{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2610.04691", "kind": "paper", "url": "https://arxiv.org/abs/2610.04691"}], "authors": ["Shiqi Yang", "Jiekai Ma", "Gaoyuan Du"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2610.04691", "source_url": "https://arxiv.org/abs/2610.04691"}, "categories": ["Factuality", "General AI", "Information retrieval", "Language & Knowledge", "cs.AI"], "description": {"en": "Retrieval-Augmented Generation (RAG) is typically evaluated under the implicit assumption that the underlying knowledge base (KB) is clean, leaving the behaviour of RAG systems under realistic KB degradation poorly characterised."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2610.04691", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2610.04691", "title": "RAGStress"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2610.04691", "modality": null, "name": "RAGStress", "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_ragstress_3c0fa068", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "10df0dd13439ed717f3ff544aa3ecfabb05c700925e58f6c8ac0030a1bde2e4f", "relation": "introduces", "review_state": "rule-auto-admitted", "reviewed_at": "2026-10-06T07:49:35.626614Z", "source_url": "https://arxiv.org/abs/2610.04691"}, "publisher": null, "released": "2026-10-03", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2610.04691", "source_url": "https://arxiv.org/abs/2610.04691"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2610-04691", "source": "claire_radar", "source_benchmark_id": "2610.04691", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "capabilities": ["Information retrieval", "Factuality"], "confidence": "High", "construction": "Transform Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "curation": {"decisionReason": "Automatically admitted from an explicit named benchmark release with evaluation-protocol evidence at or above the configured publication-confidence threshold.", "policy": "source-evidence-v2", "reviewedAt": "2026-10-06T07:49:35.626614Z", "state": "rule-auto-admitted"}, "dataStatus": "primary-source-indexed", "demo": false, "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce RAGStress, a controlled evaluation benchmark for stress-testing RAG systems under systematic KB corruption."}, "familyId": "bmf_bb2dd8dfc977", "firstSeenAt": "2026-10-06", "heat": null, "id": "bm_ragstress_3c0fa068", "indexedAt": "2026-10-06T07:49:11.852971Z", "industrySectors": [], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2610.04691", "pdf": "https://arxiv.org/pdf/2610.04691", "project": null, "report": "https://arxiv.org/abs/2610.04691"}, "metrics": [], "motivation": "Retrieval-Augmented Generation (RAG) is typically evaluated under the implicit assumption that the underlying knowledge base (KB) is clean, leaving the behaviour of RAG systems under realistic KB degradation poorly characterised.", "name": "RAGStress", "oneLine": "Retrieval-Augmented Generation (RAG) is typically evaluated under the implicit assumption that the underlying knowledge base (KB) is clean, leaving the behaviour of RAG systems under realistic KB degradation poorly characterised.", "paperTitle": "RAGStress: A controlled benchmark for evaluating retrieval-augmented generation under knowledge-base degradation", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2610.04691", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-10-03", "source": {"authors": ["Shiqi Yang", "Jiekai Ma", "Gaoyuan Du"], "categories": ["cs.AI"], "id": "2610.04691", "title": "RAGStress: A controlled benchmark for evaluating retrieval-augmented generation under knowledge-base degradation", "type": "arxiv", "url": "https://arxiv.org/abs/2610.04691"}, "sourceUpdatedAt": "2026-10-03T18:07:17Z", "topics": ["cs.AI"], "venueAttempts": []}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
