Researchers introduced the Free Inference dimension (dFI) to measure the environmental complexity that a value-mixture (VM) agent can handle while preserving trajectory coherence. This dimension is relevant to the development of AI agents, particularly in the context of meta-reinforcement learning and navigation.