When our model says a horse is 70% to place, that number is only useful if it's true. A model can be confident and wrong, or modest and right. The property that separates the two is calibration — and it matters more than most people realise.
What calibration means
A model is well-calibrated if its stated probabilities match reality: of all the horses it rates at 70%, roughly 70% should actually place. Of the ones it calls 30%, about 30% should. Plot "predicted" against "actual" and a perfectly calibrated model sits on the diagonal.
Why it matters for value
Betting value is the gap between your probability and the market's implied probability. If your numbers aren't honest, that gap is fiction — you'll "find value" that isn't there and miss the value that is. Calibrated probabilities are the foundation everything else is built on: staking, selection, and the value judgement itself.
How it's measured
The standard tools are reliability diagrams (predicted vs actual, bucketed) and summary scores like expected calibration error (ECE) and the Brier score. Lower is better; the goal is probabilities you can take literally. Raw model outputs are often over-confident, so we apply a calibration step so a "70%" really behaves like 70%.
Confidence you can read
This is why our tips carry a model probability you can actually trust, and why our track record will report calibration alongside strike rate and CLV. An honest 60% is worth far more than a hopeful 90%. See the glossary for the related terms.