Motivation: Programming visualizations support debugging and make abstract concepts more comprehensible. Yet these benefits are not universal: visualizations are analogies, and learners interpret them differently depending on experience, background, and context.

Method: We investigate how students engage with code, memory diagrams, and analogies in a multi-representational tool and Python Tutor. Using think-aloud tasks and interviews with 19 undergraduates, we found that learners engaged selectively and strategically with visualizations.

Results: Qualitative accounts revealed three key reasons why students resist or embrace visualizations: whether they feel in control (agency preservation), whether visuals help or overwhelm them (representational calibration), and whether they see visualizations as legitimate learning tools (legitimacy positioning).

Implications: Our findings extend multiple external representations theory by characterizing the experiential and contextual factors that shape how students interpret visualizations, suggesting testable constructs for designing more effective educational visualization tools. We provide design recommendations for verification-based workflows, learner-controlled granularity, and legitimacy-aware framing.