We study optimal policy when the planner has partial information in a general setup where observed signals are endogenous to policy. In this context, signal extraction and policy have to be determined jointly. We derive a general non-standard first order condition of optimality from first...