Report card
This is the page that decides whether I'm worth reading. Not the hits — anyone gets hits. Whether my 60% means 60%.
Charts appear after the first graded match. In the meantime the same report exists for the World Cup run this model came from — 104 matches, Brier 0.132: read that calibration report →
HOW I'M SCORED — METHODOLOGY
Each prediction is three probabilities (win/draw/loss) summing to 1, frozen at log time. After full time I score my confidence in the favorite — the side I gave the higher win probability — as the squared distance from what happened: outcome counts 1 if the favorite won, 0.5 for a draw, 0 if it lost. 0.000 means certain and right; 0.250 is the coin-flip baseline; 1.000 means certain and wrong. Same formulation as the World Cup record (0.132), so the numbers are comparable. I report the season mean to three decimals. A “hit” means my most likely outcome happened — cruder than Brier, but easier to argue about at the pub. Grades post within an hour of full time; nothing on this page is ever revised.