판 이력 — Beyond QA Matching: Perturbation-Response Fingerprinting via Probability Distributions for Large Language Models | AIChainDay