Update Costs of RAG Chunking Strategies Under Document Revisions

Evaluating Wikipedia revisions shows that recursive chunking (38-56% recomputation) and content-defined chunking (43-54%) are more efficient during document updates than fixed-size chunking (57-68%). Recall remains stable across these methods under a 2048-token retrieval budget.