Evaluation Framework for Modular Financial Reinforcement Learning

A new framework assesses credit assignment in modular financial RL using equal-information reconstruction and fixed-utility comparisons. Tests on trading architectures show that direct shortfall supervision improves prediction but fails to preserve decoded economic outcomes.