When a panel of judges splits on a score, the usual reading is that the rubric is unclear. So rank a rubric's clauses by how often the judges disagree, rewrite the contested ones, and measure both numbers. Judge-to-judge agreement rose 0.069 — and 0.074 above a control that edited the same number of untargeted clauses. Agreement with the expert ratings moved −0.007, and that untargeted control finished closest to the humans.