Abstract
Reproducible data audits often fail because supporting evidence is scattered across notebooks, tickets, spreadsheets and narrative review notes. This paper introduces a federated evidence graph model that links artifacts without forcing teams into a single repository. The model uses lightweight nodes for datasets, transformations, tests and review findings, allowing reviewers to trace claims across independently maintained systems.
Contribution Summary
The work defines a graph schema, a synchronization strategy and an audit view that can be adopted incrementally by distributed data teams.
Citation
Kim, D. H., and Nair, P. Federated Evidence Graphs for Reproducible Data Audits. In Proceedings of the DECaT 2024 Workshop, article DECaT-2024-02, 2024.