Framework for Evaluating Sequential Decision Making via MDP Projection

This framework characterizes AI reasoning by projecting methods onto the Markov decision process formalism and assessing autonomy and computational cost. Testing across Neurosolver, FBRL, and AutoToS reveals how graph-based, reinforcement learning, and LLM-based approaches diverge in design and performance.