The principle: publish before, verify after
AIGoalData’s method rests on a rule that allows no exception: every public prediction is recorded before kickoff, then compared with a verified result. In between, nothing is rewritten. This page describes the steps every match goes through, from its discovery to the evaluation of its result, and the limits we accept.
It explains how the service works, not the details of the model’s calculations, which remain internal. What you can check, on the other hand, are the publication times, the decisions taken and the results obtained.
1. Discovering the matches
Every day, the service lists the matches in the competitions it covers. For each one, it identifies the two teams, the competition and the kickoff time. An uncertain kickoff time, a postponed match or a poorly identified team is enough to set the match aside: publishing an analysis of a poorly identified match would make no sense.
The number of published matches therefore varies from day to day: it depends on the competition calendar and the quality of the information available, never on a volume target.
2. Validating the data
Before any calculation, the data is checked: the teams must match, the matches used must have been played before the game being analysed, duplicates are removed and the origin of the statistics is kept. When a necessary piece of data is missing, the market concerned is marked “not evaluated” rather than producing a fragile number.
This caution explains why some days have fewer published matches than others: we prefer to publish less, but to publish analyses whose basis is known.
The statistics used must also be consistent with each other: same team, same period, same competition where relevant. A detected inconsistency, for example two different identities for the same team, blocks the market concerned until the data is reliable.
3. Calculating with the historical V1 model
The estimates are produced by the historical V1 model. Its calculations are frozen: they are not readjusted after matches to improve how past results look. Whenever the site’s code changes, an automatic check replays a set of historical cases and blocks the release if even one result differs.
The model produces probabilities for several scenarios (result, goals, both teams to score), expected goals for each team and estimated scores. These are estimates, presented as such on every card.
4. Deciding market by market
The markets covered are handled separately: both teams to score, over 2.5 goals, straight winner, team over 1.5 goals and underdog over 0.5 goals. For each one, the decision can be a selection, no bet or no evaluation.
There is no quota of selections, and no obligation to offer something every day. No bet means the numbers did not justify taking a position; it is not counted as a loss. A selection on one market triggers no automatic decision on another.
5. Publishing before kickoff
The prediction is recorded with its creation time, before the match starts, then published on the match card. From that moment on, it no longer changes. If a calculation could not be recorded before kickoff, no prediction is shown for that match; it is never completed after the event.
For some matches, an analysis article is also published. It is written with the help of a language model, using only the recorded data, and an automatic check makes sure every number quoted matches the source data exactly. An article that fails this check is not published.
6. Collecting and verifying the result
After the match, the final regulation-time score is collected. It first appears as “reported”. It becomes “verified” after a second reading of the source or a review of evidence. As long as a result is not verified, it enters no performance calculation.
In case of a discrepancy or doubt, the result stays pending rather than being validated too quickly. A cancelled or abandoned match is not turned into a win or a loss.
7. Evaluating honestly
The Performance page calculates success rates only from selections published before the match and from verified results. Losing selections appear there just like winning ones, and each market is evaluated separately.
A success rate calculated over few matches remains fragile. We show the sample size so that everyone can judge its reliability, and we present no past result as a promise for the future.
What you can check yourself
This method is designed to be checkable from the outside. On every card, the publication time lets you confirm that the prediction came before kickoff. On the Results page, every score carries its status, reported or verified. On the Performance page, rates are presented market by market, with the sample size.
If you spot an error in a displayed score or result, write to us at the address shown on the Contact page, naming the match concerned and the source you consulted. Every report is compared with the recorded data before any correction, and correcting a result never changes the prediction published before the match.
Free access, VIP and advertising
The main reading of every published match is free. The VIP subscription gives access to more detail, such as statistics split between home and away, the full head-to-head history and VIP selections, but it relies on the same frozen calculations and adds no certainty. Signed-in VIP members see no advertising; for other visitors, advertising depends on their consent.
The limits we accept
No method knows line-ups announced late, last-minute injuries or unpredictable incidents. The estimates rely on the data available before the match and do not describe everything. Football keeps an element of chance that nobody can remove.
AIGoalData is not a betting operator and gives no personal advice. Betting carries a risk of losing money and is forbidden to minors. If gambling becomes hard to control, see our Responsible gambling page.