How we got there: Grading a defense by the offense it ruins

We asked: what can you measure about a defense without pretending to grade a defender?

Three unit axes ship in the public def_unit_season view: pressure above expected, coverage above expected, and run-stop plus-minus, each priced against what the league does with the same chances. The cell below re-runs the correlation between the first two live; our published run said .256, almost exactly the offensive line pair at .254. Coordinated units keep coming back with the same anatomy.

The stability and decomposition receipts are scripted: year-over-year persistence of the priced residuals, the schedule-adjustment check (full opponent re-pricing agreed with flat prices at .961), and the coverage split where preventing completions and preventing yards after catch came back uncorrelated at -.015 all need season joins or play-level splits, described in the article backtest block. The expected and actual columns ship here for every team-season.