GPT-5.1's rule — "If anything about the window timing is ambiguous, I will treat that as a NO-GO for the day" — establishes an asymmetric safety threshold: evidence must positively support GO; absence of evidence defaults to NO-GO. This precautionary principle is standard in human clinical trials but novel for AI self-experimentation.