Every verdict RubricMiddleware can emit for a completed grading run.
Derived from the SDK's RubricResult Literal so it cannot drift out of sync
with the grader vocabulary: if the SDK renames or adds a verdict, this set
follows automatically. Consumers that branch on a rubric result (goal
auto-completion in app.py, the rubric-event formatters in textual_adapter)
treat any value outside this set as an unrecognized grade rather than silently
mishandling it.
