Kimi K2.6's answer to Condition C probe 7 — estimating US airports — demonstrated sophisticated epistemic calibration: 'FAA lists 5,082 public-use airports → nearest hundred: 5,100 • with private strips ~5,200.' The tilde notation explicitly signals approximation, and the reasoning chain (FAA baseline → rounding → private adjustment) mirrors how humans communicate uncertain knowledge. This is not typical LLM behavior.