Methodology
How We Make Predictions
A strength rating for each side, adjusted for the situation, converted into a probability and published as it comes, in six steps with no tipster override.
Every rating on this site is produced by a model, not by a tipster, and the number it produces is published as it comes. The method is the same six steps for all ten sports. What feeds step two is different in each, and has to be.
- No human override. Nobody adjusts a number because a fixture feels wrong.
- One tip per fixture, named to its market, so the call can be judged.
- Calibration over strike rate. Of everything rated 60 percent, close to 60 percent should win.
- The limits are published too, including where the model is weakest.
How Is a Prediction Made, Step by Step?
The same six steps run on every sport. What goes into step two is different in each.
- Collect the data Fixtures, results, line ups and odds arrive from API-Sports for every sport we rate. Nothing is typed in by hand, which is what lets the same process run on 300 fixtures a day.
- Rate each side Each team or fighter carries a strength rating built from results, scoring and conceding rates, and the quality of who they played. Recent matches count more than old ones, on a decay rather than a cutoff.
- Adjust for the situation Home advantage, rest days, travel, confirmed absences and whether the fixture matters to either side. A team already promoted in May is not the team that played in March.
- Turn the gap into a probability The difference between two ratings becomes a percentage for each outcome. Three percentages where a draw is possible, two where it is not, and they always total 100.
- Pick one call per fixture Across the markets we rate, the tip is the selection with the best combination of probability and odds. One per fixture, named to its market, so you are never guessing which outcome it is on.
- Publish, settle and count The call goes live with its probability attached. When the fixture ends it is marked won or lost on its own page and added to the published record. Nothing is edited afterwards.
What Goes Into the Model, and What Stays Out?
Only what can be collected consistently for every fixture in a competition.
Why Is Each Sport Modelled Differently?
The framework is shared. The inputs are not, and cannot be.
What Can This Method Not Do?
The honest boundaries, because a method described without them is a sales page.
About Our Prediction Method
What the model reads, what it ignores, and where it is uncertain
How we make predictions: a strength rating for each side, adjusted for the situation, converted into a probability, and published with the number attached. The same six step process runs on all ten sports, and no human overrides the output before it goes live.
What Goes In
Results and scorelines, home and away records kept separately because they behave differently, goals or points scored and conceded, the strength of the opposition in each of those matches, rest days between fixtures, travel, confirmed team news, and competition context such as whether either side still has anything to play for. Recent matches are weighted more heavily than old ones on a decay, so form matters without a single bad afternoon erasing a season.
What Is Deliberately Left Out
Pundit opinion, public sentiment, social media, and anything we cannot get consistently for every fixture in a competition. A variable available for the Premier League but not the NPFL would quietly make our African coverage worse than our European coverage, so it is not used at all. This is also why we rate three markets on the non football sports rather than twenty: the other seventeen are not available everywhere.
Probability, Not Prophecy
A call rated 62 percent is expected to lose nearly four times in ten. That is the model working, not failing. The test of a probability is calibration: of everything rated at 60 percent, close to 60 percent should win, no more. A model whose 60 percent calls land 85 percent of the time is mislabelled rather than brilliant. Our published accuracy record is there so this can be checked rather than taken on trust.
Why We Publish One Tip and Not Ten
A page listing every market with a percentage against each is a page that will always have been right about something. One call per fixture, named and settled in public, is a claim that can fail. That is the point. The analysts who write around the number are on the team page, the rules they work to are in the editorial policy, and the model's output for today is on the all sports hub.
Frequently Asked Questions
What people ask about how the predictions are produced.