Platform-enabled research services will control, manage, and measure learner experiences within that platform. In this paper, we consider the need for research services that examine learner experiences "outside" the platform. For example, we describe an effort to conduct an experiment on peer assessment in a college writing course, where Terracotta (a research service within the learning management system) randomly assigned students to either (1) write peer assessments themselves, (2) use a generative AI tool to provide the feedback, or (3) build upon and improve the feedback provided by generative AI. This research effort was not successful, and it stands as an example of the limitations of current platform-enabled research services, and the need for infrastructure to support research beyond existing platforms.