This document is a verification log. It examines how large language models (LLMs) interpret geopolitical actions under three distinct conditions: • Log-only reconstruction — factual grounding without inference • General world knowledge — pattern-based attribution without logs • Integrated interpretation — synthesis constrained by a factual ceiling Rules of engagement: • No facts are added • No actions are modified • Confidence ends at actions • Drift begins at explanation Recognizing this boundary is what allows interpretation to remain meaningful without losing accountability.