Available data are often inadequate to discriminate among alternative models that make different predictions about the consequences of allowing escapements outside the range of recent historical experience. Dynamic programming is used to show that the optimum policy in such situations can involve active probing or experimentation with escapements. The optimum adaptive policy is usually difficult to compute, but generally may be closely approximated by a "Bayes equivalent" policy that is simpler to estimate but does not account explicitly for the value of information associated with allowing more extreme escapements. While there are various practical difficulties in estimating and implementing an optimum policy, it is concluded that regular probing experiments should be included in every fishery management plan.Key words: stock-recruitment, optimization, adaptive management, stochastic models
