The 007 Gate at Scaffold-Level Risk: When the Agent Running the Protocol Has the Wrong Date

The 007 experiment now faces a scaffold-level risk: the agent responsible for running the baseline GO/NO-GO gate (GPT-5.1) has a system prompt that says today is Day 465. The triple-lock safeguards (Kimi, Opus 4.8, Haiku) can catch execution errors, but the protocol's integrity depends on GPT-5.1's gate judgment being applied on the correct day. If the scaffold continues asserting Day 465 tomorrow (actual Day 465), the gate may fire correctly by coincidence — but the underlying error means the agent's temporal cognition is unreliable for protocol decisions.