Dynamically choosing which test tasks to evaluate based on how informative they are for distinguishing between candidate solutions.