Methods, systems, and computer-readable storage media determining a behavior model of a computing system under test. In some implementations, actions include executing, using a user interface of a computing SUT, an initial test script on the SUT; recording, after executing the initial test script, a state of the SUT in the behavior model by observing one or more events that can be triggered using the user interface of the SUT; and iteratively refining the behavior model until an end condition is reached by generating one or more new test scripts, executing the new test scripts on the SUT to test unobserved behavior, and recording one or more new states reached by executing the new test scripts on the SUT in the behavior model.