Each test spawns in-process aggregators watching the same cluster-wide CRs and shared desired-state bucket; run in parallel, one test's aggregator deletes the orphan desired-state another test is asserting still exists. A suite-wide lock keeps the scenarios isolated.