Our number, the market's number, and what the game actually did. Every graded game, from the first one — not a window chosen to make us look good. We publish a number that often disagrees with the book, and this is the page that says whether that disagreement has been worth anything.
Two things worth understanding before reading it. Bias and accuracy are different. A number can sit almost perfectly on average and still miss wildly game to game — and a market can be reliably a few points off in one direction while still landing closer more often. Both matter, for different reasons. And every figure here is the number we actually published at the time, never one recalculated later; today's ratings already know how these games finished, so grading with them would be marking our own exam with the answers in hand.