A new NSF-funded project will create an open-source framework that traces AI errors to their source, helping researchers and clinicians build more trustworthy medical AI systems