When we say 30%, it happens about 30% of the time.
We publish a win probability for every horse, then grade every prediction in public against the official results. This page is the running report card, updated November 14, 2026. Calibration measures accuracy, not profit. You can still lose money.
Historical calibration, 2023 to 2025
754,746 horses in 102,163 races from 2023 to 2025, all races the model never trained on, graded against the official results.
102,163 races · 754,746 runners
Live calibration since August 17, 2026
Every horse we have predicted since August 17, 2026, 69,392 so far, graded against the official results. Updated November 14, 2026.
9,543 races · 69,392 runners
On October 5, 2026 we moved to a new generation of our model. Every win probability published from that day on comes from it. The live chart keeps every graded prediction from both generations, so nothing has been removed from the record.
The board's implied win chances add up to more than 100%, because the takeout is built into every price. Ours add up to 100% in every race.
- 2023 to 2025
- 102,163 races · 754,746 runners
- Since August 17, 2026
- 9,543 races · 69,392 runners
Last 30 days
The same live record, restricted to recent races. Shorter windows swing more; the diagonal is still the target.
2,464 races · 18,607 runners
Last 90 days
The same live record, restricted to recent races. Shorter windows swing more; the diagonal is still the target.
9,476 races · 68,945 runners
Why not just bet the favorite?
Because the crowd is genuinely good at ranking horses. Across our 2023 to 2025 test data, the public favorite won about 39% of races, while our model's top-rated horse won about 36.8%. We publish that number because it is true, and because it explains what this model is for. The market concentrates its confidence on one horse; our model spreads probability honestly across the whole field. Both can be calibrated at the same time. Favorites win most often and still lose money over the long run, because the takeout is built into every price. Our job is not to out-pick the favorite. It is to price every horse honestly. Calibration measures accuracy, not profit. You can still lose money.
Hit rate is the wrong scoreboard
Win percentage is the number every tout leads with, so here is ours, told straight. From August 17, 2026 to November 14, 2026, our top-rated horse won 35.7% of the 9,525 races it started in. A $2 win bet on every one of them would have lost 14.7% of the money staked, which is roughly the track takeout. Hitting often is easy if you take short prices; the payoffs are just too small to cover the house's cut. That is why we don't sell hit rates, and why this page measures calibration.