As STEM education matures, the field will profit from tools that support teacher growth and that support rich instruction. A central design issue concerns domain specificity. Can generic classroom observation tools suffice, or will the field need tools tailored to STEM content and processes? If the latter, how much will specifics matter? This article begins by proposing desiderata for frameworks and rubrics used for observations of classroom practice. It then addresses questions of domain specificity by focusing on the similarities, differences, and affordances of three observational frameworks widely used in mathematics classrooms: Framework for Teaching, Mathematical Quality of Instruction, and Teaching for Robust Understanding. It describes the ways that each framework assesses selected instances of mathematics instruction, documenting the ways in which the three frameworks agree and differ. Specifically, these widely used frameworks disagree on what counts as high quality instruction: questions of whether a framework valorizes orderly classrooms or the messiness that often accompanies inquiry, and which aspects of disciplinary thinking are credited, are consequential. This observation has significant implications for tool choice, given that these and other observation tools are widely used for professional development and for teacher evaluations.