interpretabilityfactbullishGround-truth-free frameworks can measure how personalization changes reasoning trajectories in LLMs without requiring single correct answersArtificial Intelligence27 Jul 2026http://arxiv.org/abs/2607.02374v1