DBRR Increases Context Retrieval via Relational Graph Methods
August 20, 2026
Dual-Bounded Relational Recall (DBRR) improves evidence recovery by splitting a fixed retrieval budget between relevance-ranked seeds and bounded graph-adjacent context. On FullWiki questions, this method increased complete supporting-evidence recovery by 23.8 percentage points over standard top-k retrieval.
HOW THIS AFFECTS YOU
●
builderYou can increase RAG context quality without increasing your total token or retrieval budget.
●
researcherYou can improve retrieval performance by moving beyond flat top-k ranking toward relational context.