RecourseBench: A Modular Framework for Reproducible Algorithmic Recourse Evaluation

RecourseBench is a newly released modular framework designed for the reproducible evaluation of algorithmic recourse methods, addressing challenges in interoperability and systematic verification. It features a five-layer architecture—Data, Preprocessing, Model, Recourse Method, and Evaluation—along with a four-tier classification system for validating methods against originally reported results via an automated test suite. This framework integrates 28 state-of-the-art recourse methods and offers an interactive web interface, making it a significant tool for practitioners aiming to ensure reproducibility and rigor in evaluating counterfactual explanations in AI systems.