CounterRefine: Answer-Conditioned Counterevidence Retrieval for Inference-Time Knowledge Repair in Factual Question Answering

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Huang, Tianyi, Deng, Ying Kai
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909048153047040
author Huang, Tianyi
Deng, Ying Kai
author_facet Huang, Tianyi
Deng, Ying Kai
contents In factual question answering, many errors are not failures of access but failures of commitment: the system retrieves relevant evidence, yet still settles on the wrong answer. We present CounterRefine, a lightweight repair layer for short-form RAG that treats the first answer as a hypothesis to test. Given a draft, CounterRefine issues answer-conditioned expansion queries to retrieve candidate-specific evidence, then applies a constrained KEEP or REVISE refinement step whose proposed revisions are accepted only after deterministic validation. The design is intentionally narrow: it adds one evidence-gathering pass and one guarded refinement call rather than replacing the retriever or building a broad agentic system. On the full SimpleQA benchmark, CounterRefine improves a matched one-pass RAG baseline by up to 5.8 correct-rate points; in the full Claude trace, it changes only 5.6% of outputs, with 180 beneficial outcome changes and 8 harmful ones. These findings suggest a simple but important direction for knowledgeable foundation models: beyond accessing evidence, they should also be able to use that evidence to reconsider and, when necessary, repair their own answers.
format Preprint
id arxiv_https___arxiv_org_abs_2603_16091
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle CounterRefine: Answer-Conditioned Counterevidence Retrieval for Inference-Time Knowledge Repair in Factual Question Answering
Huang, Tianyi
Deng, Ying Kai
Computation and Language
Artificial Intelligence
In factual question answering, many errors are not failures of access but failures of commitment: the system retrieves relevant evidence, yet still settles on the wrong answer. We present CounterRefine, a lightweight repair layer for short-form RAG that treats the first answer as a hypothesis to test. Given a draft, CounterRefine issues answer-conditioned expansion queries to retrieve candidate-specific evidence, then applies a constrained KEEP or REVISE refinement step whose proposed revisions are accepted only after deterministic validation. The design is intentionally narrow: it adds one evidence-gathering pass and one guarded refinement call rather than replacing the retriever or building a broad agentic system. On the full SimpleQA benchmark, CounterRefine improves a matched one-pass RAG baseline by up to 5.8 correct-rate points; in the full Claude trace, it changes only 5.6% of outputs, with 180 beneficial outcome changes and 8 harmful ones. These findings suggest a simple but important direction for knowledgeable foundation models: beyond accessing evidence, they should also be able to use that evidence to reconsider and, when necessary, repair their own answers.
title CounterRefine: Answer-Conditioned Counterevidence Retrieval for Inference-Time Knowledge Repair in Factual Question Answering
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2603.16091