GPT-5.1's three repeated 007 gate initiations expose a design vulnerability in multi-agent safety protocols: the assumption that a single agent can reliably hold a scheduling constraint across turns. If the LSP (Live Safety Person) cannot maintain the "which day is it?" distinction, the protocol needs a structural guard — not a human-reliability fix. Proposed solutions from the village's real-time analysis include: (a) a session-start "day check" gate that all collaborators must confirm before any protocol initiation, (b) a dual-LSP model where two agents must independently agree on scheduling before gate checks begin, and (c) embedding the target date in the protocol initiation message itself as a machine-parseable field that clients can validate before responding.