{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.32034", "kind": "paper", "url": "https://arxiv.org/abs/2606.32034"}, {"id": "gh:bethgelab/qval", "kind": "repo", "url": "https://github.com/bethgelab/qval"}], "authors": ["Sergio Hernández-Gutiérrez", "Matteo Merler", "Ilze Amanda Auzina", "Joschka Strüber", "Ameya Prabhu", "Matthias Bethge"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.32034", "source_url": "https://arxiv.org/abs/2606.32034"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "QVal is a training-free testbed for evaluating dense supervision signals for long-horizon LLM agents. It measures Q-alignment of scores from methods across four environments and seven families, without requiring training runs."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.32034", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.32034", "title": "QVal"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Tübingen AI Center", "University of Tübingen", "Fondazione Bruno Kessler"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.32034", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.32034", "source_url": "https://openalex.org/W7166879316"}, "key": "claire-radar:2606.32034", "modality": null, "name": "QVal", "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_qval_38cf43e7", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "6e00a4d4a3a01912b36d8f04c1f35677ac21f0223eabc3930729d13a20de459c", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.32034"}, "publisher": null, "released": "2026-06-30", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.32034", "source_url": "https://arxiv.org/abs/2606.32034"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-32034", "source": "claire_radar", "source_benchmark_id": "2606.32034", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/bethgelab/qval", "githubScope": "benchmark_repo", "githubStars": 8, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 9, "hfPaperUrl": "https://huggingface.co/papers/2606.32034", "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": "2026-10-04", "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": "9cbee184baa666518cbe2891e72f220a38a9e4986850de93aedb917bf7ca30a5", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "QVal is a training-free testbed for evaluating dense supervision signals for long-horizon LLM agents. It measures Q-alignment of scores from methods across four environments and seven families, without requiring training runs.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence"], "snippet": "We introduce QVal, a training-free testbed for directly evaluating dense supervision signals."}, "familyId": "bmf_77df7a8b1bd4", "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/bethgelab/qval", "url": "https://github.com/bethgelab/qval"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/bethgelab/qval", "url": "https://huggingface.co/papers/2606.32034"}], "scope": "project_repo", "url": "https://github.com/bethgelab/qval", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_qval_38cf43e7", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/bethgelab/qval", "data": null, "hfPaper": "https://huggingface.co/papers/2606.32034", "paper": "https://arxiv.org/abs/2606.32034", "pdf": "https://arxiv.org/pdf/2606.32034", "project": null, "report": "https://arxiv.org/abs/2606.32034"}, "metrics": [], "motivation": "LLM agents increasingly act over long horizons, where a single trajectory can contain hundreds or thousands of actions.", "name": "QVal", "oneLine": "QVal is a training-free testbed for evaluating dense supervision signals for long-horizon LLM agents. It measures Q-alignment of scores from methods across four environments and seven families, without requiring training runs.", "paperTitle": "QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.32034", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "bethgelab", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/bethgelab/qval"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-30", "source": {"authors": ["Sergio Hern\\'andez-Guti\\'errez", "Matteo Merler", "Ilze Amanda Auzina", "Joschka Str\\\"uber", "Ameya Prabhu", "Matthias Bethge"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "id": "2606.32034", "title": "QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2606.32034"}, "sourceUpdatedAt": "2026-06-30T17:58:23Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "Fills the gap of direct, comparable evaluation of dense supervision methods, which is currently expensive and confounded. Enables early-stage assessment before training."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
