To what extent do the results of randomized controlled trials inform our predictions about the effectiveness of potential policy interventions? This crucial question is often overlooked in discussions about evidence-based policy. The view I defend is that the arguments that lead from the claim that a program works somewhere to a prediction about the effectiveness of this program as it will be implemented here rests on many premises, most of which cannot be justified by the results of randomized controlled trials. Randomized controlled trials only provide indirect evidence for effectiveness, and we need much more than just randomized- controlled-trial results to make reliable predictions.