
people who do this
for a living.

Every published forecast is stored before kickoff and scored after the final whistle. Against bookmaker prices on the same 1,014 matches, the model picked more winners and was twice as well calibrated — and their probabilities were still sharper overall. All three numbers are here because the third one matters too.

MeasurexGvantageBookmakers
Top pick correct49.7%49.2%
Calibration error0.0160.032
Log loss1.0191.002

Out-of-sample on the 2026 season across Argentina, MLS, Brazil, Norway and Sweden. Lower is better for calibration error and log loss.

When it says 60%, does it happen?

This is the only test that matters for a probability, and it is the one nobody shows you. Each band holds every published selection in that range, with what actually happened next to it.

10–20%
14.5%
20–30%
25.1%
30–40%
33.7%
40–50%
45.5%
50–60%
55.4%
60–70%
66.0%
what happened what the model saidEvery band with real depth lands within about a point.

29

Leagues priced

3

Markets per match

1,014

Matches graded so far

0

Tips sold

Probabilities you can check.

Every fixture we cover, every market, on web and mobile. One tier — no locked leagues, no upsell, and nothing sold as a tip.

Create your account