How accurate are our probabilities?
Our whole approach rests on one thing: when we say a horse has a 30% chance, it should win about 30% of the time. Here's the proof — measured against real results, published openly.
Across 1,822 runners the model has assessed since 24 May 2026, its predicted win probabilities land within 1.1 percentage points of what actually happened, on average.
In plain terms: when we say 30%, horses win close to 30%. That's what makes the value calculations on the race page trustworthy — the probabilities are real.
Predicted chance vs actual win rate
Every horse the model assigns a probability to, grouped by predicted chance, compared to how often those horses actually won. The closer "Actual" is to "Predicted", the more trustworthy the number.
| Predicted Chance | Actual Win Rate | Difference | Runners | Accuracy |
|---|---|---|---|---|
| 0–5% | 4.8% | +0.5pp | 104 | |
| 5–10% | 7.9% | +0.3pp | 681 | |
| 10–15% | 14.5% | +2.3pp | 477 | |
| 15–20% | 17.1% | -0.3pp | 240 | |
| 20–25% | 19.4% | -3.0pp | 170 | |
| 25–30% | 25.8% | -1.3pp | 97 | |
| 30–35% | 32.1% | -0.2pp | 53 | |
| 35%+ small sample | 32.7% | -8.3pp | 49 |
pp = percentage points. Green = within 3pp · Amber = within 7pp · Red = needs attention. Based on 1,871 runners across 10 racing days since 24 May 2026.
Accuracy by race type
Whether the probabilities hold up across different kinds of race. Predicted is the model's average chance; actual is how often those horses won.
Why we show you this
Most racing products show you their winners. We show you how accurate our numbers are — including where they're not perfect.
A well-calibrated probability is the foundation of finding value. If the model says a horse has a 25% chance and the market is offering odds that imply 15%, that gap is only meaningful if the 25% is real. This page is our evidence that it is — updated continuously from settled results.
If the numbers ever drift, you'll see it here first. That's the deal.