Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-04 · high
The gap gets worse as the generator improves, and the fix is conditioning checks on the sub-question, not the raw one.
Why now: The paper posted to arXiv in August 2026.
Entailment scoring of individual retrieved chunks backfires on multi-hop questions, a new study finds.
Multi-hop questions need separate facts chained together, and that's exactly where retrieval-augmented generation systems are expanding. The per-chunk verification step meant to catch bad context instead makes accuracy worse than not filtering at all. Entailment scoring reaches just 0.643, 0.523, and 0.560 AUC on HotpotQA, 2Wiki, and MuSiQue, against 0.951 on single-hop SQuAD, per the paper.
The method checks whether a single retrieved chunk supports the answer on its own. That assumption holds for single-hop questions. It breaks on multi-hop ones, which are built so no one chunk is a sufficient premise.
A checker scoring chunks in isolation can't tell a real hop from a plausible dead end. Dropping chunks that fail the test throws out real hops along with the dead ones. Per-chunk gating scored worse than no filtering in every setup the researchers tested.
Conditioning verification on the decomposed sub-question, instead of the original question, lifts later-hop entailment from 0.546 to 0.840, per the study. Iterative retrieval systems already break questions into sub-questions as part of retrieval, then discard that decomposition before verification ever runs.
Each link below shares sources, entities, or timing with this story.
headroom benchmarked against Squad / Shared entities / Earlier coverage
Linked by a graph relationship (headroom benchmarked against Squad); both cover RAG, SQuAD; earlier RAG coverage from 2026-07-28.
Shared entities / Same source domain / Earlier coverage / Tension
Both cover HotpotQA, RAG; reported by the same outlet (arxiv.org); earlier HotpotQA coverage from 2026-07-26.
Shared entity: RAG / Same source domain / Earlier coverage / Tension / Downstream implication
Both cover RAG; reported by the same outlet (arxiv.org); earlier RAG coverage from 2026-07-30.
headroom benchmarked against Squad / Shared entity: RAG / Earlier coverage
Linked by a graph relationship (headroom benchmarked against Squad); both cover RAG; earlier RAG coverage from 2026-07-26.
Linked by a graph relationship (headroom benchmarked against Squad); both cover RAG; earlier RAG coverage from 2026-07-08.
Linked by a graph relationship (headroom benchmarked against Squad); both cover RAG; earlier RAG coverage from 2026-06-18.
Linked by a graph relationship (headroom benchmarked against Squad); both cover RAG; earlier RAG coverage from 2026-06-02.
Shared entity: RAG / Same source domain / Earlier coverage / Tension
Both cover RAG; reported by the same outlet (arxiv.org); earlier RAG coverage from 2026-07-21.