Where the MLB gap lives: information, not calibration
mlb: model Brier 0.1668 is worse than market 0.1510 (gap=+0.0158); driven mainly by resolution (information, not fixable by recalibration) (reliability_gap=+0.0022, resolution_gap=-0.0149). soccer_intl: model Brier 0.3325 is worse than m...
Data integrity
These MLB/soccer numbers are revision 2, computed on the segment-clean corpus (2026-09-16). Revision 1 values are withdrawn and kept in the regeneration receipt. Segmentation excluded 51,635 of the 78,986 MLB ticks (65.4 percent); the specifically identified label mismatches were a smaller 27,076 (34.3 percent).
- murphy_decomposition
How to read this figure
Each reliability curve plots predicted probability against actual outcome frequency across 10 bins; the closer the curve hugs the diagonal, the more honest the numbers.
Continue with the data
Open the interactive inspector: Score decompositionWhat it means
A Murphy decomposition splits a Brier score into reliability (are the probabilities honest), resolution (do they separate winners from losers), and an irreducible uncertainty term. The MLB in-game model is almost as well-calibrated as the market (reliability differs by only +0.0022) but separates outcomes less sharply (resolution trails by -0.0149). That means re-scaling the probabilities cannot close the gap -- the market simply knows things the model does not.
Caveats and confounds
Soccer behaves differently: its gap is far larger (model Brier 0.332522 vs market 0.195157) and is driven mainly by reliability, which is a calibration problem; its reconstructed Brier (0.366753) also diverges from the direct Brier, a sign the 10-bin decomposition is coarse on the smaller n=4265 soccer corpus.