In order to provide answers to theoretical and methodological challenges in the study of English-Medium Instruction, researchers increasingly borrow relevant frameworks and practical approaches from outside the familiar field of (socio)linguistics. In this context, Lipsky’s notion of Street-Level Bureaucracy (originally designed for the study of public service work) can be used as as novel approach to studying the multilingual practice dimensions of EMI classroom interaction. Existing work has demonstrated how Lipsky’s framework (1980) allows us to consider on-the-ground language policy construction by EMI content lecturers, describing the lecturer as a lens through which institutional language policy is reflected or – at times – refracted in situated classroom practice (De Soete & Slembrouck, 2023). In this presentation, I investigate how negotiation and discretionary transformation of language policy are brought about interactionally in the EMI classroom, leading to different degrees and forms of monolingual/multilingual conduct. The dataset consists of 23 classroom recordings, which capture the interactions between six engineering lecturers (3 bio-science engineers and 3 industrial design engineers) and their students in two EMI programs at a Belgian university. The interactional data are examined through a combination of Lipsky’s model and Goffmanian frame analysis (1974). This interdisciplinary approach brings into focus the distribution of translanguaging practices across lecturers, disciplinary contexts, and didactic formats. Specifically, I focus on who initiates translanguaging, when and how this happens, which didactic purpose it appears to serve, and whether it is negotiated through meta-pragmatic commentary. The analysis reveals how conceptualizations of didactic roles, considerations of disciplinary literacies, and micro-level pragmatic concerns converge to inform the lecturers’ stance vis-à-vis the institutional language policy dictates, occasionally leading to negotiation and discretionary transformation of that policy. As such, the findings suggest the viability and sociolinguistic relevance of an enactment-of-policy-in-interaction perspective which is equally a matter of disciplinary induction.