Measured, not claimed
When we say 68%,
it happens
67%
of the time.
Every probability on this site has been graded against what actually happened, out of sample, 3,963,231 times. The plot below is that grading, not a promise about it.
said 3% · landed 2.5% · 500,947 results
said 8% · landed 7.8% · 351,384 results
said 13% · landed 12.7% · 297,487 results
said 18% · landed 18.0% · 273,908 results
said 23% · landed 23.1% · 269,268 results
said 28% · landed 27.6% · 236,292 results
said 33% · landed 32.6% · 190,543 results
said 38% · landed 37.3% · 184,975 results
said 43% · landed 42.4% · 168,931 results
said 48% · landed 47.8% · 214,298 results
said 53% · landed 51.8% · 194,270 results
said 58% · landed 57.3% · 145,691 results
said 63% · landed 62.3% · 140,279 results
said 68% · landed 67.5% · 130,544 results
said 73% · landed 72.5% · 133,334 results
said 78% · landed 76.9% · 126,064 results
said 83% · landed 81.8% · 116,505 results
said 88% · landed 86.8% · 106,264 results
said 93% · landed 91.6% · 89,275 results
said 98% · landed 96.5% · 92,972 results
Next up 10,635 fixtures priced ahead
Al Shabab
Al Khaleej
Brisbane Roar
Perth Glory
Strongest on the board ranked by evidence, not by odds
The record every out-of-sample selection, settled
3,462,284
selections settled
42.14%we said they would land
42.15%actually landed
0.01points apart
Graded by a model fitted only on matches played before each one it priced, so the number on the left is a forecast rather than hindsight. Where a model has stopped beating its own baseline, the page for that sport says so above its numbers. See it broken down →
What you can do here
Scan the board
Every market across every league, ranked by how much evidence sits behind
the call rather than by how big the price is.
Price one match
Open a fixture and the model reads both sides' whole recorded history,
then prices every market on it and shows its working.
Check a slip you already have
Paste a booking code. Each leg is judged against our own number, and the
weak ones can be swapped for something the record supports.
Every figure on this page is a model estimate, not a certainty. A 70% call loses three times in ten, and this page exists to show you exactly how often. Please bet responsibly, and only what you can afford to lose.