Relevance Is Not Sufficient Evidence: Detecting Evidence Gaps Before Generation in RAG
cs.CL, cs.IR, cs.LG
Submitted: 2026-09-27
Updated: 2026-09-27
Comments: 22 pages, 7 tables, 2 figures
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- mMARCO: A Multilingual Version of the MS MARCO Passage Ranking Dataset
- Semantics Delivery Network: Rethinking Web Retrieval Infrastructure for LLM Agents
- When Confidence Takes the Wrong Path: Diagnosing Retrieval-State Lock-In in RAG
- Language Models (Mostly) Know What They Know
- TASR: Training-Free Adaptive Stopping for Iterative Retrieval
- AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions
- RoBERTa: A Robustly Optimized BERT Pretraining Approach
- Stop-RAG: Value-Based Retrieval Control for Iterative RAG
- SURE-RAG: Sufficiency and Uncertainty-Aware Evidence Verification for Selective Retrieval-Augmented Generation
- Qwen2.5 Technical Report
- The Hallucination Tax of Reinforcement Finetuning
- Llama 2: Open Foundation and Fine-Tuned Chat Models
- TruthRL: Incentivizing Truthful LLMs via Reinforcement Learning
- Corrective Retrieval Augmented Generation
- LLM-Specific Utility for Retrieval-Augmented Generation
- GRACE: Reinforcement Learning for Grounded Response and Abstention under Contextual Evidence
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering