A passing test suite is weak evidence unless the detector is proven to fail and the defect can be reproduced.
Solution
Built seeded and fixed versions of a synthetic workspace-invite flow, then ran the same 12 assertions to isolate a role-reset state defect after inline email validation.
Built a 12-check QA evidence pack that caught one seeded state defect, then passed 12/12 against the corrected reference build with reproducible artifacts.