Feature Vector

The ordered set of numeric inputs a model scores a game with — market prices, splits, ratings, weather, rest and more, each reduced to a number. Every prediction here is generated from a versioned feature vector so results stay reproducible.

• Calibration — How well a model's stated probabilities match reality — a well-calibrated 60% pick wins about 60% of the time. Shown on a reliability curve.
• Edge — The model's estimated advantage over the market — the gap between its projected win probability and the probability implied by the price.