mlsanityData to 7 Oct 2026

Regime Boundary Alignment for Evidence-Gated Question Answering

arXiv 2609.37491 ↗First posted 25 Sep 2026cs.CL cs.AIcited by 0

On the map: Retrieval-augmented generation → RAG robustness and adaptive retrieval

Abstract (arXiv)

Retrieval-augmented language models are expected to answer from the retrieved evidence, but in practice they often keep answering when that evidence is missing. We trace this behavior to the training signal: answer-focused fine-tuning assigns no target to unsupported contexts, so it cannot distinguish a reader that abstains from one that guesses, and unsupported answering stays near 100% even as supported accuracy improves. We introduce Regime Boundary Alignment (RBA), which trains a single reader on matched variants of the same question and gold answer. The reader is trained to produce the gold answer when the context supports it, including when conflicting evidence is also present, and to abstain when the correct support is removed; inference is ordinary decoding, with no verifier, threshold, or regime label. On three multi-hop QA datasets across three seeds, RBA reduces the unsupported-answer rate by more than sixty percentage points relative to conflict-focused training while matching its supported accuracy. On a held-out TriviaQA retrieval-miss slice, the same reader reduces unsupported answering from 100% to below 1% while also improving supported accuracy. These results indicate that evidence-gated answering must be learned on both sides of the support boundary.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in RAG robustness and adaptive retrieval, one of 15 directions in Retrieval-augmented generation. Builds on 35 of its 38 references that are in the map. First posted on arXiv 25 Sep 2026.

Builds on

35 of 38 references.

Closest ancestors · 7

Same direction, newest first

  1. 2025FaithfulRAG: Fact-Level Conflict Modeling for Context-Faithful Retrieval-Augmented Generation10 Jun
  2. 2025Resisting Contextual Interference in RAG via Parametric-Knowledge Reinforcement5 Jun
  3. 2025Divide-Then-Align: Honest Alignment based on the Knowledge Boundary of RAG27 May
  4. 2025Retrieval-Augmented Generation with Conflicting Evidence17 Apr
  5. 2024Adaptive Contrastive Decoding in Retrieval-Augmented Generation for Handling Noisy Contexts2 Aug
  6. 2023Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection17 Oct
  7. 2023Benchmarking Large Language Models in Retrieval-Augmented Generation4 Sep

Other references · 28

  1. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  2. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,686
  3. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,383
  4. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,895
  5. 2020REALM: Retrieval-Augmented Language Model Pre-Trainingcited by 2,126
  6. 2018Know What You Don't Know: Unanswerable Questions for SQuADcited by 1,538
  7. 2020Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Stepscited by 1,259
  8. 2020Leveraging Passage Retrieval with Generative Models for Open Domain Question Answeringcited by 1,203
  9. 2021MuSiQue: Multihop Questions via Single-hop Question Compositioncited by 893
  10. 2023Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generationcited by 795
  11. 2022Teaching Models to Express Their Uncertainty in Wordscited by 731
  12. 2023Making Retrieval-Augmented Language Models Robust to Irrelevant Contextcited by 364
16 more
  1. 2021Entity-Based Knowledge Conflicts in Question Answeringcited by 349
  2. 2023Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflictscited by 333
  3. 2023Trusting Your Evidence: Hallucinate Less with Context-aware Decodingcited by 319
  4. 2024The Power of Noise: Redefining Retrieval for RAG Systemscited by 275
  5. 2020Selective Question Answering under Domain Shiftcited by 228
  6. 2023Chain-of-Note: Enhancing Robustness in Retrieval-Augmented Language Modelscited by 180
  7. 2022DisentQA: Disentangling Parametric and Contextual Knowledge with Counterfactual Question Answeringcited by 93
  8. 2023Self-Evaluation Improves Selective Generation in Large Language Modelscited by 74
  9. 2024AdaCAD: Adaptively Decoding to Balance Conflicts between Contextual and Parametric Knowledgecited by 35
  10. 2024Discerning and Resolving Knowledge Conflicts through Adaptive Decoding with Contextual Information-Entropy Constraintcited by 27
  11. 2024Assessing "Implicit" Retrieval Robustness of Large Language Modelscited by 9
  12. 2025SAFER: Risk-Constrained Sample-then-Filter in Large Language Modelscited by 7
  13. 2025Toward Faithful Retrieval-Augmented Generation with Sparse Autoencoderscited by 6
  14. 2025KnowGuard: Knowledge-Driven Abstention for Multi-Round Clinical Reasoningcited by 5
  15. 2025Can LLMs Refuse Questions They Do Not Know? Measuring Knowledge-Aware Refusal in Factual Taskscited by 5
  16. 2025High Accuracy, Less Talk (HALT): Reliable LLMs through Capability-Aligned Finetuningcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering
  2. 2026TrustMargin: Training-Free Arbitration between Parametric Memory and Retrieved Evidence in Large Language Models
  3. 2026Context Convergence Improves Answering Inferential Questions
  4. 2025Improving Context Fidelity via Native Retrieval-Augmented Reasoning
  5. 2026PAVE: Premise-Aware Validation and Editing for Retrieval-Augmented LLMs
  6. 2024ALR2: A Retrieve-then-Reason Framework for Long-context Question Answering
  7. 2026When Evidence Conflicts: Uncertainty and Order Effects in Retrieval-Augmented Biomedical Question Answering
  8. 2026PassiveQA: A Three-Action Framework for Epistemically Calibrated Question Answering via Supervised Finetuning