This dissertation analyzes the implementation of performance programs in local bureaucracies by investigating the constitutive organizational routines of performance programs. Calibrated with CompStat (short for Computer Statistics), one of the most celebrated and contested police performance programs of the last three decades, I adapt the Carnegie Mellon tradition in organizational studies to analyze programs as sets of interrelated routines, which can be adopted in toto or adapted when implemented in new organizational settings. This organizational understanding of performance programs, in turn, enables me to explore the ways in which (i) certain NYPD “internal” routines are strongly interconnected and hence able to diffuse as a “performance package” across numerous police departments (paper 1); (ii) a subset of those routines are used strategically by supervisors and subordinates in a way which may give rise to second-order effects that are self-defeating (paper 2); and (iii) other NYPD routines, in this case “stops” of individuals on city streets, are patterned in a way which enforces residential racial segregation (paper 3). Mixing document analysis, interviews, direct observation, surveys, agent-based models, and random effects models, the dissertation contributes to our understanding of policy reinvention and evolutionary public policy, the distorting effects and limitations of public management, and the nature of interdependence between routines within the organization, on the one hand, and outside the organization, i.e., street-level bureaucrats’ interactions with citizens, on the other hand.