We need someplace to collect known failures, previous regressions, edge cases that are difficult to construct from generics, and similar. Introduce this here.
