Methodology

What ProofPitch shows

For each match, we show the model’s estimated chance of an outcome, a calculated price and a minimum-odds threshold. These are tools for understanding a match and its price, not an instruction to bet.

How a forecast is calculated

The model estimates how many goals each team is expected to score, using past results, opponent strength and quality-of-chance data (expected goals). From these expected goals it calculates probabilities for outcomes and lines: 1X2, totals, both teams to score and handicaps. It is a goal model of the Poisson type. It estimates probabilities and can be wrong in individual matches.

Where the data comes from

Results and quality-of-chance data come from public football data providers. Bookmaker prices come from The Odds API, using Pinnacle as the reference when it lists the match and the median of bookmakers otherwise. Stored prices retain their source and capture time.

Strong, Good and Careful

Every forecast gets a level from the model's own assessment and the reliability of the league. Strong and Good are the forecasts we count in the statistics. Careful forecasts are shown for reference and are not counted.

Calculated price and minimum odds

The price on a forecast card is calculated from the market price with a 5% margin added. It is not a bookmaker quote and may differ from the price available to you. The minimum odds mark the threshold below which, under the model’s assumptions, the selection stops making sense. A price above that threshold does not guarantee profit.

How forecasts are recorded

New forecasts are entered in a journal before the match with their time, chance and price. The journal uses a hash chain: each day receives a fingerprint, so altering a record afterwards would break the chain. After the match, wins, losses and returned stakes are added to the record.

Live results and retrospective checks

Live results are forecasts recorded before the match. Retrospective checks are the model applied to matches that had already been played; they show how it would have behaved, not what was promised in advance. We never mix them.

As of 3 November 2026: 116 forecasts of the Strong and Good levels are checked, and 42 of them were recorded before the match. The numbers are still small, so conclusions are premature.

We do not change the model's parameters until 300 live forecasts have been recorded, so that results cannot be tuned after the fact.

Calibration: do percentages match reality?

Calibration compares stated probabilities with observed outcomes. Among forecasts assigned about 70%, roughly how many were successful over a sufficiently large sample? The app’s statistics show this comparison, including where the model misses. Read sample sizes, dates and the type of record alongside percentages.

What the model does not know

The current model does not use confirmed line-ups or injury information. Missing information, data errors and unexpected events can change the result. A 70% probability is not certainty: across comparable cases, about 30% would not produce that outcome if the estimates were well calibrated. It does not mean exactly seven successes in the next ten matches.

How articles are made

Articles are prepared with AI assistance. The ProofPitch editorial process compares rules and figures with sources and recalculates educational examples. Sources are linked and their verification dates are given. Educational calculations must be distinguished from actual model results.

Partner links and money

The app contains a paid bookmaker partner link. It is marked, is for adults only and does not determine forecasts. ProofPitch earns from partners and the Premium plan; neither determines the probability assigned by the model.

Corrections

If you find an error, use the contact page. Corrections should identify the date of the change.

Important

ProofPitch provides analytics, not betting advice. Results are not guaranteed. 18+ or the higher legal age where you live. Betting involves a risk of loss.