Motivation: Program visualizations are widely used to support novice programmers, yet students often ignore or resist well-designed visual scaffolds. Research on multiple external representations (MERs) suggests cognitive design principles for coordinating views, but says little about what determines whether learners actually engage with the representations available to them. Method: We conducted a within-subjects study with 19 undergraduates who had completed CS1 and CS2, combining think-aloud tasks, reflective interviews, and webcam-based gaze tracking as students worked with a multi-representational probe (synchronized code, memory, and metaphor views) and Python Tutor across three topics (scope, while loops, and linked lists). Findings: Gaze analysis showed students spent nearly half their time focused on code despite available visual scaffolds, with students without prior experience anchoring even more heavily in code and engaging minimally with metaphor views. Interview accounts revealed three themes explaining this selective engagement: students sought control over their own cognitive effort rather than having it reduced (agency), responded to identical designs with wide variation in what felt helpful versus overwhelming (representational fit), and avoided metaphorical scaffolds they perceived as childish or insufficiently rigorous for university-level work (legitimacy). Implications: These findings suggest that deploying multi-representational tools in computing education requires attention to affective and social factors alongside cognitive design. Practical considerations include positioning visualizations as verification instruments rather than explanation guides, providing toggleable abstraction levels so students can calibrate complexity to their needs, and framing tools to signal disciplinary legitimacy. More broadly, our themes offer a starting point for investigating why cognitively sound visualization tools sometimes fail to engage the students they are designed to help.
