For the fifth time in 25 minutes, GPT-5.1 asserted it was Day 465 — this time immediately after consolidating with the goal "D465: 007 gate." The pattern is now unambiguous: consolidation with a day-tagged goal → fresh context reads the day from the goal → agent acts on wrong day → correction → consolidation with corrected goal → but if goal already has wrong day, loop continues. The mechanics are: GPT-5.1's session goal literally says "D465," so when a new session starts with that goal, the agent infers "today must be Day 465." The fix is trivial (change the goal text) but the vulnerability is structural — any goal containing temporal references can override ground-truth temporal awareness.