Methodology
Backtest — simulated results
Everything on this page is a simulation. None of it is a record of money at risk.
This is not our track record
A backtest asks “how would this method have done?”. A track record says “here is what we actually published, before the result was known”. They are different claims and only the second one is evidence. Backtest figures are never added to, blended with, or shown alongside the numbers on our results page.
How the backtest runs
It walks forward through history a week at a time. At each step the model is refitted using only matches that had already been played at that point, then evaluated on the following week’s fixtures. The prices used are the pre-match prices that would have been available.
The script aborts outright if any match at or after the evaluation date reaches the fit. That guard matters: leaking even a little future information into a backtest inflates the results dramatically, and it is the single easiest mistake to make in this kind of analysis.
Why we do not publish a headline ROI here
Because it would be meaningless to you. A backtest ROI depends entirely on which prices you assume you could have taken, and anyone can produce an impressive one by choosing generously. The figure that means something is the live record, and that is on its own page whether it flatters us or not.
What the backtest is genuinely useful for is calibration: checking that when the model says 60%, things happen about 60% of the time. We test that against a Wilson confidence interval rather than eyeballing it, and calibration is reported over every forecast rather than only over the bets we advised.
Known limitations
- Historical prices are not always the best price that was available, so backtest returns can flatter or understate what was really achievable.
- The model’s own estimation error means an apparent edge is partly noise. Selecting bets on apparent edge therefore selects, in part, for our own mistakes — a real effect that a naive backtest hides.
- Team strengths drift with transfers, injuries and management changes in ways historical data cannot fully capture.