Reviews eval scenarios and skills with judgement, covering what a deterministic gate cannot - whether a scenario discriminates, whether a rubric is fair to the unit it tests, whether a recording shim is faithful, and whether a skill's description and body guide an agent. Also grades the criteria an eval run left pending. Use when asked to review a pull request that touches a skill or an eval scenario, to judge an eval run, or to check that a set of evals actually proves what it claims.