This article addresses issues of consistency verication of learner proles in adaptive serious games. More precisely, our research objective is to propose models and tools that allow the user (learner, teacher or expert, depending on the context of application) to create coherent proles consistent with domain knowledge. Our approach has been conceived and developed in the context of the platform GOALS. GOALS, as Generator Of Adaptive Learning Scenarios, is an online platform which allows the generation of learning scenarios, keeping into account the educational and entertaining aspects of serious games. For this, the knowledge in GOALS is organized into three layers: the domain concepts, the pedagogical resources, and the game resources. The prole is represented by a set of couples in the form , where attribute corresponds to a concept, and value represents the learner competence in that concept. The prole is initialized by the user. During the game session, the prole is updated automatically according to dependencies among dierent domain concepts. In order to verify the learner proles validity, we use a rule-based system which veries, for every type of relation between concepts, the values between the source and the target concept. In this article, we present the formalization of our approach, as well as, its evaluation.