Calibration-based accuracy scores for every major sportsbook. Built from real closing lines and outcomes across NBA, NHL, and MLB.
A perfectly calibrated book prices lines where the over hits at exactly the implied probability. The gap between cyan and amber reveals systematic bias across all books.
Every betting line makes a prediction. When a book prices an over at -130, they're implying a ~56.5% probability. We test that claim against reality.
For each line, we take the canonical side (home team for moneylines/spreads, over for totals/props), compute the devigged implied probability from the closing price, then check if that side actually hit.
Calibration Error = |actual hit rate - average implied probability|. A book with 0.000 calibration error prices lines perfectly. Every point of error means their prices are systematically off.
Bias tells you the direction. Negative bias (most books) means the book overestimates probability - they shade toward the public side, extracting vig from recreational bettors.
We convert calibration error into a devig weight using: weight = 1 / (1 + 15 x CalErr). This weight determines how much each book's line contributes to our fair value calculation. Circa (weight 0.999) has 5x the influence of Kalshi (weight 0.209).
Data source: 12,979,710 graded lines from 74 sportsbooks across NBA, NHL, and MLB. Historical closing line data covering the 2024-25 and 2025-26 seasons.