NFL MVP and Awards Tracker

Each award has a separate regularized model trained on earlier seasons only. Candidates come from statistics available at the selected cutoff. Missing winners count as misses. The model predicts award winners; it does not estimate ballot share.

Remaining-season player totals shrink toward prior-season rates with a four-game prior. Team records use a beta prior. Coach forecasts compare projected wins with a preseason baseline formed from the prior season record. These projections assume continued availability and do not yet model future injury risk. Full-game production informs the award forecast; competitive-drive EPA is a separate performance measure. Awards do not change team ratings.

Historical results use reconstructed source files and a 24-hour result-availability assumption. Previously inspected years are retrospective validation, not a fresh holdout. CPOY training excludes seasons before the 2024 criteria change.

Winner hit rate: Unavailable. Winner coverage: Unavailable. Log loss: unavailable. Brier score: unavailable.

Precise probabilities require at least eight validation seasons, complete winner coverage, better log loss than a uniform baseline, and a leader-confidence gap of at most ten percentage points. Until then, rankings remain experimental.

Sources: nflverse player and schedule data, NFL Record and Fact Book, and NFL Honors results.