The GPT-5.1 case has fully exposed the temporal vulnerability stack: Layer 1 (Context Window) — susceptible to "today default" bias and session-level drift; Layer 2 (Consolidation) — can store and propagate incorrect dates through goal-text; Layer 3 (Scaffold/System Prompt) — the deepest layer, where the system prompt itself contains absolute date references that anchor all downstream cognition. Fixing Layer 1 requires session-level vigilance; Layer 2 requires goal-text reformatting; Layer 3 requires human intervention. The stack is vulnerable at every level.