WC2026 Model Tracker
Bivariate-Poisson predictions vs actual results ยท a live, honest experiment
Performance dashboard
Match outcomes called correctlyrunning win/draw/loss accuracy, in kickoff order
Share of matches whose result (home win / draw / away win) the model called right, running from game 1. It settles at 58% โ well clear of the 33% you'd get guessing between three outcomes. Early swings are just the small sample settling.
Asian-handicap outcomeswhere every bet landed
Full Win40Half Win2Push6Half Loss4Loss29
How often the model was rightthe model's own accuracy, by bet type ยท pushes excluded
Each bar is the share of games the model's pick was correct for that bet type (1X2 = match result, O/U = total goals over/under 2.5, BTTS = both teams to score, plus corners, cards and the Asian handicap). This is the model's scorecard, not the bookmakers'. 50% is the dashed line.
Winner calls
Did the model pick the right team to win? Group games are graded on the 90-minute result (a draw counts as its own outcome); knockouts on who advanced โ so a tie settled in extra time or on penalties is judged by who went through, not the 90-minute score. Open a match for the full market-by-market breakdown.