To ensure multi-agent based simulation models reproducibility, particular attention must be payed on its possible implementation ambiguities. This concerns every aspect of simulation's architecture, including how the agent selects the actions it performs, and on which agents it is performed. On this point, classical agent-centered design methodologies leave room to implicit design choices. We show in this paper how an interaction-centered design methodology provides guidelines to elicit these choices, and to remove possible implementation ambiguities underlying agent design. To illustrate this issue, we study which ambiguities underly even simple models and how our interaction-based methodology makes them appear.