{"record": {"aliases": ["RealMath-Eval: Why SOTA Judges Struggle with Real Human Reasoning"], "artifacts": [{"id": "arxiv:2606.10254", "kind": "paper", "url": "https://arxiv.org/abs/2606.10254"}, {"id": "gh:richarmd/realmath-eval", "kind": "repo", "url": "https://github.com/RicharMd/RealMath-Eval"}, {"id": "hf:richarmd/realmath-eval", "kind": "dataset", "url": "https://huggingface.co/datasets/RicharMd/RealMath-Eval"}], "authors": ["Yiteng Mao", "Kenan Xu", "Yijia Lyu", "Wenhao Li", "Jianlong Chen", "Xiangfeng Wang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.10254", "source_url": "https://arxiv.org/abs/2606.10254"}, "categories": ["Language & Knowledge", "Mathematics & Formal Science", "Reasoning"], "description": {"en": "A benchmark of authentic high-school exam responses for evaluating how accurately language-model judges assess human mathematical reasoning."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.10254", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.10254", "title": "RealMath-Eval"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.10254", "modality": null, "name": "RealMath-Eval", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_realmath-eval-why-sota-judges-struggle-wit_c2d9ad9e", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "590254acd13e8c03a27f6823285e6fbb04d9cb036bdfd6fd972ec57ec2c87a1a", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:10:41.268638Z", "source_url": "https://arxiv.org/abs/2606.10254"}, "publisher": null, "released": "2026-06-08", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.10254", "source_url": "https://arxiv.org/abs/2606.10254"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-10254", "source": "claire_radar", "source_benchmark_id": "2606.10254", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["RealMath-Eval: Why SOTA Judges Struggle with Real Human Reasoning"], "annotation": "Machine Generated", "applicationDomains": ["Mathematics & Formal Science"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "The benchmark addresses LLM-as-judge reliability in education, but the supplied evidence does not establish likely near-term attention.", "score": 3}, "capabilities": ["Reasoning"], "confidence": "Medium", "construction": "Original Synthetic", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:10:41.268638Z", "inputHash": "b31a18fd1f001535135aab3139ef17f1c6a7f086efb16f35bee9438682570643", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "RealMath-Eval: Why SOTA Judges Struggle with Real Human Reasoning", "canonicalNameSource": "paper_title", "decisionReason": "The source formally names the benchmark, defines the evaluation task, reports a comparable MSE against expert grading, and provides public code and data paths.", "evaluationEvidence": {"reuse": "Code available at https://github.com/RicharMd/RealMath-Eval , Data available at https://huggingface.co/datasets/RicharMd/RealMath-Eval", "scoring": "Mean Squared Error ($\\sim$2.96) against expert human grading", "task": "their ability to \\emph{evaluate} the diverse reasoning processes of real human students"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:10:41.268638Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A benchmark of authentic high-school exam responses for evaluating how accurately language-model judges assess human mathematical reasoning.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "To bridge this gap, we introduce \\textbf{RealMath-Eval}, a rigorously annotated benchmark of 224 real-world exam responses from high schools."}, "familyId": "bmf_8b13a6bf2d6f", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_realmath-eval-why-sota-judges-struggle-wit_c2d9ad9e", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/RicharMd/RealMath-Eval", "data": "https://huggingface.co/datasets/RicharMd/RealMath-Eval", "paper": "https://arxiv.org/abs/2606.10254", "pdf": "https://arxiv.org/pdf/2606.10254", "project": null, "report": "https://arxiv.org/abs/2606.10254"}, "metrics": [], "motivation": "While Large Language Models (LLMs) have achieved near-perfect performance in \\emph{solving} high-school mathematics, their ability to \\emph{evaluate} the diverse reasoning processes of real human students remains under-examined.", "name": "RealMath-Eval", "oneLine": "A benchmark of authentic high-school exam responses for evaluating how accurately language-model judges assess human mathematical reasoning.", "paperTitle": "RealMath-Eval: Why SOTA Judges Struggle with Real Human Reasoning", "primaryDomain": "Mathematics & Formal Science", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.10254", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-06-08", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Yiteng Mao", "Kenan Xu", "Yijia Lyu", "Wenhao Li", "Jianlong Chen", "Xiangfeng Wang"], "categories": ["cs.AI", "cs.CL"], "id": "2606.10254", "title": "RealMath-Eval: Why SOTA Judges Struggle with Real Human Reasoning", "type": "arxiv", "url": "https://arxiv.org/abs/2606.10254"}, "sourceUpdatedAt": "2026-06-08T23:40:34Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "It tests evaluation behavior on real student work and exposes a measured gap between grading authentic and synthetic reasoning."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
