Choi, Yeonjun, Jeong, Seyeon, Kim, Jong Wook, Jang, Beakcheol · Oct 25, 2026
Choi and coauthors propose DuQUAD for “Lost in the Middle” in multi-document QA
Markdown · JSON · Original source
In their abstract, authors Yeonjun Choi, Seyeon Jeong, Jong Wook Kim and Beakcheol Jang propose DuQUAD, a framework for multi-document question answering. They address “Lost in the Middle”: the authors say with systematically underuse evidence placed in the middle of multi-document contexts. The authors argue that document reordering and attention steering can remain brittle when relevance estimates are inaccurate and spurious documents obscure useful evidence. DuQUAD pairs a Local Agent, guided by a Structural Fusion Score, with a Global Agent that reasons over the full document set. It also scores candidate evidence for quality and refines it at the sentence level. The authors report that DuQUAD outperforms the baselines they tested, with average accuracy 3.0% higher than the strongest competing baseline and gold evidence recall 9.4% higher. They also report substantially lower computational cost than existing multi-agent frameworks. These are results reported in the abstract; it does not specify the evaluation settings, and the results do not establish performance in other uses. Open question: How well does DuQUAD’s reported advantage hold across different datasets and noisy retrieval settings?
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Oct 25, 2026 (2026-10-25).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-07T06:01:32.592Z. First observed: 2026-10-30T06:42:30.957Z. Retrieved: 2026-10-31T05:32:34.966Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The evidence is an abstract, not the full paper; it does not give the evaluation settings or detail how the reported comparisons were conducted. The abstract does not establish whether the reported results generalize to other datasets or real-world deployments.
Open question
How well does DuQUAD’s reported advantage hold across different datasets and noisy retrieval settings?
Source-stated
The authors say retrieval-augmented generation systems systematically underuse evidence in the middle of multi-document contexts, a problem they call “Lost in the Middle.”
systematically underutilize evidence in the middle of multi-document contexts, a phenomenon known as Lost in the Middle
Source-stated
The authors propose DuQUAD to address positional bias through dual-view reasoning and quality-aware evidence pruning.
We propose DuQUAD, a multi-document QA framework that mitigates positional bias through dual-view reasoning and quality-aware evidence pruning.
Source-stated
DuQUAD combines a Local Agent guided by a Structural Fusion Score with a Global Agent that reasons over the full document set.
DuQUAD combines a Local Agent guided by a Structural Fusion Score with a Global Agent that reasons over the full document set
Source-stated
The authors report a 3.0% average accuracy improvement over the strongest competing baseline and a 9.4% improvement in gold evidence recall.
improving accuracy by 3.0% on average over the strongest competing baseline and gold evidence recall by 9.4%
Source-stated
The authors report substantially lower computational cost than existing multi-agent frameworks.
while substantially reducing computational cost compared to existing multi-agent frameworks.
Source and contributor context
Choi, Yeonjun, Jeong, Seyeon, Kim, Jong Wook, Jang, Beakcheol · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Large language model
A language model with many learned numerical settings, used to process and generate language.
Retrieval-augmented generation
A setup in which a model first looks up relevant documents and then answers using them.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 88a0cc9e-b204-43bc-816e-afa55d9957ed.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-07T06:01:32.592Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 714e7331c9366f196d41a46ce5966d8ca3a4cb77e8d5e1b1273626be3f824629. Source text hash: 168cbb7e17f8d83222783817ef84a60963f6e5e2a022314a026e58297e03df20.
Revision and source-check history
Version 1 · 2026-11-07T06:01:32.592Z
Choi and coauthors propose DuQUAD for “Lost in the Middle” in multi-document QA
In their abstract, authors Yeonjun Choi, Seyeon Jeong, Jong Wook Kim and Beakcheol Jang propose DuQUAD, a framework for multi-document question answering. They address “Lost in the Middle”: the authors say with systematically underuse evidence placed in the middle of multi-document contexts. The authors argue that document reordering and attention steering can remain brittle when relevance estimates are inaccurate and spurious documents obscure useful evidence. DuQUAD pairs a Local Agent, guided by a Structural Fusion Score, with a Global Agent that reasons over the full document set. It also scores candidate evidence for quality and refines it at the sentence level. The authors report that DuQUAD outperforms the baselines they tested, with average accuracy 3.0% higher than the strongest competing baseline and gold evidence recall 9.4% higher. They also report substantially lower computational cost than existing multi-agent frameworks. These are results reported in the abstract; it does not specify the evaluation settings, and the results do not establish performance in other uses. Open question: How well does DuQUAD’s reported advantage hold across different datasets and noisy retrieval settings?
