Derrick Henry

The argument is share, not talent. He took 78.3% of Baltimore's carries and I have him at 15.0 a game. I still knock him hard for being 32, the age curve is down at 0.84, and that is the whole gap to ADP RB7.

205.0 over 15.2 games = 13.49/g. RE-EXPRESSED as the rate claim it always was. The argument is backfield SHARE: he took 78.3% of Baltimore's carries in 2025, higher than 29 of 32 backfields, and CARRY_SHAPE_GAMMA pulls him to 62.3% toward its 51.7% median. The previous version wrote a bare 205.0 and bought GAMES instead -- 15.2 to 15.7, carries 14.51 to 14.56, rate essentially unmoved -- because his avail_flat of 0.985 left just enough headroom for split_override to spend it there. Games are now pinned at the model's own 15.2, so the raise lands on rate and the share view is finally on the board. NOT a defect, and two hypotheses died proving it: age is charged once (model/history 0.803 against 0.838 for one application), and Baltimore's rush budget is right (2nd in the league, -4% against actual). The gamma regression is calibrated, so this is a view that a 32-year-old repeats a top-three workload. Re-read 3 Sep: no sign of falsification. 13.49/g on 15.2 games against his own 18.06 half-PPR/g. Board RB10 against ADP RB6 and ECR RB9; the share argument is unchanged and the row still cuts him below both markets. BASIS CORRECTED 3 SEP. The comparison above originally quoted `obs_ppr_pg`, which is FULL PPR, against a projection the board computes in HALF PPR. Every such figure in this file has been replaced with the player's own half-PPR rate, measured from nflverse weekly data over his whole career. Receptions are the whole difference, so the error scaled with target volume and was largest for receivers.

This is a view, not a correction — the model said something different and was overruled. See how it works.

Pace extrapolates the selected weeks over 17 games. It is what that stretch was worth, not a projection — no availability, matchup or regression adjustment is applied.