Exams have traditionally been given to evaluate students but in recent years, with the appearance of freely accessible on-line tutoring, quizzes can also be used as a learning tool. In systems where students can request quiz items until a satisfactory grade is obtained, new probabilistic approaches are required for dealing items to students and new statistical approaches are required for analysis, including student evaluation. This paper applies various models to a case study to quantify the actual learning involved in using such a dynamic system and demonstrates the need for new statistical approaches. Responses to repeatedly assigned questions, some with random numbers, are analyzed to distinguish rote learning from learning with understanding. It is seen that models designed for evaluation are inadequate for dynamic systems which are a part of a learning experience in contrast to static systems which merely provide data storage and snapshot evaluation. (Contains 7 tables.)