The finding that expert political forecasters perform barely better than chance, and worse than simple statistical rules, has been repeated often enough to become a genre of commentary in its own right. What the commentary usually omits is the structure of the finding: the aggregate concealed considerable variation, and the variation tracked style of reasoning rather than credentials, subject knowledge or access to information. Forecasters who held multiple partial explanations, updated in small increments and treated their own predictions as revisable did substantially better than those working from a single organising framework, and the difference was stable across domains. This is a more useful result than the headline and less congenial to everyone: it does not vindicate the sceptics about expertise, and it offers no comfort to the experts either.