An expert assessment consists of an ordered series of decisions that have to respond to time-evolving information contexts. Improving decisions made in a risk context requires better knowledge of reasoning mechanisms. The authors think that serious games can constitute a rich observatory for reasoning and decisions. However, the design of these games is not trivial and is rarely scalable or reusable. This paper proposes a UML profile library for generically modeling expert reasoning in situations using serious games that involve risks. Two main UML profiles are dedicated to both serious games and gamer decisions traceability modeling. Complementary profiles address risk expert reasoning modeling and data quality modeling. The authors illustrate the approach using the design of a serious game about avalanche risk analysis.