{"paper":{"title":"Courtroom-Style Multi-Agent Debate with Progressive RAG and Role-Switching for Controversial Claim Verification","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Courtroom-style debate with progressive retrieval raises claim verification accuracy to 81.7 percent on Check-COVID.","cross_cats":["cs.AI","cs.MA"],"primary_cat":"cs.CL","authors_text":"Hasan Mahmud, Masnun Nuha Chowdhury, Md Kamrul Hasan, Nusrat Jahan Beg, Syed Rifat Raiyan, Umme Hunny Khan","submitted_at":"2026-03-30T14:23:15Z","abstract_excerpt":"Large language models (LLMs) remain unreliable for high-stakes claim verification due to hallucinations and shallow reasoning. While retrieval-augmented generation (RAG) and multi-agent debate (MAD) address this, they are limited by one-pass retrieval and unstructured debate dynamics. We propose a courtroom-style multi-agent framework, PROClaim, that reformulates verification as a structured, adversarial deliberation. Our approach integrates specialized roles (e.g., Plaintiff, Defense, Judge) with Progressive RAG (P-RAG) to dynamically expand and refine the evidence pool during the debate. Fur"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"In zero-shot evaluations on the Check-COVID benchmark, PROClaim achieves 81.7% accuracy, outperforming standard multi-agent debate by 10.0 percentage points, with P-RAG driving the primary performance gains (+7.5 pp).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the courtroom structure and progressive evidence retrieval provide genuine robustness against hallucinations and biases, rather than overfitting to the specific Check-COVID dataset or the models used in the evaluation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"PROClaim achieves 81.7% accuracy on Check-COVID claim verification by combining courtroom roles, progressive RAG, and multi-judge aggregation, outperforming standard multi-agent debate by 10 percentage points.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Courtroom-style debate with progressive retrieval raises claim verification accuracy to 81.7 percent on Check-COVID.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b01a1d1361af76f11f31edabed60fd06b358c0d26678dee9679e30685acb84d5"},"source":{"id":"2603.28488","kind":"arxiv","version":3},"verdict":{"id":"20c5ddb0-5737-497a-8178-c90c6adca83a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T22:09:54.875625Z","strongest_claim":"In zero-shot evaluations on the Check-COVID benchmark, PROClaim achieves 81.7% accuracy, outperforming standard multi-agent debate by 10.0 percentage points, with P-RAG driving the primary performance gains (+7.5 pp).","one_line_summary":"PROClaim achieves 81.7% accuracy on Check-COVID claim verification by combining courtroom roles, progressive RAG, and multi-judge aggregation, outperforming standard multi-agent debate by 10 percentage points.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the courtroom structure and progressive evidence retrieval provide genuine robustness against hallucinations and biases, rather than overfitting to the specific Check-COVID dataset or the models used in the evaluation.","pith_extraction_headline":"Courtroom-style debate with progressive retrieval raises claim verification accuracy to 81.7 percent on Check-COVID."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.28488/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":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}