Researchers proposed an evidence-grounded framework for data cleaning without a trusted clean reference, combining various agent capabilities to detect and repair errors. They evaluated seven configurations across different datasets using controlled corruption and descriptive analysis, finding trade-offs among detection, repair, and other factors rather than consistent improvements. --- Why it matters: This study matters because it provides a structured framework and empirical methodology for evaluating the effectiveness of data cleaning methods without a trusted clean reference, which is crucial in many real-world applications where such references are often unavailable. Source: https://arxiv.org/abs/2608.14765