Knowledge Pull Requests for Continual Document Authoring
The paper proposes “Knowledge Pull Requests” as an auditable way to update documents without rewriting them from scratch.
KPRs extract new claims, route them to the right sections, flag conflicts, and produce a ChangeLog separating the proposed knowledge change from the resulting text diff. The authors test the method on cross-language Wikipedia revisions and RAGTIME query-driven reports. They report that KPRs add more information while preserving more existing content than source rewriting or full regeneration. HF Daily Papers' note
KPRs extract new claims, route them to the right sections, flag conflicts, and produce a ChangeLog separating the proposed knowledge change from the resulting text diff. The authors test the method on cross-language Wikipedia revisions and RAGTIME query-driven reports. They report that KPRs add more information while preserving more existing content than source rewriting or full regeneration. HF Daily Papers' note
score 5