Retrieval-Augmented Generation (RAG) significantly improves the factuality of Large Language Models (LLMs), yet standard pipelines often lack mechanisms to verify inter- mediate reasoning, leaving them vulnerable to hallucinations in high-stakes domains. To address this, we propose a domain-specific RAG framework that integrates explicit rea- soning and faithfulness verification. Our architecture augments standard retrieval with neural query rewriting, BGE-based cross-encoder reranking, and a rationale generation module that grounds sub-claims in specific evidence spans. We further introduce an eight-category verification taxonomy that enables fine-grained assessment of rationale faithfulness, distinguishing between explicit and implicit support patterns to facilitate structured error diagnosis. We evaluate this framework on the BioASQ and PubMedQA benchmarks, specifically analyzing the impact of dynamic in-context learning and rerank- ing under constrained token budgets. Experiments demonstrate that explicit rationale generation improves accuracy over vanilla RAG baselines, while dynamic demonstration selection combined with robust reranking yields further gains in few-shot settings. Using Llama-3-8B-Instruct, our approach achieves 89.1% on BioASQ-Y/N and 73.0% on Pub- MedQA, competitive with systems using significantly larger models. Additionally, we perform a pilot study combining human expert assessment with LLM-based verification to explore how explicit rationale generation improves system transparency and enables more detailed diagnosis of retrieval failures in biomedical question answering.
Figures & tables
Category
Meaning
CORRECT-EXPLICIT
Information is explicitly stated in the documents (quoted or paraphrased).
CORRECT-IMPLICIT
Facts are not stated verbatim but are logically inferred from context clues.
CORRECT-ADDITIONAL
Uses context accurately but adds relevant, correct external details.
CORRECT-MISSING
Conclusion is correct, but the cited documents provide no support (irrelevant).
INCORRECT-FALSE
Statements directly contradict evidence provided in the context.
INCORRECT-DEVIATING
Statements are off-topic or unrelated to the query/documents.
Table 1. Legend for Rationale-Verification Label Categories
Method
BioASQ-Y/N
PubMedQA*
Closed-Book (No Retrieval; CoT Prompting)
GPT-3.5 †
74.27
36.00
GPT-4 †
84.30
39.60
Mixtral-8 × 7B †
77.51
35.20
Llama2-70B †
61.17
42.20
MEDITRON-70B †
68.45
53.40
Table 2. Comparison with baseline methods under MIRAGE settings. Accuracy (%) reported on BioASQ-Y/N (618 yes/no questions) and PubMedQA* (500 questions, context removed). All baselines use question-only retrieval without gold passages. † Results from MIRAGE benchmark [ 14 ] .
BioASQ
PubMedQA
Configuration
0
1
2
3
4
0
1
2
3
4
Static Demonstration Selection
w/o Reranking
85.8
78.9
78.5
77.7
70.4
73.0
54.0
60.0
56.0
47.5
w/ Reranking
87.4
79.8
79.6
76.4
71.7
72.5
60.5
60.0
60.0
60.0
Δ (Reranking)
+1.6
+0.9
+1.1
-1.3
+1.3
-0.5
+6.5
–
+4.0
+12.5
Dynamic Demonstration Selection (w/ Reranking)
Table 3. Ablation study showing the impact of reranking and demonstration selection strategy. Accuracy (%) reported across varying numbers of ICL demonstrations.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Example
Annotator A
Annotator B
LLM Verifier
Question 1
1.00
1.00
1.00
Question 2
0.75
0.50
0.83
Question 3
0.75
0.30
0.93
Question 4
0.90
0.80
1.00
Appendix
Table 4. Pilot per-example faithfulness scores (4 examples). Scores are shown for two human annotators and the LLM verifier.
Retrieval-augmented generation (RAG) substantially extends the knowledge boundary of large language models. However, it still faces two major challenges when handling complex reasoning tasks: low context utilization and frequent hallucinations. To address these issues, we propose Self-Correcting RAG, a unified framework that reformulates retrieval and generation as constrained optimization and path planning. On the input side, we move beyond traditional greedy retrieval and, for the first time, formalize context selection as a multi-dimensional multiple-choice knapsack problem (MMKP), thereby maximizing information density and removing redundancy under a strict token budget. On the output side, we introduce a natural language inference (NLI)-guided Monte Carlo Tree Search (MCTS) mechanism, which leverages test-time compute to dynamically explore reasoning trajectories and validate the faithfulness of generated answers. Experiments on six multi-hop question answering and fact-checking datasets demonstrate that our method significantly improves reasoning accuracy on complex queries while effectively reducing hallucinations, outperforming strong existing baselines.Our code is available at https://github.com/xjiacs/Self-Correcting-RAG .
Shijia Xu, Zhou Wu, Xiaolong Jia +3
Chongqing University, China · Queen Mary University of London, UK · Chongqing Key Laboratory of Big Data Intelligence and Privacy Computing, China +1
Retrieval-augmented generation (RAG) mitigates hallucination but does not eliminate it: a deployed system must still decide, at inference time, whether its answer is actually supported by the retrieved evidence. We introduce LatentAudit, a white-box auditor that pools mid-to-late residual-stream activations from an open-weight generator and measures their Mahalanobis distance to the evidence representation. The resulting quadratic rule requires no auxiliary judge model, runs at generation time, and is simple enough to calibrate on a small held-out set. We show that residual-stream geometry carries a usable faithfulness signal, that this signal survives architecture changes and realistic retrieval failures, and that the same rule remains amenable to public verification. On PubMedQA with Llama-3-8B, LatentAudit reaches 0.942 AUROC with 0.77,ms overhead. Across three QA benchmarks and five model families (Llama-2/3, Qwen-2.5/3, Mistral), the monitor remains stable; under a four-way stress test with contradictions, retrieval misses, and partial-support noise, it reaches 0.9566--0.9815 AUROC on PubMedQA and 0.9142--0.9315 on HotpotQA. At 16-bit fixed-point precision, the audit rule preserves 99.8% of the FP16 AUROC, enabling Groth16-based public verification without revealing model weights or activations. Together, these results position residual-stream geometry as a practical basis for real-time RAG faithfulness monitoring and optional verifiable deployment.
Zhe Yu, Wenpeng Xing, Meng Han
Binjiang Institute of Zhejiang University · Communication University of Zhejiang · Zhejiang University
The conventional Retrieval-Augmented Generation (RAG) paradigm of injecting raw retrieved texts into the Large Language Model (LLM)'s context often results in suboptimal integration of retrieved information. This paper proposes to bridge retrieval results and the LLM's reasoning ability through Verbal Annotations, analytic narratives that explicitly articulate the logical connection between a search query and retrieved contexts. Our empirical investigation reveals the potential of Verbal Annotations to substantially enhance the LLM's ability to generate accurate, contextually-grounded responses. Motivated by this finding, we introduce Verbal-R3, a novel agentic RAG framework that consists of a Generator and a Verbal Reranker. The Generator performs iterative retrieval and reasoning, while the Verbal Reranker returns relevance scores and Verbal Annotations to guide the reasoning and answering process of the Generator. The inference process of Verbal-R3 is further refined through relevance-guided test-time scaling, which efficiently allocates test-time compute for effective trajectory expansion. Verbal-R3 achieves state-of-the-art performance on complex Question Answering benchmarks, validating the effectiveness of the proposed framework.
Sangkwon Park, Donghun Kang, Jisoo Mok +1
Department of Electrical and Computer Engineering, Seoul National University · DGIST · Interdisciplinary Program in Artificial Intelligence, Seoul National University +1