The paper reports whichever outcome the markets price highest, so it is wrong whenever the favourite loses, which is often and by design. Accuracy alone therefore says little. What matters is whether the confidence attached to a claim means anything: of the things called ninety per cent likely, roughly nine in ten should happen.

The Brier score measures exactly that, over every claim at once. Zero is perfect, 0.25 is what you would score by calling everything a coin flip, and lower is better.