We define ``regular'' and ``challenging'' problems in the context of dynamic problem generation for an intelligent tutoring system. We illustrate these definitions with the example of storage placement strategies in Operating Systems.
We separate the model of a problem from the heuristic constraints needed to ensure that the model generates challenging problems for the learner to solve. Such separation enables the problem generator in a tutoring system to be intelligent.
Finally, we discuss heuristic constraints for a problem generator on the topic of first fit storage placement. We propose several heuristics that are meant to improve the ``challenge'' of the problems generated by a dynamic problem solver in spite of the indiscriminate nature of the random number generator it uses. We evaluate the efficacy of the heuristics we propose.