Abstract The author studies the evolution of the number of coexisting beliefs in a financial market. Crucially, he undertakes to do so in a framework where the paradigms, beliefs, and models driving agents behavior are left totally unspecified; i.e., the author does not make any parametric or non-parametric model assumptions. The overreaching aim of this exercise is to characterise the dynamic of the variety of beliefs in an auction-based financial market independently of any assumptions on agents behaviors. The resulting framework may be seen as an abstract agent-based model. In a computer experiment the authors exhibits a cycle between two states, so that either all agents act according to the same belief, or there is no leading belief; i.e., there is one dominating belief, or none. Further, the author finds that the frequency of this cycle is positively linked to the quality of the information available to the agents.