A court AI pipeline has no lineage from source docket records to model input features. What should the auditor recommend?

A court AI pipeline with no lineage from source docket records to model input features impairs auditability, incident investigation, and accountability for justice-system AI. The auditor should recommend establishing end-to-end data lineage from source dockets to AI input features so each feature can be traced to originating records, transformations, and owners when outcomes are challenged or errors appear. Maintaining version history for feature transformation code in the shared model repository fails by concept: code version history does not by itself trace each input feature back to its originating docket record or capture the end-to-end lineage from source to feature. Reconciling AI input features against monthly case count summaries fails by concept: summary-level case counts do not establish the traceable link from individual docket records to specific model input features, and consistency with aggregates does not prove feature-level provenance. Documenting lineage from model outputs to judicial reports fails by concept: tracing outputs downstream to reports does not establish the upstream link from source dockets to model input features, which is the missing provenance the audit requires. Exam caveat: AAIA AI data-management items emphasize end-to-end lineage for AI inputs in municipal justice systems—not code version history alone, monthly aggregate reconciliations, or output-to-report tracing that bypasses the source-to-feature link. Operational check: implement lineage capture from docket extract through transforms to feature store, retain lineage with model versions, test reconstruction for sample features, and include lineage evidence in court AI assurance reviews.