Despite decades of research in the Science of Team Science, a persistent gap remains between how interdisciplinary collaboration is designed and how it is actually enacted. Existing work has largely emphasized structural solutions (e.g., governance models, institutional supports), without specifying the actual interactional mechanisms through which scientists coordinate meaning, integrate knowledge, and sustain collaboration in practice. This study addresses that gap by advancing an experience-based account of interdisciplinary teamwork grounded in scientists’ own prioritizations of interaction formats and behaviors. Using a two-part survey of researchers with extensive experience in large-scale, multi-institution collaborations (e.g., AI, robotics, human–machine teaming), we combine structured ranking tasks with qualitative accounts of collaboration practices. This design enables us to move beyond retrospective narratives to identify what scientists actually prioritize when forced to make tradeoffs among competing interaction formats and team behaviors. Across analyses, a striking pattern emerges: scientists privilege interactional infrastructure over formal structure. Regular, cross-team working meetings are consistently ranked as the most critical mechanism for successful collaboration, outperforming leadership meetings, governance processes, and formal training interventions. Similarly, the highest-valued behaviors are not strategic or procedural, but fundamentally communicative. Scientists emphasize the importance of careful listening, adapting explanations across disciplinary boundaries, and making expertise accessible to others. In contrast, canonical “team process” behaviors emphasized in the literature (e.g., explicit reflection on team functioning) are seen as less important. At the same time, nearly all respondents report persistent coordination breakdowns rooted not in technical complexity, but in epistemic misalignment; that is, incompatible standards of evidence, and unarticulated disciplinary assumptions. Crucially, the work required to resolve these tensions (e.g., building shared vocabulary, aligning conceptual models) is described as essential yet systematically under-resourced. We argue that these findings surface a critical misalignment in the field: while SciTS has emphasized designed coordination (structures, roles, processes), scientists themselves emphasize emergent coordination enacted through repeated, high-quality interaction. This shifts the locus of explanation from formal mechanisms to what can be conceptualized as in situ knowledge integration practices. Scientists prioritize and value the moment-to-moment communicative work through which shared understanding is constructed. These findings contribute to the Science of Team Science in three ways. First, they provide empirical grounding for a shift from structural to interactional accounts of interdisciplinary collaboration. Second, they highlight the central role of communicative competencies as core teamwork processes, suggesting a reconceptualization of “coordination” as an ongoing, distributed cognitive activity rather than a formally managed process. Third, they identify integrative roles and recurring interaction forums as key leverage points for intervention. More broadly, this work reframes interdisciplinary teamwork as a problem of epistemic alignment under constraint, where success depends less on formal design and more on the capacity of teams to continuously negotiate meaning in the flow of work. Future research should build on this perspective by linking interactional patterns to performance outcomes and experimentally testing interventions that scaffold shared understanding, translation, and integration in real-world scientific teams.