RESEARCH · RESEARCH · #537
GVD: unified framework for governed versioning and deduplication (arXiv:2609.17696v1)
The arXiv preprint introduces GVD (Governed Versioning and Deduplication), a locally runnable pipeline that unifies cross-document version-family linking with rule-level duplicate, contradiction and refinement detection under auditable update policies. On an enterprise corpus of 120 documents (140 ingestions across 59 version families) the authors report F1 = 0.97 for version-family construction and F1 = 0.94 for rule-level consistency, with Counterfactual Span Probing (CSP) improving rule consistency from 0.90 to 0.94.
KEY POINTS
- The arXiv preprint introduces GVD (Governed Versioning and Deduplication), a locally runnable pipeline that unifies cross-document version-family linking with rule-level duplicate, contradiction and refinement detection under auditable update policies.
- On an enterprise corpus of 120 documents (140 ingestions across 59 version families) the authors report F1 = 0.97 for version-family construction and F1 = 0.94 for rule-level consistency, with Counterfactual Span Probing (CSP) improving rule consistency from 0.90 to 0.94.
- GVD matters because it provides an auditable, policy-driven way to reduce redundant reviews and surface only consequential changes across evolving enterprise documents without relying on external LLMs.
WHY IT MATTERS
GVD matters because it provides an auditable, policy-driven way to reduce redundant reviews and surface only consequential changes across evolving enterprise documents without relying on external LLMs.