How to Check a Football Prediction Site's Track Record: A 6-Point Checklist

A track record is only worth something if it was fixed before kick-off and cannot be edited afterwards, if it counts every call (not the best ones), and if it is measured against something hard to beat, such as the closing price of a sharp bookmaker. Hit rate on its own tells you very little.

Use the six checks below on any football prediction website. If a site fails the first two, its record cannot be verified, however good it looks.

1. Published before kick-off, with proof. A screenshot or a list made after the match proves nothing. Look for a timestamp an outside party controls (an independent tip-tracking service, a public post that cannot be edited) or a cryptographic hash published before the match.

2. Every call counted. The record should include every prediction of the same type, not a selection. 'Our best wins this month' is marketing, not a record.

3. The price at the time. A call on a 1.20 favourite and one on a 4.00 outsider are very different. A record without the odds or probability at the time cannot tell skill from simply picking favourites.

4. Calibration on every match. For probability-based predictions: do calls rated 60% come true about 60% of the time, across thousands of matches? This is a stronger test than hit rate because it covers every outcome, not the chosen ones.

5. A benchmark. The sharpest betting markets are very well calibrated. An honest site shows how its model compares with the market's margin-free price, and says so when the market is better.

6. No guarantee language. 'Sure win', 'fixed matches' or 'guaranteed profit' are warning signs on their own. Football is uncertain, and any honest record has losing runs.

Before kick-off, every PunditRival call is turned into a SHA-256 hash, chained to the previous one, and the hash is published. After kick-off the secret salt is published, and anyone can recompute the hash in the browser on the Verify page. If it matches, the call is exactly what was committed before the match.

The model itself was tested walk-forward on more than ten thousand matches in nine European leagues: each prediction only used results from before that match. Calibration by probability band and the comparison with Pinnacle's closing prices are published on the Method page and in our accuracy guide. The market was slightly more accurate than our model, and every match page shows both side by side.

Graded results are shown on the Results page for 24 hours after each match. PunditRival takes no bets and sells no tips; the Premier League prediction game is free to enter (18+).

Some services record tipsters' picks with the odds at the time and keep the full history, so you do not have to trust the tipster. They are useful for judging tipsters who post individual bets. For probability models that price every match, calibration and log loss over all matches are the better yardsticks.