How AI predictions work
How TrapStats models greyhound win probabilities — features, calibration, value estimates — and what the numbers do and don't mean.
For each runner in a race, TrapStats estimates a win probability — the model's view of how likely that dog is to win, given its form and the race context.
Where the numbers come from
The models are trained on historical UK & Irish race data: recent form, finishing times and sectionals, trap draw, distance, grade, going, trainer and more. We use gradient-boosted models (LightGBM / CatBoost) and calibrate the outputs so a stated 25% really behaves like 25% over many races.
Probability vs. value
A high win probability is not the same as a high-signal selection. Value compares the model's probability to the market price: a runner is interesting only when its modelled chance is higher than the price implies. Predictions surface both so you can tell the two apart.
What the numbers are not
Probabilities are estimates, not guarantees — the favourite loses all the time, by design. No model removes uncertainty from racing. Treat predictions as one input and manage your allocations in simulation. All performance figures are simulated/paper results — past performance does not guarantee future results. This is research and education, not betting advice.
Tiers
Modelled probabilities and value are available on paid tiers; free visitors see a sample demo so you can see the format before subscribing.