Most football sites show you a scoreline. Fewer will tell you where it came from, how it was checked, or what happens when it turns out to be wrong.
This page explains all three: where our data comes from, how our model produces the numbers you see on match and team pages, and what we do when we find an error in the historical record. If you only read one section, read Corrections — it's the part most sites don't have.
Gooolll licenses its match data from two professional providers:
We pay for both. We do not scrape other football sites, and we do not republish another publisher's editorial content as our own.
Providers disagree with each other more often than you would expect — on kick-off times, on how a knockout tie is recorded, on which of two duplicate fixture entries is canonical. Where they disagree, we resolve to one canonical record rather than showing you both. Our current database holds 108,537 completed matches with full scorelines across 46 competitions, plus a separate national-team history reaching back to 1930.
Everything labelled Gooolll Model Read on a match, team or competition page is produced by
our own model — currently version elo-poisson-v1. It is not licensed from a provider, and it
is not a repackaged betting-market price. It runs on two independent signals, each doing the job
it is suited to.
Every team carries a single rating that moves after each match, up when they outperform expectation and down when they underperform. A team's first rating is 1500. The rating gap between two sides is converted into a win expectation using the standard logistic Elo curve.
How much a result moves a rating depends on what was at stake. A World Cup knockout tie carries a K-factor of 60; a World Cup group match 50; a continental championship knockout 50; a Champions-League-class club knockout 42; a top-five domestic league 32; an international friendly just 18. A pre-season friendly should not reshape a team's rating the way a World Cup quarter-final does, and in our model it doesn't.
Draws are handled with an empirical curve rather than a fixed number: two evenly matched sides draw roughly 28% of the time, and that probability tapers as the rating gap widens, with a floor around 15%.
Separately from Elo, each team carries an attack rating and a defense rating: an exponentially-weighted moving average of goals scored and conceded relative to the league average, with roughly a 16-match memory. These give an expected-goals figure for each side, and the most likely scoreline follows from treating the two as independent Poisson variables.
A home team gets 65 Elo points and a 10% scoring boost — but only when the home advantage is real. National-team matches at World Cups, continental championships and neutral-venue friendlies get zero home advantage, because nobody is at home. The exception is a host nation playing on its own soil, which does receive it. Pages carrying a neutral-venue fixture say so explicitly.
These apply small, capped adjustments — at most 35 rating points for recent form and 30 for head-to-head history, and never more than 60 combined. They adjust a single prediction and are never written back into a team's stored rating.
The model was fitted by walking every completed match in our database in chronological order, oldest first. For each match, the prediction was made using only the ratings that existed before that match was played, and form and head-to-head inputs drawn strictly from earlier fixtures. The result was then scored, and only afterwards were the ratings updated.
That ordering matters: it means the model was never tested on a match it had already learned from. Predictions are scored on two measures — straight accuracy against a three-way random baseline of 33.3%, and Brier score, which rewards being confident when right and punishes being confident when wrong.
4,449 teams currently carry a rating.
We would rather tell you this than have you find out:
Licensed data arrives with errors in it. The most common one we deal with is knockout ties: some sources fold a penalty-shootout result into the 90-minute score, so a match that finished 1–1 and went to penalties is recorded as, say, 4–3. Left alone, that corrupts the historical record, the head-to-head numbers shown on match pages, and every rating the model derives from them.
Our approach:
Our national-team history is the deepest part of the archive and has had the most correction work done on it. We currently hold 497 penalty shootouts recorded separately from the 90-minute score — each one a knockout tie whose result is stored the way it was actually decided rather than folded into the scoreline.
Found something wrong? We would genuinely like to know. Report it through our contact page with the match and what you believe the correct result is, and we will check it against independent sources and correct it if you're right. Corrections are made to the underlying record, not patched on the page.
Articles published under Gooolll Desk are produced with AI assistance and reviewed before publication. Our standing rules:
Model version elo-poisson-v1. Data licensed from football-data.org and API-Football.
Figures on this page last verified 26 July 2026.
More about the site on our About page.
← Back to Gooolll