I propose a general framework to analyze the socially optimal response to criminal and civil wrongs. Each wrongful act, detected with some probability, generates a random piece of evidence and a consequent posterior probability of guilt for each individual. I show that with no assumptions on individual behavior, if the government prefers that, all else equal, innocent individuals face less punishment, incapacitation, rehabilitation, and reparations, then for each individual and each response the optimal policy must be non-decreasing in that individual’s posterior probability of guilt. This implies that the optimal response for each individual depends solely on the quality of evidence against them, and not on their rank in the suspect list. I argue that distinguishing only between “innocent” and “guilty” significantly limits the scope of a multifaceted justice system, and may bias the system towards retributive rather than rehabilitative intuitions.