Tag
This paper investigates using governance records from machine-verifiable workflows as supervision for training language models to perform structured workflow repair, demonstrating that verifier-selected self-training enhances execution efficiency and validity.