{"record": {"aliases": ["OCR-MetaReasoning Benchmark"], "artifacts": [{"id": "arxiv:2608.30678", "kind": "paper", "url": "https://arxiv.org/abs/2608.30678"}, {"id": "gh:gengxuli/ocr-metareasoning", "kind": "repo", "url": "https://github.com/gengxuli/OCR-MetaReasoning"}], "authors": ["Gengxu Li", "Yuan Wu", "Yi Chang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.30678", "source_url": "https://arxiv.org/abs/2608.30678"}, "categories": ["General AI", "Reasoning", "Vision & 3D"], "description": {"en": "Evaluates MLLMs on text-rich image understanding by separating deduction, induction, and abduction across 1,500 samples with automatic answer scoring and reasoning-process compliance metrics."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.30678", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.30678", "title": "OCR-MetaReasoning"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2608.30678", "modality": null, "name": "OCR-MetaReasoning", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_ocr-metareasoning-benchmark_8b06e709", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "e11fbf6da9f8f5c27ca27c73eb3f510967635efb5e903603d3c34d728ade3132", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-09-02T05:23:11.715633Z", "source_url": "https://arxiv.org/abs/2608.30678"}, "publisher": null, "released": "2026-08-31", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.30678", "source_url": "https://arxiv.org/abs/2608.30678"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-30678", "source": "claire_radar", "source_benchmark_id": "2608.30678", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["OCR-MetaReasoning Benchmark"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-09-02", "attemptedAt": "2026-09-02T05:12:51.621571Z", "githubRepo": "https://github.com/gengxuli/OCR-MetaReasoning", "githubScope": "benchmark_repo", "githubStars": 1, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2608.30678", "observedAt": "2026-09-02T05:12:51.621571Z", "signalStatus": {"githubStars": {"observedAt": "2026-09-02T05:12:51.621571Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-09-02T05:12:51.621571Z", "state": "fresh"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Topical text-rich multimodal benchmark with balanced taxonomy and dual metrics likely to draw interest from MLLM evaluation community.", "score": 65}, "capabilities": ["Reasoning"], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-02T05:23:11.715633Z", "inputHash": "8921d3ccbf14a61d82a57a4b5dcf99bd5d70c95bbad1534d9cf697e2a0fc2717", "model": "deepseek-v4-pro"}, "curation": {"canonicalNameEvidence": "We introduce OCR-MetaReasoning, a controlled single-image benchmark", "canonicalNameSource": "abstract", "decisionReason": "Formally named benchmark with code, dataset, and evaluation scripts publicly available; defines clear scoring contract including MRMS and RPCS.", "model": "deepseek-v4-pro", "reviewedAt": "2026-09-02T05:23:11.715633Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Evaluates MLLMs on text-rich image understanding by separating deduction, induction, and abduction across 1,500 samples with automatic answer scoring and reasoning-process compliance metrics.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "score_submission", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce OCR-MetaReasoning, a controlled single-image benchmark that treats deduction, induction, and abduction as distinct directions and separates final-answer correctness from reasoning-process compliance."}, "familyId": "bmf_7e47bb1391c8", "firstSeenAt": "2026-09-01", "heat": null, "id": "bm_ocr-metareasoning-benchmark_8b06e709", "indexedAt": "2026-09-02T05:21:06.631990Z", "industrySectors": [], "links": {"code": "https://github.com/gengxuli/OCR-MetaReasoning", "data": null, "paper": "https://arxiv.org/abs/2608.30678", "pdf": "https://arxiv.org/pdf/2608.30678", "project": null, "report": "https://arxiv.org/abs/2608.30678"}, "metrics": [], "motivation": "Text-rich image understanding requires multimodal large language models (MLLMs) to organize OCR (Optical Character Recognition)-grounded evidence across words, layout, fields, charts, and visual correspondences.", "name": "OCR-MetaReasoning", "oneLine": "Evaluates MLLMs on text-rich image understanding by separating deduction, induction, and abduction across 1,500 samples with automatic answer scoring and reasoning-process compliance metrics.", "paperTitle": "OCR-MetaReasoning Benchmark: Evaluating the Meta-Reasoning Ability of MLLMs in Text-Rich Image Understanding", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.30678", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-09-02T05:21:06.631990Z"}, "publications": [], "ranking": {"30d": {"confidence": "High", "coverage": 0.85, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 30}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.10323645750017778, "value": 1}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.1228568942814387, "value": 1}}, "confidence": "High", "coverage": 0.85, "rank": 189, "score": 17}, "method": "attention-ranking-v13", "rank": 189, "score": 17}, "90d": {"confidence": "Medium", "coverage": 0.7, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.09434039834931938, "value": 1}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.1228568942814387, "value": 1}}, "confidence": "Medium", "coverage": 0.7, "rank": 394, "score": 22}, "method": "attention-ranking-v13", "rank": 394, "score": 22}}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-08-31", "source": {"authors": ["Gengxu Li", "Yuan Wu", "Yi Chang"], "categories": ["cs.CL"], "id": "2608.30678", "title": "OCR-MetaReasoning Benchmark: Evaluating the Meta-Reasoning Ability of MLLMs in Text-Rich Image Understanding", "type": "arxiv", "url": "https://arxiv.org/abs/2608.30678"}, "sourceUpdatedAt": "2026-08-31T12:20:05Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "Disentangles extraction from reasoning and outcome from process, enabling targeted diagnosis of multimodal reasoning failures in OCR-grounded tasks."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
