Document repositories evolve continuously; guidelines and policies are revised, superseded, and re‑uploaded, causing the same content to appear in different wording. Newer versions may refine or contradict earlier ones. These inconsistencies belong to the collection as a whole rather than any single document. Existing approaches treat version linking, deduplication, and contradiction detection as isolated pairwise tasks and stop after labeling a pair.
The GVD framework unifies cross‑document version linking with rule‑level conflict resolution under an auditable update policy. Incoming documents are assigned to version families via bidirectional rule alignment; their rules are then compared against the family memory to detect duplicates, contradictions, asymmetric refinements, and new knowledge. Counterfactual Span Probing (CSP) resolves related pairs that inference mistakenly classifies as neutral.
Relation‑specific policies suppress duplicates and elevate only consequential changes for review, while preserving version lineage as an audit trail. The entire pipeline runs locally without any large language model.
In experiments on 120 enterprise documents processed as 140 ingestions across 59 version families, GVD achieved an F1 of 0.97 for version‑family construction and 0.94 for rule‑level consistency, with CSP raising rule consistency from 0.90 to 0.94.
Review