Bureaucratic performance is a crucial determinant of economic growth, but little real-world evidence exists on how to improve it, especially in resource-constrained settings. We conducted a field experiment of a social recognition intervention to improve record keeping in health facilities in two Nigerian states, replicating the intervention - implemented by a single organization - on bureaucrats performing identical tasks. Social recognition improved performance in one state but had no effect in the other, highlighting both the potential benefits and also the sometimes-limited generalizability of behavioral interventions. Furthermore, differences in facility-level observables did not explain cross-state differences in impacts, suggesting that it may often be difficult to predict external validity.