There is no ranking on this page
We have not installed, tested or measured a single football prediction app, so we are not going to tell you which is best. Anyone who publishes that ranking without running the apps’ predictions against results for a season is telling you about their affiliate deals, not about the apps.
What follows is more useful and much cheaper: six checks that separate a product with a model behind it from a product with a slogan behind it. They apply to apps, to websites and to this site.
Six things to check
In rough order of how much they tell you.
Does it give you a probability, or a pick?
A pick is unfalsifiable one match at a time — it was right or it was unlucky, and you cannot tell which. A probability commits: called at 60%, it has to land about six times in ten across many matches, and it can be checked. Products that only ever show you a selection have chosen the version that cannot be scored.
Does it say what the number is made of?
Not “advanced algorithms” or “AI-powered” — the actual inputs. Which matches, how far back, weighted how, with what data. If the description would fit on any product in the category, it is describing nothing.
Is there a backtest, with a date and a sample size?
A record on a season the model was not fitted on. Look for the number of predictions scored and, just as importantly, the number skipped and why. A record where nothing was ever skipped and nothing was ever wrong is a record of a marketing decision.
Does it show calibration, or only a hit rate?
This is the check that catches most of them. Always backing the favourite produces a high hit rate and tells you nothing — a model earns its keep by getting the size of the probability right, not by picking the obvious side. If a product will only quote you a percentage of correct picks, it is quoting the number that is easiest to make look good.
Can the accuracy claim come out badly?
Ask what would have to happen for it to be false. “85% accuracy” with no period, no market and no sample cannot be false, which is why it is written that way. A claim worth reading names the season, the market and the count.
Does it separate the price from the result?
“Value” is a statement that a price is wrong, not that an outcome is certain — a genuine edge on a 25% shot still loses three times in four. Anything that presents an edge as a likely win has either misunderstood its own number or is counting on you to.
Run the checks on us
Probabilities on every fixture rather than picks, the inputs and constants written out on the mathematics page, and a dated backtest on a hold-out season with the skipped fixtures counted and the calibration error published league by league on the accuracy page — including the league that missed our own tolerance. Check number four is the one we would point at first.
There is no app. BCCFC is a website, and everything it does works in a browser.