판 이력 — Hidden Reliability Risks in Large Language Models: Systematic Identification of Precision-Induced Output Disagreements | AIChainDay