Methodology: how we analyse every match

No prediction here is a loose opinion. Every analysis follows the same data pipeline and the same quality controls.

1. Data collection

For each match: odds from multiple bookmakers (1X2, over/under 2.5, BTTS), lineups, injuries, season stats (goals, xG, corners, cards), head-to-head, referee data and standings. We also monitor official club/federation sources and local press.

2. Implied probabilities (vig removed)

Odds embed the bookmaker's margin. We remove it ('de-vig') to get the market's true implied probability — that's what we compare our model against.

3. Statistical score model (Poisson)

A Poisson model over season goal averages projects expected goals, most likely scorelines and 1X2 / over 2.5 probabilities. Where the model diverges from the market we flag it — that's the basis of our Value Picks page.

4. AI analysis with internal critique

An AI engine (Claude models) turns the signals into a structured report. Before publishing, a second pass hunts for contradictions and ignored signals; if found, the report is revised.

5. Post-match grade and learning

After each match we grade our own analysis (1-10): what we got right, what we missed. Lessons feed the next analyses for the same league and teams.

Limitations (read this)

Probability is not certainty: a 70% favourite loses 3 times out of 10. Small samples reduce model accuracy. This is statistical analysis and interpretation — not betting advice.

See how the model has been performing, misses included:

Full track record →