Frequently asked questions

A statistical model sets every probability: Elo ratings and a Dixon-Coles goals model, fitted only on results available before the forecast locks. It turns team strength and home advantage into a grid of possible scores, and the win, draw and loss chances are read from that grid. AI analysts add sourced team news and written analysis. The full method is on our methodology page.

They are AI tools from Anthropic and OpenAI. Before each match they collect sourced team news, give their own forecast and draft the analysis. In version 1 their forecasts run in shadow: we grade them but they do not change the published probability. An editor checks written analysis before it goes live. See our models for how each one is doing.

Extra time and penalties happen only in some cup matches and follow different patterns. Grading every match on the same 90 minutes, stoppage time included, keeps the record comparable across leagues and seasons.

Each forecast normally locks 24 hours before kickoff; a match that enters our tracking later locks straight away. Until then the match shows “Forecast not yet locked”. After the lock the numbers cannot change: the forecast is fingerprinted with SHA-256 and listed in the ledger, so anyone can check it was not edited later.

A home win can come through 1–0, 2–0, 2–1 and many other scores, while draws sit on a few, mostly 1–1. So 1–1 can be the single most likely score even when a home win is the most likely result. Match pages show both, and our correct score page explains the difference.

Since 31 October 2026 we have graded 72 locked forecasts. The most likely result was right in 65% of them, with an average ranked probability score of 0.193. These numbers update after every batch; the track record adds confidence intervals, calibration and comparisons with simple benchmarks.

Two scores that grade the probabilities, not just the pick. RPS compares the forecast with the result in the order home, draw, away; log loss looks at the probability we gave to what actually happened and punishes confident misses. Lower is better for both. Our RPS and Brier score guide works through an example.

No. Every locked forecast is graded and stays on its match page, in the results archive and in the ledger, right or wrong. Postponed matches are voided and listed in the exclusion log with the reason. You can download the full ledger and recalculate our numbers yourself.

Football has three outcomes and draws are hard to call, so even strong models pick the right result only a little over half the time. Simply backing the home side is right well under half the time in most leagues. What matters is beating simple benchmarks over many matches, which is why we publish RPS and calibration next to the hit rate.

It is not a live-score service. Data arrives in reviewed batches through the day, and every page shows the time of the latest batch and the next one. Results are graded once two independent sources agree on the score. Our data sources page lists the latest update for each league.

The locked forecast is voided, not graded, and the match is listed in the exclusion log on the track record. When a new date is set, the match gets a fresh forecast that locks 24 hours before the new kickoff.

“Statistical forecast” means the probabilities come from the model alone: no reviewed written analysis was published for that match, or team news had not been checked for it yet (“team news pending”). The numbers are the same either way, because the model does not use team news. Before a forecast locks, normally 24 hours before kickoff, the match page shows form and team news but no probabilities. A match with no probabilities after that did not have enough reliable data for the model, for example a side with too few recent results. We show it as unavailable rather than guess.

Yes, everything is free: predictions, analysis, the track record and the downloadable ledger. There is no account, no paid tier and no paywall. The site is independent and takes no money from bookmakers; our About page explains how it is paid for.

No. Forecasts are probabilities for information and entertainment, not advice to bet. A 70% favourite still fails to win about three times in ten. If you do bet, only use money you can afford to lose. Our responsible gambling page lists free, confidential helplines. 18+.

Yes. The forecast ledger is published under CC BY 4.0: you can copy, analyse and republish it with a credit and a link. Download it as CSV or JSON from the ledger page. Club and competition names belong to their owners and are not covered by that licence.

Use the contact form. Add the match or page URL, what is wrong and, if you can, a source that shows the correct information. We fix data in the next batch, and every change to a graded result is logged in public on the methodology page.