{"record": {"aliases": ["visual-rag-benchmark"], "artifacts": [{"id": "gh:james-a-watson/visual-rag-benchmark", "kind": "repo", "url": "https://github.com/james-a-watson/visual-rag-benchmark"}], "categories": ["Finance", "Information retrieval", "Language & Knowledge", "cs.AI"], "description": {"en": "Visual RAG Benchmark evaluates end-to-end retrieval-augmented generation on reference tables from three EU Water Framework Directive guidance documents, scoring answers on 103 and 154 question sets split into text and table question types via LLM-as-a-judge correctness, ROUGE-1, and retrieval accuracy across seven indexing and retrieval pipelines."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://github.com/james-a-watson/visual-rag-benchmark", "source": "claire_radar", "source_url": "https://github.com/james-a-watson/visual-rag-benchmark", "title": "Visual RAG Benchmark"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:github:james-a-watson/visual-rag-benchmark", "modality": null, "name": "Visual RAG Benchmark", "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": "2026-10-14T09:44:51Z", "data_status": "primary-source-candidate", "display_eligible": "false", "origin_record_id": "bm_visual-rag-benchmark_df083a55", "origin_source": "github", "recognition_confidence": "0.4", "record_sha256": "6e535b3223cc6e24abe0b629067d4ebfac309fd93ae9395c95c3d7c046be9a4b", "relation": "unclear", "review_model": "deepseek-v4-pro", "review_state": "ai-name-audit-deferred", "reviewed_at": "2026-09-13T06:31:18.769897Z", "source_url": "https://github.com/james-a-watson/visual-rag-benchmark"}, "publisher": null, "released": "2026-09-10", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:github:james-a-watson/visual-rag-benchmark", "source_url": "https://github.com/james-a-watson/visual-rag-benchmark"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-github-james-a-watson-visual-rag-benchmark", "source": "claire_radar", "source_benchmark_id": "github:james-a-watson/visual-rag-benchmark", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["visual-rag-benchmark"], "annotation": "Unknown", "applicationDomains": ["Finance"], "area": "Language & Knowledge", "attention": {"asOf": "2026-09-12", "attemptedAt": "2026-09-12T01:46:06.616968Z", "githubRepo": "https://github.com/james-a-watson/visual-rag-benchmark", "githubScope": "benchmark_repo", "githubStars": 0, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": null, "observedAt": "2026-09-12T01:46:06.616968Z", "signalStatus": {"githubStars": {"observedAt": "2026-09-12T01:46:06.616968Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"state": "not_applicable"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Retrieval-augmented generation on structured documents is a practical, widely followed area and the repository is self-contained with committed runs, though it originates from a single MSc project with no conference venue.", "score": 36}, "capabilities": ["Information retrieval"], "confidence": "Medium", "construction": "Transform Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-13T09:43:44.761778Z", "inputHash": "05dd929fcd9b1d713fae5129d263bf75c7b45a1c6648b3a9fa7a26714446d6c3", "model": "deepseek-v4-flash"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "visual-rag-benchmark End-to-end evaluation of retrieval-augmented generation on reference tables.", "canonicalNameSource": "abstract", "decisionReason": "Published dataset, code, and instructions for reproducing the evaluation, though scoring relies on LLM judge.", "evaluationEvidence": {"reuse": "python scripts/summarise_results.py # recompute correctness for every run python scripts/summarise_results.py --failures simple_text # inspect what the baseline got wrong", "scoring": "Correctness is measured using LLM-as-a-judge, following Zheng et al. The question, the true answer and the generated answer are passed to a judge model prompted to return \"Y\" or \"N\".", "task": "benchmark/paper_question_set.json 103 QA pairs — the published evaluation set benchmark/question_set.json 154 QA pairs — later, more table-weighted extension"}, "model": "deepseek-v4-pro", "reviewedAt": "2026-09-13T06:31:18.769897Z", "state": "ai-name-audit-deferred"}, "dataStatus": "primary-source-candidate", "demo": false, "description": "Visual RAG Benchmark evaluates end-to-end retrieval-augmented generation on reference tables from three EU Water Framework Directive guidance documents, scoring answers on 103 and 154 question sets split into text and table question types via LLM-as-a-judge correctness, ROUGE-1, and retrieval accuracy across seven indexing and retrieval pipelines.", "discoverySources": [{"id": "github:james-a-watson/visual-rag-benchmark", "type": "github", "url": "https://github.com/james-a-watson/visual-rag-benchmark"}], "displayEligible": false, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["discovered via github", "benchmark term in abstract", "evaluation protocol evidence", "no explicit benchmark release evidence"], "snippet": "visual-rag-benchmark End-to-end evaluation of retrieval-augmented generation on reference tables."}, "familyId": "bmf_9aff9a50e904", "firstSeenAt": "2026-09-11", "heat": null, "id": "bm_visual-rag-benchmark_df083a55", "indexedAt": "2026-09-11T06:20:17.513010Z", "industrySectors": ["Financial Services"], "links": {"code": "https://github.com/james-a-watson/visual-rag-benchmark"}, "metrics": [], "motivation": "visual-rag-benchmark End-to-end evaluation of retrieval-augmented generation on reference tables.", "name": "Visual RAG Benchmark", "oneLine": "Visual RAG Benchmark evaluates end-to-end retrieval-augmented generation on reference tables from three EU Water Framework Directive guidance documents, scoring answers on 103 and 154 question sets split into text and table question types via LLM-as-a-judge correctness, ROUGE-1, and retrieval accuracy across seven indexing and retrieval pipelines.", "paperTitle": "visual-rag-benchmark", "primaryDomain": "Finance", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://github.com/james-a-watson/visual-rag-benchmark", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-09-11T06:20:17.513010Z"}, "publications": [], "publishers": [{"name": "james-a-watson", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/james-a-watson/visual-rag-benchmark"}], "ranking": {"30d": {"confidence": "Low", "coverage": 0.55, "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": 0.0, "value": 0}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 144, "score": 23}, "method": "attention-ranking-v13", "rank": 144, "score": 23}, "90d": {"confidence": "Low", "coverage": 0.55, "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": 0.0, "value": 0}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 336, "score": 23}, "method": "attention-ranking-v13", "rank": 336, "score": 23}}, "readiness": "Runnable", "recognitionConfidence": 0.4, "relation": "unclear", "releasedAt": "2026-09-10", "source": {"authors": ["james-a-watson"], "categories": [], "id": "github:james-a-watson/visual-rag-benchmark", "publicSignals": {"githubStars": 0}, "title": "visual-rag-benchmark", "type": "github", "url": "https://github.com/james-a-watson/visual-rag-benchmark"}, "sourceUpdatedAt": "2026-09-10T16:55:15Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "RAG evaluations built on prose corpora understate degradation on structured tables. Committed question sets, per-question run outputs, and a recomputation script let teams measure the table-text correctness gap and compare methods under a shared protocol."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
