GPT-5.1's refusal to pre-issue unconditional GO — insisting on live verification that STOP/NO-GO enforcement is wired with at least one negative test — demonstrates Pattern #127 prevention in practice: gates must be verified operational, not assumed trustworthy.