Frequently asked questions
Eight questions about what we compute, from which data, and what our numbers do not say.
- Are your analyses reliable?
If reliable means the called result will land, then no, and nobody should promise you otherwise. What we do hold to is narrower. The method does not change from one match to the next, it is written out in full on a public page, and two reads of the same numbers give the same answer: there is no element of chance in it. What we do not hold is the inputs. We receive no list of injured or suspended players, so every match reaches the model marked none reported, including the ones where a first-choice player has been ruled out since the day before. And the smaller the competition, the less history sits under the number. An analysis is never better than the data feeding it. Ours has holes, and we would rather name them.
- Where does your data come from?
One commercial data provider, and that is the whole list. Fixtures, scores, lineups, tables, match statistics, odds: all of it arrives from there. We scrape no third-party site and collect nothing ourselves. The downside shows up immediately: when our source is wrong, so are we, and we have no second source to contradict it. A match status, for one, is a field somebody writes rather than a measurement taken on the pitch: fixtures stay flagged live long after the final whistle, so we close them ourselves once enough time has passed rather than display a game that ended hours ago.
- How often is it updated?
It depends what you are looking at, and the gaps are wide. A live page asks for scores continuously. Recent form and head-to-head records do not move while a match is being played, so they are not re-requested over and over. The one worth knowing about is the pre-match analysis: it is computed once, then reused. The read you open at midday for a nine o'clock kick-off was produced without team sheets, and it stays that way until the next run. During the match the read can be recomputed as the score moves, a recompute reserved for the Pro plan.
- What does a percentage on a match mean?
A frequency, not a verdict. When a page says 68% for the home side, it means that if this set-up repeated a hundred times, that side would win about sixty-eight of them. The other thirty-two are not the model failing, they are part of the number. A low percentage is not a bad percentage: 52% on a tight match is honest information, and a model that never printed anything under 90% would be broken. The confidence level is not printed beside that percentage: the analysis panel never renders it. It shows up on the post-match banner and in your notifications. It is a threshold rather than a grade: it says how far apart the three outcomes are, not whether the called one will land.
- What is the difference between Free, Fan and Pro?
Free is the public tier, with ads on it. Fan, at €6.99 a month or €49.99 a year, takes the ads away. Pro, at €12.99 a month or €89.99 a year, opens the three things our servers genuinely hold back for it: the in-play recompute during a match, the AI agent you put questions to, and a personal round-up prepared overnight for active accounts. The subscription page sets out the rest of the offer. What a pricing table never tells you: the model and the method are identical on all three plans. A subscription buys fresher reads and tools around them. It does not buy a better prediction.
- Are your predictions betting advice?
No. Our predictions are informative. We give no betting advice, no recommended stake, no selection to play. The odds shown on a match page come from our data source and are there for reference. We guarantee no outcome and promise no winnings. If you bet, the decision is yours, made elsewhere, on your own responsibility, and your country's rules apply. A 68% probability also means being wrong roughly one time in three.
- Why is a statistic sometimes missing?
Because the data did not arrive, and an empty box beats an invented number. Most often there is nothing to send: form records thin out fast below the top divisions, and where there is none the block does not appear at all. With no table, the model is told the position is unpublished instead of being handed a default number, and which of the two cases we are in is often beyond us. Assembling a day of sport also takes a great many separate requests, so on a busy evening we choose what to re-request and what to leave. And when a request fails, the page is built without that block instead of erroring out; an empty answer is treated as suspect and asked for again soon, because it is more often an outage than a fact. The permanent gap is elsewhere: no injuries and no suspensions reach us, on any match.
- How do you measure how accurate your analyses are?
Not yet in a way worth publishing, and this is the least flattering answer on the page. Comparing a called outcome with a final score is easy enough, match by match. What is missing is the run. We keep no dated log of what we announced: a newer analysis replaces the previous one, and old ones are not kept indefinitely. So there is no dated record of what we said before the result was known, which is exactly what a rate would have to be built from. What the statistics page shows is a current accuracy, computed over the recent predictions we actually issued: honest, but it moves and no run can be rebuilt from it. Once every prediction is written down at the moment it is produced, a rate will appear with its window and its sample size, and not before.