Production, with a baseline.
Compare observed production with each player’s earlier results and the opposing unit’s history. These are chronological conditional-rate benchmarks, not talent grades.
Observed workload is part of the calculation. Expected yards condition on the actual attempts, carries or receptions in the recorded game. This is not a pregame workload or yardage forecast. Historical publication times remain unverified.
Find a comparison
The default view shows FBS programs and category positions: QB for passing, RB for rushing, and WR/TE for receiving. Choose All reported positions to include other roles. Positions are separately captured roster labels; a statistic category does not establish a player’s position. Receiving opportunities are receptions, not targets.
Observed usable-row totals include clean positive-workload records, including rows without a fitted benchmark. Paired actuals, both expectations and residuals use exactly the same games with a known opponent and finite comparison rates. Excluded or zero-workload rows are outside these totals.
The own-prior baseline shrinks a player’s earlier production toward the league baseline. The opponent-adjusted model also accounts for earlier opponents and the current opposing unit; their difference evaluates that full adjustment process.
How the benchmark performs
Experimental replayThis panel evaluates the selected statistic category across the full eligible held-out cohort, including FBS and observed FCS offenses. Player, position, team and sample filters above do not change these validation results.
MSE is squared rate error; MAE is absolute rate error. Weighted errors use observed opportunities. Equal-row MAE gives each player-game the same weight. Lower error is better, and every model uses the same eligible paired rows.
Uncertainty in the difference
The 95% intervals estimate own-prior weighted MSE minus opponent-adjusted weighted MSE, resampling whole games or weeks. A positive difference favors the adjusted model. An interval spanning zero does not clearly favor either model.
Interval coverage, subgroup checks and development diagnostic
Empirical interval coverage counts player-game rows equally. Widths are in yards per observed opportunity. These validation intervals are not individual talent-confidence scores.
Subgroups
Subgroups describe the supplied report; they do not replace the overall promotion gate.
Rolling development diagnostic
Development rows support calibration. These results are not independent final-test evidence and do not activate promotion.