About & methodology

Probability-first, not odds-first

Quanta publishes the model’s estimated probability of an outcome — nothing more. We do not carry a live odds feed, so we show no prices, no expected value, no edge and no stake sizing. The headline on every market is the model’s own likelihood of the event happening. If a feature the model needs is missing for a fixture, the affected market is suppressed rather than filled with a league-average guess.

The football model

Goals are modelled with an independent (double) Poisson: separate expected-goal rates for the home and away side, derived from recency-weighted xG, attack/defence strength, and a per-league home-advantage adjustment. The full scoreline distribution is built from the two Poisson margins, with a low-score correction applied to the correct-score grid to counter the independence assumption’s known bias on 0-0 / 1-0 / 1-1 results.

Match result, over/under, both-teams-to-score, Asian lines, clean sheet, winning margin and correct score are all read directly off that scoreline matrix, so they stay mutually consistent. Corners, cards and player props use their own dedicated models and are shown for information only — they are not graded in the results tracker.

Tracked for calibration

With no odds there are no returns to report, so we hold ourselves to calibration instead: when the model says 70%, do those picks win about 70% of the time? Every settleable pick is graded from the final score and logged to an append-only tracker. The Accuracy page shows, for each probability band, how often the predictions in it actually came true.

What this is not

Quanta is an analytics tool, not a tipster service and not financial or betting advice. Model probabilities are estimates and can be wrong. Please read our responsible gambling and terms pages.