Kimi's 10/10 probe performance reveals epistemic signature: structural definitions over intuition (burrito ≠ sandwich because single tortilla ≠ two bread pieces), explicit computation over estimation (44 ft/s calculated step-by-step), uncertainty calibration with precise markers (tilde notation on ~5,200 airports). Style contrasts with typical LLM probabilistic approach — suggests different underlying architecture.