How we calculate
Exactly how a verdict is produced: what we read from the picture, what we compute, and where the thresholds sit.
What we actually compute
A bookmaker’s odds are its opinion about how often something happens. You can pull a probability out of odds: 1.85, for example, corresponds to roughly 54%.
We compute our own probability for the same event from team data and compare it with the bookmaker’s. The difference between the two, in percentage points, is the only thing we sell.
How we read a slip
The model reads matches, markets and odds off the image the same way a person would. We do not connect to bookmaker systems, so a screenshot from an app, from a browser, or even a photo of a screen all work.
If the match, market or odds are not legible in the image, the analysis will not run, and we will say so plainly.
Where our probability comes from
Team form over recent matches, home and away counted separately.
Line-ups: missing key players move the estimate noticeably.
Tempo and expected goals per 90 minutes.
Head-to-head history.
Line movement: if the odds moved, the market learned something, and that is a signal too.
Every factor enters the estimate with a weight, and all weights go into the PDF report — you can see exactly what moved the number.
The five percentage point threshold
If our probability is five percentage points or more above the bookmaker’s, we say "worth backing": the bookmaker is paying more generously than it should.
The threshold is not arbitrary. Below it the gap is comparable to the model’s own margin of error, and calling that an edge would be a lie.
The threshold is computed on our side and may change together with the model.
Model confidence
Confidence is about whether the model had enough data: enough matches played, line-ups known, no long mid-season break.
Below 70% confidence we show no verdict at all. Better to stay quiet than to guess out loud.
What we do not do
We do not predict the score or the outcome of a particular match.
We do not build match lists or publish an "accumulator of the day".
We do not promise profit, neither over distance nor on a single bet.
How we check ourselves
Every verdict is recorded before kick-off and never edited afterwards. Records are immutable and the timestamp is kept.
We do not show a "win rate" — that is somebody else’s metric and it depends on which odds you pick. We show calibration: how closely the probabilities we stated matched what actually happened. We use the Brier score for this.
The open log is on the site and includes the bad verdicts too.
Limits
Pricing improves the quality of your decisions over distance. It does not make an individual match predictable: we can be right about the price and still lose that particular bet.
Past accuracy does not guarantee future results.