Intel / article

Universal Pathologies, Conditional Consequences: A Triple-Robustness Analysis of RAG for Multi-Hop Traceability

Source: arxiv.org
Author: Meftun Akarsu, Burak Ozdemir
Date: Published Aug 7, 2026

Context

Finding relevant documents and citing the right evidence are different problems. In this retrieval study, a graph-based system first found similar passages, then followed links to additional material before writing an answer. Across the tested settings it cited roughly 11–15 identifiers per answer, but only 12–23% matched the 's expected evidence.

The consequences depended on the document collection. Extra aerospace-requirement references could introduce conflicting material, while extra Wikipedia paragraphs often remained relevant enough for an automated judge to accept the answer as supported. The result makes citation precision and answer support separate targets. It concerns the tested pipeline and collections, rather than establishing a universal verdict on graph-based retrieval.

Terms in this piece · Glossary

benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.

Key quotes

“GraphRAG underperforms vector RAG on citation precision in many reports, but where and why have remained corpus-bound.”

“Over-citation is architecturally universal: GraphRAG emits 11-15 IDs per answer at citation precision 0.12-0.23 and retrieval recall 0.68-0.87 across all three settings.”

“Its faithfulness consequence is corpus-conditional: in typed-edge DO-178C, GraphRAG faithfulness collapses 74%->40% across hops; on Wikipedia chains the same pipeline rises 42%->58% because over-cited paragraphs remain topically supporting.”

“We argue triple-robustness is the minimum bar for trustworthy RAG architecture claims.”

Read the source arxiv.org