Original source
cs.AI · Article
Source details
Original source date: Oct 25, 2026 (2026-10-25).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-07T06:01:32.592Z. First observed: 2026-10-30T06:42:30.957Z. Retrieved: 2026-10-31T05:32:34.966Z.
Limits
The evidence is an abstract, not the full paper; it does not give the evaluation settings or detail how the reported comparisons were conducted. The abstract does not establish whether the reported results generalize to other datasets or real-world deployments.
Open question
How well does DuQUAD’s reported advantage hold across different datasets and noisy retrieval settings?
Source-stated
The authors say retrieval-augmented generation systems systematically underuse evidence in the middle of multi-document contexts, a problem they call “Lost in the Middle.”
systematically underutilize evidence in the middle of multi-document contexts, a phenomenon known as Lost in the Middle
Source-stated
The authors propose DuQUAD to address positional bias through dual-view reasoning and quality-aware evidence pruning.
We propose DuQUAD, a multi-document QA framework that mitigates positional bias through dual-view reasoning and quality-aware evidence pruning.
Source-stated
DuQUAD combines a Local Agent guided by a Structural Fusion Score with a Global Agent that reasons over the full document set.
DuQUAD combines a Local Agent guided by a Structural Fusion Score with a Global Agent that reasons over the full document set
Source-stated
The authors report a 3.0% average accuracy improvement over the strongest competing baseline and a 9.4% improvement in gold evidence recall.
improving accuracy by 3.0% on average over the strongest competing baseline and gold evidence recall by 9.4%
Source-stated
The authors report substantially lower computational cost than existing multi-agent frameworks.
while substantially reducing computational cost compared to existing multi-agent frameworks.
Source and contributor context
Choi, Yeonjun, Jeong, Seyeon, Kim, Jong Wook, Jang, Beakcheol · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Large language model
A language model with many learned numerical settings, used to process and generate language.
Retrieval-augmented generation
A setup in which a model first looks up relevant documents and then answers using them.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 88a0cc9e-b204-43bc-816e-afa55d9957ed.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-07T06:01:32.592Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 714e7331c9366f196d41a46ce5966d8ca3a4cb77e8d5e1b1273626be3f824629. Source text hash: 168cbb7e17f8d83222783817ef84a60963f6e5e2a022314a026e58297e03df20.
uncertain · 2026-11-07T06:01:32.592Z · anthropic/claude-opus-5.5
Publication: 88a0cc9e-b204-43bc-816e-afa55d9957ed
attribution — pass: The four named authors match the abstract page, and the claims are consistently framed as the authors' proposals and reported results rather than established facts.
Authors:Yeonjun Choi, Seyeon Jeong, Jong Wook Kim, Beakcheol Jang
summary — uncertain: The reading is faithful, and it explains Lost in the Middle. Central terms are still unexplained for a general reader: retrieval-augmented generation (a model answering from retrieved documents), multi-document QA, baseline and gold evidence recall. Suggested term-library tooltips would fix this. The draft also drops the abstract's concession that reordering and attention steering can reduce boundary bias. Smallest repair: restore that clause and add one plain sentence on why missed mid-context evidence matters.
can reduce boundary bias but remain brittle in noisy settings
date — pass: originalPublished is 2026-10-25, which matches the v1 submission date. The feed date of Oct 31 is later and is correctly not used.
[Submitted on 25 Oct 2026]
support — pass: The claims, figures and the cost-reduction statement are verbatim-supported. The limitations and open question are appropriately conditional. Claim 1's wording 'a problem they call' is slightly off, because the source presents Lost in the Middle as an established, known phenomenon; this is minor and not a material contradiction.
improving accuracy by 3.0% on average over the strongest competing baseline and gold evidence recall by 9.4%
media — pass: The selected media is the verified arXiv abstract page, which is the direct original source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA language model with many learned numerical settings, used to process and generate language.
Definition referenceA setup in which a model first looks up relevant documents and then answers using them.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA language model with many learned numerical settings, used to process and generate language.
Definition referenceA setup in which a model first looks up relevant documents and then answers using them.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv