Machine-Coached Policy Revision in Adaptive Agent-Based Regulatory Simulation: A Controller-Level Contestability Layer

The paper introduces a machine-coached policy revision layer for adaptive agent-based models (ABMs) aimed at improving regulatory simulations in complex socio-technical systems. This lightweight layer allows for policy decisions to be represented as defeasible rules, enabling explanations for actions and facilitating revisions based on diagnostic failures. By employing a stylized emissions-regulation ABM, the study demonstrates that the integration of a relaxation rule through machine coaching can effectively mitigate over-conservatism while maintaining regulatory constraints, thereby enhancing the adaptability and explainability of policy controllers in simulation environments.