How we grade Betwizzard's own predictions against what actually happened.
A Brier score measures how well a predicted probability matched what actually happened, across many predictions. Lower is better. 0 means every prediction was perfectly confident and correct. 1 means every prediction was perfectly confident and wrong. 0.25 is what you'd get by guessing 50% on everything — the score a coin flip earns.
For each resolved market, take the predicted probability (0 to 1), subtract the actual outcome (1 if it happened, 0 if it didn't), and square the result. Average that across every resolved market in the category:
Brier score = average( (predicted probability − actual outcome)² )
A Brier score computed on a handful of resolved markets is mostly noise — a few lucky or unlucky calls can swing it wildly. We only show a score once a category has at least 20 resolved markets in the trailing 90 days. Below that, we show the actual count instead of a number that would look more authoritative than it is.
Only predictions Betwizzard classified as "modeled" — meaning Grok found genuine market-specific evidence, not just a category-level guess — are graded here. A "heuristic" (category base-rate) estimate is never held to this standard, because it was never presented as more than a guess in the first place.