Our prediction methodology

Inputs we use

Every match estimate starts from public data: full-time results and goal counts for the last two seasons, shots and shots on target, home and away splits, rest days between fixtures, confirmed team news, and the current market price at several bookmakers. Where a league has thin or unreliable data — some domestic cups and lower divisions — we say so on the page rather than publishing a confident number we cannot support.

From data to probability

We model goals scored as a Poisson-style process. Each team gets an attacking and a defensive rating derived from recent scoring rates adjusted for opponent strength, then a home advantage adjustment for the specific league rather than a single global constant. Those ratings produce an expected goals figure for each side, which expands into a scoreline distribution. Market probabilities — 1X2, over/under, both teams to score, correct score — are read off that distribution.

Recent matches are weighted more heavily than old ones, and small samples are pulled toward the league average so a team with three matches played is not treated as if it has thirty.

Removing the bookmaker margin

Bookmaker odds do not sum to 100% — the excess is the margin. Before comparing our number with the market we strip that margin proportionally to get the implied fair probability. A pick is only labelled value when our probability exceeds the margin-free market probability by a meaningful gap, not by rounding noise. You can run the same calculation yourself with our no-vig calculator and implied probability tool.

Confidence ratings

Confidence reflects how stable the estimate is, not how likely we think a win is. A high confidence rating means the inputs are complete and the model output is not sensitive to one missing player or one unusual result. A low rating means the data is thin or contradictory. A high-confidence 55% is still a 45% chance of being wrong.

What this method cannot do

It cannot know about unreported injuries, dressing-room disputes, weather at kickoff or motivation in a dead rubber. It cannot predict individual matches reliably — it produces probability distributions that only make sense across many bets. And it cannot make betting profitable on its own: prices move, margins compound, and losing runs are normal.

We do not publish guaranteed outcomes, and any site that does is selling you something. Read responsible play and our editorial policy for the standards that sit behind this method.

Review and correction

Model outputs are reviewed by Chibueze Musa before publication, and pages carry the date of their last substantive update. When we find an error in an input or a published probability, we correct the page and note what changed.