{"paper":{"title":"CoCoReviewBench: A Completeness- and Correctness-Oriented Benchmark for AI Reviewers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"CoCoReviewBench evaluates AI reviewers using expert discussions to measure both completeness and correctness.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dehao Huang, Derek F. Wong, Hexuan Deng, Min Zhang, Ruina Hu, Xiaopeng Ke, Xuebo Liu, Yichen Li, Yue Wang","submitted_at":"2026-05-08T15:44:26Z","abstract_excerpt":"Despite the rapid development of AI reviewers, evaluating such systems remains challenging: metrics favor overlap with human reviews over correctness. However, since human reviews often cover only a subset of salient issues and sometimes contain mistakes, they are unreliable as gold references. To address this, we build category-specific benchmark subsets and skip evaluation when the corresponding human reviews are missing to strengthen Completeness. We also leverage reviewer--author--meta-review discussions as expert annotations and filter unreliable reviews accordingly to strengthen Correctn"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"we introduce CoCoReviewBench, which curates 3,900 papers from ICLR and NeurIPS to enable reliable and fine-grained evaluation of AI reviewers. Analysis shows that AI reviewers remain limited in correctness and are prone to hallucinations, and highlights reasoning models as more effective reviewers.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That reviewer-author-meta-review discussions provide reliable expert annotations for correctness and that skipping evaluation when human reviews are missing strengthens completeness without introducing selection bias.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CoCoReviewBench curates 3,900 conference papers with category subsets and expert discussion annotations to evaluate AI reviewers on completeness and correctness, showing they are limited and prone to hallucinations while reasoning models perform better.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CoCoReviewBench evaluates AI reviewers using expert discussions to measure both completeness and correctness.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"50484b090c7c76e8972c930de18f653350a830a05c5df0cc441bd34cd73794d9"},"source":{"id":"2605.07905","kind":"arxiv","version":2},"verdict":{"id":"bc9fd92b-95be-4107-91c7-cfc63b0fdd92","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-11T03:35:25.842977Z","strongest_claim":"we introduce CoCoReviewBench, which curates 3,900 papers from ICLR and NeurIPS to enable reliable and fine-grained evaluation of AI reviewers. Analysis shows that AI reviewers remain limited in correctness and are prone to hallucinations, and highlights reasoning models as more effective reviewers.","one_line_summary":"CoCoReviewBench curates 3,900 conference papers with category subsets and expert discussion annotations to evaluate AI reviewers on completeness and correctness, showing they are limited and prone to hallucinations while reasoning models perform better.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That reviewer-author-meta-review discussions provide reliable expert annotations for correctness and that skipping evaluation when human reviews are missing strengthens completeness without introducing selection bias.","pith_extraction_headline":"CoCoReviewBench evaluates AI reviewers using expert discussions to measure both completeness and correctness."},"integrity":{"clean":false,"summary":{"advisory":1,"critical":1,"by_detector":{"doi_compliance":{"total":2,"advisory":1,"critical":1,"informational":0}},"informational":0},"endpoint":"/pith/2605.07905/integrity.json","findings":[{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.18653/v1/2023.acl-long.277.URL) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":4,"audited_at":"2026-05-19T11:25:30.935354Z","detected_doi":"10.18653/v1/2023.acl-long.277.URL","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null},{"note":"Identifier '10.2340/aos.v84.43' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","detector":"doi_compliance","severity":"critical","ref_index":7,"audited_at":"2026-05-19T11:25:30.935354Z","detected_doi":"10.2340/aos.v84.43","finding_type":"unresolvable_identifier","verdict_class":"cross_source","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T15:31:18.389065Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T11:25:30.935354Z","status":"completed","version":"1.0.0","findings_count":2}],"snapshot_sha256":"532b6e22e37f00816ad08e9a2c896b46527cc68f3e919d30990ab9d27d45bc26"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"a7ad63cec115e6dbf5d245717437547be6b6c51aa10292abca473981eab90736"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}