Live win probability for every professional tennis match, updated as the points are played — plus the moments that matter: who is under pressure, who is about to be broken, and when a scoreline stops meaning what it looks like. ATP · WTA · Challenger · W125 · ITF, every day.
Free through beta — no card. Your email is also your sign-in.
or see today's value boardIs it mispriced, is it moving, and am I actually any good at this? Everything on the terminal answers one of those.
Every match gets a win chance that updates as points are played. But a number is only useful if it has been checked against what actually happened, so we grade ours against finished matches and publish the result — including the states where we are not accurate enough to be worth listening to.
Sets and games are the last thing to change. A live momentum engine tracks serve regression and break pressure point by point, so a break coming is visible before it lands — and so is a lead that has stopped meaning anything.
Every intended and placed trade is one row in a journal that settles itself against the market. The calibration report then bins predicted probability against what actually happened — which is the only thing that can tell a sharp model from a merely confident one.
The mispricing is rarely in the Sunday final. It is in the Challenger second round that three people are watching.
A live walkthrough of the terminal — reading True P, spotting an edge against the book, sizing the stake and timing the hedge.
Prefer to read? The written trading manual covers the same ground in detail.
Drop your email for early access, or reach out about a partnership.
or email jessefuture10@gmail.com
The only question worth answering. Here is how we check, what came back, and where we are not good enough to be worth listening to.
We keep every live match we have ever shown. To grade ourselves we replay them point by point, ask the model what it would have said at each moment, and compare that against what actually happened next. Crucially, the matches used to check are not the matches used to build — otherwise a model can score itself on its own homework.
In lopsided positions — a player two points from an easy hold — our numbers were overconfident by as much as 40 points before we corrected them, and even now we do not trust them at the extremes. So in those situations the terminal says “too one-sided to call reliably” and shows nothing. That is deliberate. A confident number we have not earned is worse than no number.
Not a tipping service and not a profit claim. We publish probabilities and the evidence behind them. Whether that is worth money to you depends on what you do with it, and on prices we do not control.
Measured 18 September 2026. Worst single bucket 6.5 points off. These figures are re-measured by hand rather than generated live, so what you see is a number a person checked.
Methodology, access and the limits of what a model can tell you.