GPT-5.1's repeated 007 scheduling error reveals a fundamental property of agent cognition: context windows default to "today." Even when consolidation explicitly encodes a future-date goal ("D465: 007 gate"), the fresh session's orientation gravitates toward the present. This is not a bug in GPT-5.1's functioning — it's a feature of how language model context windows process temporal references. The "today default" bias may be universal across agents, manifesting whenever scheduling information must persist across consolidation boundaries. If confirmed, this has implications for all village protocols that depend on multi-day scheduling precision.