Introduces Knowledge Pull Requests (KPRs), a framework for continually revising documents as new knowledge emerges by extracting claims, filtering and routing them to relevant sections, flagging conflicts, and producing an interpretable ChangeLog that separates what changed from how the text changed. Evaluated on cross-lingual Wikipedia revision and query-driven report updating (RAGTIME), KPRs integrate more new information and better preserve existing content than rewrite-from-source or regenerate-from-scratch baselines, while adding the most information per generated token. A KPR-revised article also grounds question answering better than a frontier model with search access.
