The model is 3.1% behind the sharpest book, and that is the whole story
Scored on 9,487 matches priced before they were played, the model's Brier is 0.5962 against Pinnacle's 0.5783. It beats 0 of 5 benchmarks. Here is what that does and does not mean.
Why a losing benchmark is still worth publishing
A model that trails a sharp closing line by two percent of Brier is not a failed model; it is an honest one. The closing price contains every piece of team news, every injury and every stake placed by everyone who knew something. What this model offers instead is a coherent price for every market on every fixture, with a stated band and a coverage guarantee, at moments when no closing line exists yet.
Where the money actually is, if it is anywhere
The books disagree with each other more often than any of them disagrees with the truth. Taking the best price rather than the average is worth more than the model's entire forecasting edge, and when the best prices contradict each other outright there is a position that cannot lose. Those are execution facts, not forecasts, and they are the part of this business that survives contact with a real book.
Walk-forward: each match priced before it was played, then scored against the price each firm actually showed, with the overround removed by the power method. Each book is scored only on the fixtures it priced and the model on that same subset.
Beating a retail book is ordinary and beating a sharp closing line is not something this model does. The arbitrage figures come from an envelope of recorded prices that may not have been simultaneously available, and books restrict accounts that only take them.