How to evaluate a football prediction site

Start with evidence, not the loudest win

A screenshot of a winning selection cannot show whether the forecast was published before kickoff, whether the quoted price was available or how many losing calls were omitted. Evaluation begins with a record that makes those questions answerable.

Look for a defined method and scope

The site should explain what it forecasts, which evidence is available before the match, how probabilities are produced and where the method does not apply. A useful explanation is specific enough to challenge without pretending that proprietary code must be published.

• Named markets and settlement rules.
• A clear time boundary for inputs and publication.
• Defined competitions, exclusions and unavailable-data handling.
• Model-version or methodology change notes.
• An honest account of uncertainty and limitations.

Require predictions to exist before the result

A credible audit trail preserves the original selection, probability, price and publication time. Results should be appended later under fixed settlement rules. Corrections should remain visible rather than silently replacing the original record.

Check whether the metrics fit probability forecasts

Hit rate changes with the difficulty and price of selections. Probability forecasts should also be tested for calibration and with proper scoring rules such as the Brier score. Profit or return figures need their period, total stakes, pricing basis and complete sample.

• Count and sample period.
• Win, loss, void and unresolved rules.
• Average forecast probability and calibration.
• A proper probability score where appropriate.
• The price source, timestamp and availability.
• Results split by materially different model versions.

A practical evaluation checklist

• Can you identify exactly what was predicted?
• Was the forecast recorded before kickoff?
• Are losses and voids as visible as wins?
• Can you inspect the complete period rather than a chosen streak?
• Are probabilities evaluated, not merely labelled confident?
• Are quoted odds and their timing explained?
• Are exclusions, unavailable data and corrections disclosed?
• Does the site avoid guarantees and pressure tactics?

Red flags that weaken a performance claim

• Guaranteed-win or certain-profit language.
• Only winning screenshots or an unexplained best-of list.
• A performance period chosen after results are known.
• Predictions that can be edited without a visible history.
• Profit claims without stakes, prices, availability or settlement rules.
• Testimonials used in place of verifiable evidence.
• Urgency designed to bypass careful review or spending limits.

Past evidence still has limits

A well-documented historical record can show how a method behaved under stated conditions. It cannot establish that future performance will match the past, that every user obtained the quoted price or that betting will be profitable.