{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.27887", "kind": "paper", "url": "https://arxiv.org/abs/2605.27887"}, {"id": "gh:agenticfinlab/portbench", "kind": "repo", "url": "https://github.com/AgenticFinLab/portbench"}, {"kind": "website", "url": "https://portbench.github.io/"}], "authors": ["Yuxuan Zhao", "Sijia Chen", "Ningxin Su"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.27887", "source_url": "https://arxiv.org/abs/2605.27887"}, "categories": ["Finance", "Language & Knowledge", "cs.AI"], "description": {"en": "PortBench evaluates LLM-driven portfolio management via a static QA dataset (6,269 questions across seven task templates) and a dynamic five-stage allocation pipeline, spanning six asset classes over ten years. Scoring includes a dual-layer correlation score and CEPS, with evaluation under three stress regimes and investor profiles."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.27887", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.27887", "title": "PortBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2605.27887", "modality": null, "name": "PortBench", "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": "unknown", "origin_record_id": "bm_portbench_32129f61", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "530e2c5d41432c3acf79ee28493294aee924d90879e220d13f949b2a15f7072e", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.27887"}, "publisher": null, "released": "2026-05-27", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.27887", "source_url": "https://arxiv.org/abs/2605.27887"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-27887", "source": "claire_radar", "source_benchmark_id": "2605.27887", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Finance"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/AgenticFinLab/portbench", "githubScope": "benchmark_repo", "githubStars": 10, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2605.27887", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "b514623ab90b9974c54f11299d7b16c95e80565ce3ccbe1091959579bfd21376", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "PortBench evaluates LLM-driven portfolio management via a static QA dataset (6,269 questions across seven task templates) and a dynamic five-stage allocation pipeline, spanning six asset classes over ten years. Scoring includes a dual-layer correlation score and CEPS, with evaluation under three stress regimes and investor profiles.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "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 PortBench, a benchmark spanning six heterogeneous asset classes over ten years."}, "familyId": "bmf_1aa870031c3f", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/AgenticFinLab/portbench", "url": "https://github.com/AgenticFinLab/portbench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/AgenticFinLab/portbench", "url": "https://huggingface.co/papers/2605.27887"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/AgenticFinLab/portbench", "url": "https://portbench.github.io/"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/AgenticFinLab/portbench", "url": "https://arxiv.org/abs/2605.27887"}], "scope": "benchmark_repo", "url": "https://github.com/AgenticFinLab/portbench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_portbench_32129f61", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Financial Services"], "links": {"code": "https://github.com/AgenticFinLab/portbench", "data": null, "hfPaper": "https://huggingface.co/papers/2605.27887", "paper": "https://arxiv.org/abs/2605.27887", "pdf": "https://arxiv.org/pdf/2605.27887", "project": "https://portbench.github.io/", "report": "https://arxiv.org/abs/2605.27887"}, "metrics": [], "motivation": "Large language models (LLMs) have shown strong performance across diverse financial tasks, yet portfolio management (PM) remains poorly benchmarked.", "name": "PortBench", "oneLine": "PortBench evaluates LLM-driven portfolio management via a static QA dataset (6,269 questions across seven task templates) and a dynamic five-stage allocation pipeline, spanning six asset classes over ten years. Scoring includes a dual-layer correlation score and CEPS, with evaluation under three stress regimes and investor profiles.", "paperTitle": "PortBench: A Correlation-Aware, Full-Pipeline Benchmark for LLM-Driven Portfolio Management", "primaryDomain": "Finance", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.27887", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "AgenticFinLab", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/AgenticFinLab/portbench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-27", "source": {"authors": ["Yuxuan Zhao", "Sijia Chen", "Ningxin Su"], "categories": ["cs.AI", "q-fin.PM"], "id": "2605.27887", "title": "PortBench: A Correlation-Aware, Full-Pipeline Benchmark for LLM-Driven Portfolio Management", "type": "arxiv", "url": "https://arxiv.org/abs/2605.27887"}, "sourceUpdatedAt": "2026-08-10T10:05:45Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Existing financial benchmarks often ignore cross-asset correlations and the full portfolio management pipeline. PortBench addresses this gap by providing a comprehensive, reusable evaluation for LLM capabilities in realistic portfolio management, enabling comparison across models and informing practical deployment decisions."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
