The concept of query learning (due to Dana Angluin) is used and developed in several different communities today. The TU Dortmund team originates in the area of program verification. Active learning in this community is used to produce exact and complete models of a system under test. Having in mind this application of learning, we entered the ZULU challenge with the goal to improve in (1) saving membership queries and (2) finding counterexamples, while still producing exact models. In both disciplines we proceeded in two steps: first finding good generic solutions and then customizing these to reflect the specific profile of the ZULU problems. Our main contribution is a highly configurable learning algorithm, which can mimic most of the known algorithms, and a new approach to steering the search for counterexamples using a monotone growing hypothesis annotated with coverage information.