mlsanityData to 7 Oct 2026

Evaluating and Guarding Citation Faithfulness in Agentic Scientific Synthesis

arXiv 2607.20527 ↗First posted 10 Jul 2026cs.AIcited by 0

On the map: LLM agents and tool use → Web, search, and research agents

Abstract (arXiv)

Agentic LLM systems such as OpenScholar and PaperQA2 read the scientific literature and return cited answers, and both they and their benchmarks already check whether those citations hold, with a fixed attribution model or human graders. Neither audits the reliability of that check itself. We show it is not reliable, and that this matters. On identical agent outputs the measured unsupported-citation rate ranges from about 3% to about 18% depending only on the verifier's strictness, and although verifiers agree on which citations are supported, they disagree on which to flag (negative-specific agreement 0.27 to 0.30), so no single flag set is trustworthy and cross-paper comparison is invalid without a named verifier and protocol. We present a gold-anchored evaluation protocol and a deployable guard that make this behavior measurable and bounded. The protocol validates the verifier, measures re-attribution, and calibrates a guarantee against human gold rather than another model's verdict; the verifier is a swappable instrument chosen on cost (recall 0.94 on the supported class, held out), and re-attribution is a commodity step where a deterministic BM25 matches the best open generator. The guard adds a split-conformal layer placing a distribution-free, finite-sample bound on truly unsupported citations that slip past a chosen flagging rule, a guarantee on catch rate rather than conclusion correctness. The bound holds on held-out gold, and we identify and quantify the condition governing its transfer to deployment, calibration-negative difficulty, with a concrete recalibration recipe, left untested by prior conformal-factuality work. Validated across four open 27-35B models and three agentic pipelines on public benchmarks (SciFact, QASA, PubMedQA), with confidence intervals on every headline number, the protocol and guard ship as an open single-GPU kit.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Web, search, and research agents, one of 16 directions in LLM agents and tool use. Builds on 22 of its 50 references that are in the map. First posted on arXiv 10 Jul 2026.

Builds on

22 of 50 references.

References · 22

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  2. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
  3. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,193
  4. 2023Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflectioncited by 1,800
  5. 2024τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domainscited by 1,280
  6. 2024Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluatorscited by 865
  7. 2017Selective Classification for Deep Neural Networkscited by 799
  8. 2024LLM Evaluators Recognize and Favor Their Own Generationscited by 651
  9. 2023ChemCrow: Augmenting large-language models with chemistry toolscited by 609
  10. 2023Enabling Large Language Models to Generate Text with Citationscited by 584
  11. 2024Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Modelscited by 431
  12. 2024AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agentscited by 319
10 more
  1. 2024Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Modelscited by 301
  2. 2021Measuring Attribution in Natural Language Generation Modelscited by 253
  3. 2018Categorizing Variants of Goodhart's Lawcited by 132
  4. 2024Empowering Biomedical Discovery with AI Agentscited by 117
  5. 2025Evaluating LLM-based Agents for Multi-Turn Conversations: A Surveycited by 56
  6. 2025Rating Roulette: Self-Inconsistency in LLM-As-A-Judge Frameworkscited by 37
  7. 2024AttributionBench: How Hard is Automatic Attribution Evaluation?cited by 26
  8. 2025CiteEval: Principle-Driven Citation Evaluation for Source Attributioncited by 19
  9. 2024Correctness is not Faithfulness in RAG Attributionscited by 18
  10. 2026CiteAudit: You Cited It, But Did You Read It? A Benchmark for Verifying Scientific References in the LLM Eracited by 14

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Who is the Agent to Blame? Localizing Faithfulness and Citation Mistakes in Agentic Deep Research
  2. 2026ReCite: Agentic Reasoning for Faithful Citation
  3. 2026Read the Paper, Write the Code: Agentic Reproduction of Social-Science Results
  4. 2026Traxia: A Framework for Verifiable, Agent-Native Scientific Publishing
  5. 2025BadScientist: Can a Research Agent Write Convincing but Unsound Papers that Fool LLM Reviewers?
  6. 2026OpenNovelty: An LLM-powered Agentic System for Verifiable Scholarly Novelty Assessment
  7. 2026Agentic Publication Protocol: An Attempt to Modernize Scientific Publication
  8. 2026Cited but Not Verified: Parsing and Evaluating Source Attribution in LLM Deep Research Agents