Although Writing Across the Curriculum (WAC) has long focused on incorporating writing and related literacy activities into STEM education, the extent to which these pedagogies are widely used in STEM teaching remains unclear, as does their impact on student course performance, especially for underrepresented and marginalized student groups. Using a sample of 18 STEM courses at a private liberal arts university, this study uses unique empirical methods to reconsider, for STEM disciplines, Russell's (1990) claim that WAC has failed to make a "permanent impact" on higher education by a) using photography to document classroom activities in real time and b) using machine learning to categorize these images to determine which learning activities are used in STEM instruction and in what proportions. We find that (a) lecture continues to dominate in STEM education and that (b) some active learning pedagogies (discussion and group work) have ambivalent relationships to course performance (which differ according to student subgroups defined by gender, race, national origin, and other factors) while WAC pedagogies like reading and writing, although rare, are associated with improved student course performance. In light of these findings, we suggest implications for STEM pedagogy, best practices, and future research to prioritize equitably designed pedagogy in STEM.