> How can you certify something which doesn't behave the same twice, and more importantly we don't understand how it works 100%? By verifying that all of its p

> How can you certify something which doesn't behave the same twice, and more importantly we don't understand how it works 100%? By verifying that all of its possible behaviors conform with the "it works" spec, regardless of which of those behaviors it chooses. Monitoring with a known-safe fallback is the easiest case.