Everything below this is a measurement, and measurements are only worth anything if you know what was measured. Three minutes here and the record further down will read plainly.
Every fixture is simulated thousands of times using the squads, recent form and conditions we have. A 62% win probability means that side came out ahead in about 62 of every 100 of those runs. It does not mean they will win, and it is not a recommendation to back them.
A bookmaker's price is a probability with their margin built in. Take the margin out and you have what the market genuinely thinks. We publish that next to our own number, so the disagreement between them is visible rather than hidden inside a tip.
Each projection is committed to a file while the game is still ahead of us, then settled against the official result. Nothing is added afterwards and nothing is quietly dropped. A record you can edit after the fact is not a record.
When our number is higher than the market's on the side we favour, that distance is what we call the gap. It is the most interesting thing on the page and the most easily misread. A gap means the two estimates differ — it does not mean ours is the right one. Most of the time the market is closer, which is exactly what the record below shows.
Picking the favourite clears 50% most weeks. Any model should manage it, and a number above half proves very little on its own.
The market price is the sharpest public estimate of a result that exists. Consistently landing closer than it is the only thing that would justify selling a signal rather than a tool.
Which is why nothing here is sold as an edge. You are paying for the projections, the market comparison and the tools around them — not for a tip. The moment a sport does clear it, this page will say so, with the sample size beside it.
Each number below was locked to a committed file while the game was still ahead of us, then settled against the official result. Nothing is added afterwards and nothing is quietly removed. Losses appear at the same size as wins, because a record that hides them is an advertisement.
Win rate is how often the projected winner actually won. Draws are excluded. It is shown with a 95% confidence interval — the range the true rate plausibly sits in given how many games have settled. When that interval spans 50%, the sample cannot yet tell you the model is better than a coin flip, and we say so rather than quoting the midpoint.
Brier score measures how well-calibrated the probabilities are, not just whether the pick won. Lower is better; 0.25 is what you get by saying 50% to everything. It is the honest measure for a probability model, because a model can be right often and still be badly calibrated.
Skill vs market compares our Brier score against the bookmaker favourite locked in the same rows. Above zero means we were better calibrated than the closing market. This is the number that decides whether we are allowed to claim an edge — and no sport has cleared it yet.
Market-derived means the published probabilities come from de-vigged bookmaker odds rather than a model of ours. For those sports the graded pick and the market favourite are the same side by construction, so the percentage measures the market, not us. It is labelled every time it appears.
We publish no profit figure, no ROI and no unit total, and we claim no edge over the market on any sport — because on the evidence above we do not have one we can prove. When a sport's skill score clears the market with a confidence interval that stays above zero, we will say so here and show the working.
This section lives there too — you were sent a direct link to it.