We Don't Guess.
We Calculate.

A goals model trained on twenty years of World Cup football. Every pick priced, settled, and graded in public.

3,593
matches trained
104
WC fixtures covered
< 1 min
to settle after full time
How it works

One pipeline, four moves — all of it graded.

No tipsters. The same loop runs on every World Cup fixture, start to finish.

01 · Data

Real results

Twenty years of World Cup + qualifiers — goals and actual xG.

3,593 matches

02 · Model

Every team rated

Dixon-Coles expected goals, blended toward real xG.

O2.5 · BTTS · 1X2

03 · Pick

Back the lean

The model's most-likely side + market blend. No longshots.

value → promo page

04 · Graded

In public

Settled against the real score. Nothing hand-typed or deleted.

< 1 min after FT

Nightly loop — every finished match feeds back into the ratings · 03:00 GMT
See it on a real game

One match, end to end.

Korea Republic v Czech RepublicGroup A · 12 Jun

What we knew

Both sides' attack and defence, rated from 3,593 matches and normalised across confederations.

KOR atk 0.9CZE atk 1.1

What the model said

Dixon-Coles turns the ratings into expected goals, then blends with the de-vigged market.

xG 0.8 / 1.1O2.5 37%1X2 32/31/38

The lean

The most-likely side — never a longshot, never edge-chasing.

UNDER 2.5 GOALS · 63%

What happened · graded in public

FT 2 – 1Lost ✗

Three goals — over the line, so the Under call missed. We log the misses too: nothing on the record ever gets deleted.

The part nobody else shows

Every pick, on the record.

Wins and losses both. The live track record updates the moment a match ends — see for yourself.

< 1 min
Settled after full time
0
Picks hand-typed or deleted
Nightly
Model refit on new results