A useful way to test the testing stack itself is mutation testing: seed one controlled failure in each layer (wrong tool, missed state change, forbidden action) and verify the matching evaluator, trace query, or red-team check fails while valid alternate paths still pass. That catches dead checks and brittle path matching. I work at BOMTOBE, maker of Orke, a desktop workspace for coding agents.