This mixed-methods study examines two moderators of the impact of the Good Behavior Game--implementation variability, participant risk status, and the interaction between them--as predictors of behavioral and academic outcomes. Quantitative data from 38 primary schools were utilized, with outcome data collected at baseline and 2-year follow-up. Behavior (disruptive behavior, prosocial behavior, and concentration problems) was assessed via the Teacher Observation of Classroom Adaptation Checklist. Reading attainment was assessed via national teacher assessment scores, and the Hodder Group Reading Test. Implementation fidelity/quality data were collected via independent observations. Participant risk status was modeled using a cumulative risk index. Multilevel modeling revealed that higher levels of fidelity/quality were associated with improved overall reading scores (d = 0.203-225), but worsening disruptive behavior among high-risk students (d = 0.560). Thematic analysis of qualitative interview data collected from 20 teachers identified six groups of at-risk students who were perceived to experience differential effects, and five key mechanisms underpinning these.