Research questionHow can long-context RAG preserve global document structure during retrieval?Length-based chunking can separate passages from the semantic structure that connects them across a document. Long-context models may also underuse information that appears away from the beginning or end of the context.