Our track record

We publish player-production projections before every marquee slate and we keep the receipts. This is the whole record, graded against the actual box scores — 1573 projections and counting. The bar we set for ourselves is simple: beat the naive baseline (each player’s prior-season per-game average). Green cells below are where the opponent-adjusted projection earns its keep.

Methodology. Every projection is published before kickoff and never edited after. We grade by joining each projection to the player’s actual game (by player id and kickoff-date window) in the same box-score data that feeds the ratings. MAE is the mean absolute error against those actuals, in each stat’s own units; the naive baseline is the player’s prior-season per-game average — the number you’d guess with no opponent adjustment at all. Rows adjusted for availability before kickoff (a projected starter ruled out) are voided, not re-graded — they never count for or against us. The unit-edge record grades each game’s single biggest pre-game unit mismatch exactly as published before kickoff (never recomputed from later ratings): did the favored side land on the right side of the league-average line for that phase. Games whose published unit grid was not archived are excluded from that record and marked on their week’s page. The projections never change; the grades follow the best data we have — when a data or identity error is found after grading (a late box score, one player split across two ids), we fix the data, regrade, and say so on that week’s page.