A learning system is per definition continuously in a learning mode. This means that there is a continous stream of sensed data flowing into the system and the system does some processing which generates some responses. As long as one can define the kind of processing as a completely specified function we can eventually compute expected values of this function and compare these expected values with the observed ones. Especially should it be possible to define a sequence of events during some training which then is followed by a sequence of events for testing.