The goal of the project is to improve the procedure that produces the selection region. We provide a training set with signal/background labels and with weights, a test set (without labels and weights), and a formal objective representing an approximation of the median significance (AMS) of the counting test. The objective is a function of the weights of selected events. It is expected that significant improvements are possible by revisiting some of the ad hoc choices in the standard procedure, or by incorporating the objective function or a surrogate into the classifier design.