{"paper":{"title":"Facet-Level Tracing of Evidence Uncertainty and Hallucination in RAG","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Hallucinations in RAG systems arise mainly from how retrieved evidence is integrated during generation rather than from retrieval failures.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Markus Schedl, Monorama Swain, Passant Elchafei, Shahed Masoudian","submitted_at":"2026-04-10T09:59:43Z","abstract_excerpt":"Retrieval-Augmented Generation (RAG) aims to reduce hallucination by grounding answers in retrieved evidence, yet hallucinated answers remain common even when relevant documents are available. Existing evaluations focus on answer-level or passage-level accuracy, offering limited insight into how evidence is used during generation. In this work, we introduce a facet-level diagnostics framework for QA that decomposes each input question into atomic reasoning facets. For each facet, we assess evidence sufficiency and grounding using a structured Facet x Chunk matrix that combines retrieval releva"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"hallucinations in RAG systems are driven less by retrieval accuracy and more by how retrieved evidence is integrated during generation, with facet-level analysis exposing systematic evidence override and misalignment patterns that remain hidden under answer-level evaluation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That automatic decomposition of questions into atomic facets combined with NLI-based faithfulness scoring reliably captures whether and how evidence is used or overridden during generation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Facet-level analysis of RAG systems on medical QA and HotpotQA shows hallucinations stem primarily from evidence integration and override failures during generation, not from retrieval inaccuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Hallucinations in RAG systems arise mainly from how retrieved evidence is integrated during generation rather than from retrieval failures.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f14b32f3698266917a34f5c7875a11c4e27a5bb6b0500ee6e89721a3cca5a613"},"source":{"id":"2604.09174","kind":"arxiv","version":2},"verdict":{"id":"0ccf916c-6be8-4b0d-93e5-f558a2699a7b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T17:56:29.868503Z","strongest_claim":"hallucinations in RAG systems are driven less by retrieval accuracy and more by how retrieved evidence is integrated during generation, with facet-level analysis exposing systematic evidence override and misalignment patterns that remain hidden under answer-level evaluation.","one_line_summary":"Facet-level analysis of RAG systems on medical QA and HotpotQA shows hallucinations stem primarily from evidence integration and override failures during generation, not from retrieval inaccuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That automatic decomposition of questions into atomic facets combined with NLI-based faithfulness scoring reliably captures whether and how evidence is used or overridden during generation.","pith_extraction_headline":"Hallucinations in RAG systems arise mainly from how retrieved evidence is integrated during generation rather than from retrieval failures."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.09174/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0389b5fb664821386237ee476a91e6624e154de8066af1c2f2855a6bbc79be01"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}