{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.25485", "kind": "paper", "url": "https://arxiv.org/abs/2607.25485"}, {"id": "gh:amazon-science/patientagentbench", "kind": "repo", "url": "https://github.com/amazon-science/PatientAgentBench"}], "authors": ["Korosh Vatanparvar", "Ashutosh Joshi", "Maria Xenochristou", "Mohammad Abuzar Hashemi", "Prasad Kasu", "Deepak Bansal", "Daniel Lopez-Martinez", "Anchal Nema", "Ramya Ganesan", "Will Kimbrough", "Alex Woody", "Yadunandana Rao", "Dilek Hakkani-Tur", "Wilko Schulz-Mahlendorf"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.25485", "source_url": "https://arxiv.org/abs/2607.25485"}, "categories": ["Biology & Drug Discovery", "Language & Knowledge", "cs.AI"], "description": {"en": "A benchmark framework that evaluates patient-facing healthcare agents in simulated, tool-using conversations with patients, scored by an LLM jury against clinician-grounded criteria."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.25485", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.25485", "title": "PatientAgentBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.25485", "modality": null, "name": "PatientAgentBench", "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_patientagentbench_7194d3bc", "origin_source": "arxiv", "recognition_confidence": "0.75", "record_sha256": "8f10f71832854bb81fd4d918ca4495a03dbee905a73084989dd144d7a52eab0d", "relation": "introduces", "review_model": "claude-code-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T16:37:47.209552Z", "source_url": "https://arxiv.org/abs/2607.25485"}, "publisher": null, "released": "2026-07-28", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.25485", "source_url": "https://arxiv.org/abs/2607.25485"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-25485", "source": "claire_radar", "source_benchmark_id": "2607.25485", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Biology & Drug Discovery"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Healthcare agent evaluation is an active area, and the released code allows others to reuse it.", "score": 62}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T16:37:47.209552Z", "inputHash": "a79160dfa5ec66d185522e46f49b2efd8688459f5a78e4bb90a8caee12dc9633", "model": "claude-code-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "PatientAgentBench: A Benchmark Framework for Evaluating Patient-Facing Health AI Agents", "canonicalNameSource": "paper_title", "decisionReason": "Formally named benchmark with 1,200 scenarios, a defined six-dimension jury scoring rubric, and released code, applied across 10 models from four families.", "evaluationEvidence": {"reuse": "We release the framework as a reproducible, clinician-validated evaluation standard", "scoring": "Each conversation is scored by an LLM-as-a-Jury across six dimensions via over a hundred conversation-agnostic, clinician-grounded criteria", "task": "evaluates a foundation model, wrapped in an agent with a sandbox of healthcare tools, conversing with a simulated patient"}, "model": "claude-code-default", "reviewedAt": "2026-09-24T16:37:47.209552Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A benchmark framework that evaluates patient-facing healthcare agents in simulated, tool-using conversations with patients, scored by an LLM jury against clinician-grounded criteria.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "evaluation protocol evidence", "public artifact URL"], "snippet": "Health AI is evolving from answering questions to agentic systems that converse with patients, reason about health records, and act on their behalf."}, "familyId": "bmf_8c89b90f7901", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_patientagentbench_7194d3bc", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Pharma & Biotech"], "links": {"code": "https://github.com/amazon-science/PatientAgentBench", "data": null, "paper": "https://arxiv.org/abs/2607.25485", "pdf": "https://arxiv.org/pdf/2607.25485", "project": null, "report": "https://arxiv.org/abs/2607.25485"}, "metrics": [], "motivation": "Health AI is evolving from answering questions to agentic systems that converse with patients, reason about health records, and act on their behalf.", "name": "PatientAgentBench", "oneLine": "A benchmark framework that evaluates patient-facing healthcare agents in simulated, tool-using conversations with patients, scored by an LLM jury against clinician-grounded criteria.", "paperTitle": "PatientAgentBench: A Benchmark Framework for Evaluating Patient-Facing Health AI Agents", "primaryDomain": "Biology & Drug Discovery", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.25485", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "publishers": [{"name": "Amazon Science", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/amazon-science/PatientAgentBench"}], "readiness": "Runnable", "recognitionConfidence": 0.75, "relation": "introduces", "releasedAt": "2026-07-28", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Korosh Vatanparvar", "Ashutosh Joshi", "Maria Xenochristou", "Mohammad Abuzar Hashemi", "Prasad Kasu", "Deepak Bansal", "Daniel Lopez-Martinez", "Anchal Nema", "Ramya Ganesan", "Will Kimbrough", "Alex Woody", "Yadunandana Rao", "Dilek Hakkani-Tur", "Wilko Schulz-Mahlendorf"], "categories": ["cs.AI", "cs.CL"], "id": "2607.25485", "title": "PatientAgentBench: A Benchmark Framework for Evaluating Patient-Facing Health AI Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2607.25485"}, "sourceUpdatedAt": "2026-07-28T09:24:04Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "It targets multi-turn, tool-using patient interactions such as triage, clinical safety, and workflow accuracy, rather than static medical question answering, and it reports jury–clinician agreement."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
