Pacy
Data methodology
How Pacy collects, checks and updates scores, and how its match predictions are built, tested and graded.
This page explains how a score gets from the match to your screen, how Pacy makes sure it is the right score, and how our predictions are built, tested and graded. If something here stops being true, the page is wrong and we change it.
Seconds, not minutes
Live matches are re-read every few seconds. Upcoming and finished matches are refreshed on slower cycles.
One match, one record
A match reported by several sources is merged into a single record, so it appears on Pacy once.
Honest when late
If updates stop arriving, the page says the data is delayed. It does not pass off an old score as live.
Predictions on the record
Every prediction is kept and graded against the result, and the model is retuned against that record every night.
How a score reaches the page
- Collected. Pacy collects fixtures, scores and match events continuously from established sports data sources.
- Identified. Teams and competitions are matched through a maintained catalogue of their names and IDs. Two clubs with similar names are never merged on spelling alone, and a women's or youth fixture is never mixed up with the senior men's side.
- Reconciled. When sources disagree, Pacy shows the account it can corroborate rather than the one that arrived first. A source that keeps proving unreliable is removed.
- Published. The reconciled record goes on the scoreboard, the match page, and the team, competition and date pages that list it.
- Settled. Once a match is final it is settled and left alone. Postponed and cancelled matches are marked as such.
How often it updates
- A live match is re-read every few seconds.
- An upcoming fixture is re-read on a slower cycle, because nothing about it is changing yet.
- A finished match is settled and left alone, unless a correction is needed.
- Finished matches keep their page, so a result stays readable long after the day it was played.
Timings
Every time on Pacy is shown in one zone, stated on the page. A match is filed under the calendar day it kicks off in that zone, so a late-night fixture does not jump to the following day.
When Pacy is behind
If updates stop arriving, the page says so rather than continuing to show the last score as current. A stale score shown as live is the worst thing a scoreboard can do, so we would rather tell you we are waiting.
How predictions are made
Pacy's win probabilities come from a rating model. Every team has a rating within its competition. After each result the rating moves by how surprising that result was: beating a stronger side moves it more than beating a weaker one. The gap between two teams' ratings, plus home advantage, becomes a chance of a home win, a draw (in sports that have them) and an away win.
How quickly ratings react is set per sport. A single baseball game says less about a team than a single football match, so baseball ratings move more slowly.
How the model is tested
The model is refitted every night on recent results. We tested longer histories and found they made predictions worse, because old seasons anchor a team to form it no longer has. A refitted model is used only if it beats a simple baseline on matches it was not trained on.
Once a sport has enough graded games, its probabilities are also calibrated against that record. The aim is that a 60% chance means the favourite wins about six times in ten.
How predictions are graded
When a match finishes, the prediction is scored against the result. Only the last estimate made before kick-off counts, so a forecast that was updated many times is still graded once. The record is used to retune the model, not to decide which results to show.
The limits of a prediction
A probability is an estimate, not a promise. Some sports are harder to predict than others, and an upset is part of sport, not a fault in the model. Every probability on Pacy is labelled as Pacy's estimate. It is not betting advice, and Pacy takes no money from bookmakers.
Corrections to data
A score corrected after publication is marked as corrected on the match, so a reader who saw the wrong figure knows it changed. See the corrections policy, or report a wrong score to [email protected].