The data on which a model is trained and which is used to generate predictions is often more important than the model architecture in determining how the model will perform once productionized. Although “Model Cards” exist to document the accuracy, bias, and limitations of various models; ML practitioners lacked a similar artifact to understand the data pipelines underlying an ML-driven system. As such, Tagliabue et al. have released “DAG Cards,” which automatically document an ML pipeline from code comments, introspection, and third-party APIs. DAG Cards include information about pipeline dependencies, parameters, and input files; training and validation loss; behavioral tests; and other flow and run-level data. DAG Cards have been implemented in Metaflow, a popular data science framework.