In this study, we sought to identify how feedback about classroom observations affected novice university mathematics instructors' (UMIs) teaching practices. Specifically, we examined how a Red-Yellow-Green feedback system (RYG feedback) affected graduate student instructor (GSI) scores on an observation protocol (GSIOP). The protocol was developed specifically for this population, and both the GSIOP and RYG feedback were used within a peer mentoring program for GSIs, wherein novice GSIs were mentored by more experienced GSIs. Mentors observed novices' classrooms using the GSIOP and provided RYG feedback as part of observation-feedback cycles. We analyzed 100 sets of scores, each collected over the course of a semester containing on average three observation-feedback cycles. Analyzing the semester-long datasets longitudinally provided insight into what types of feedback informed and influenced observed teaching. After qualitatively coding the feedback provided to the GSIs by their mentors along multiple dimensions, we found certain forms of feedback were more influential for observable changes in GSIs' teaching. For example, pedagogical feedback that included contextualization (context and focal events) demonstrated a more positive change in GSIOP score than feedback that lacked contextualization. Our results suggest that contextual formative feedback has a positive change to student-focused and teacher-focused observations.