{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.13020", "kind": "paper", "url": "https://arxiv.org/abs/2606.13020"}], "authors": ["Pierre Beckmann", "Marco Valentino", "Andre Freitas"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.13020", "source_url": "https://arxiv.org/abs/2606.13020"}, "categories": ["General AI", "Reasoning", "Vision & 3D"], "description": {"en": "SciR evaluates LLMs on deduction, induction, and causal abduction in scientific settings, with tasks generated from formal objects and rendered into multi-document scientific discourse. Difficulty is controlled along extraction and inference axes, with verifiable answers."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.13020", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.13020", "title": "SciR"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["École Polytechnique Fédérale de Lausanne", "Idiap Research Institute", "University of Sheffield"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.13020", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.13020", "source_url": "https://openalex.org/W7164502235"}, "key": "claire-radar:2606.13020", "modality": null, "name": "SciR", "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_scir_5305af40", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "7ed2c646bd758c03372622d926248e5b6e393f9b2bd24a23a94cb74a6c8ba72b", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.13020"}, "publisher": null, "released": "2026-06-11", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.13020", "source_url": "https://arxiv.org/abs/2606.13020"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-13020", "source": "claire_radar", "source_benchmark_id": "2606.13020", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Machine Generated", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.13020", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-01", "state": "stale"}}, "status": "stale"}, "capabilities": ["Reasoning"], "confidence": "High", "construction": "Original Synthetic", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "3e4c1dad6af2a8ed0bc0ad7f976f06ef9e4ea462214a544d574a00883e5f946d", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SciR evaluates LLMs on deduction, induction, and causal abduction in scientific settings, with tasks generated from formal objects and rendered into multi-document scientific discourse. Difficulty is controlled along extraction and inference axes, with verifiable answers.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce SciR, a benchmark that combines multi-paradigm reasoning with controllable scientific rendering, anchored on three paradigmatic scientific problems."}, "familyId": "bmf_fdd9b59047b7", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_scir_5305af40", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.13020", "paper": "https://arxiv.org/abs/2606.13020", "pdf": "https://arxiv.org/pdf/2606.13020", "project": null, "report": "https://arxiv.org/abs/2606.13020"}, "metrics": [], "motivation": "Three paradigmatic forms of inference recur across scientific reasoning: deduction, induction, and causal abduction.", "name": "SciR", "oneLine": "SciR evaluates LLMs on deduction, induction, and causal abduction in scientific settings, with tasks generated from formal objects and rendered into multi-document scientific discourse. Difficulty is controlled along extraction and inference axes, with verifiable answers.", "paperTitle": "SciR: A Controllable Benchmark for Scientific Reasoning in LLMs", "primaryDomain": "General AI", "publication": {"evidence": "Accepted at NeurIPS 2026 (Evaluations & Datasets track). v2: camera-ready version with corrected induction scoring and new analyses.\"", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.13020", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "NeurIPS 2026 (Evaluations & Datasets track)", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-11", "source": {"authors": ["Pierre Beckmann", "Marco Valentino", "Andre Freitas"], "categories": ["cs.AI"], "id": "2606.13020", "title": "SciR: A Controllable Benchmark for Scientific Reasoning in LLMs", "type": "arxiv", "url": "https://arxiv.org/abs/2606.13020"}, "sourceUpdatedAt": "2026-06-19T12:48:22Z", "topics": ["Reasoning"], "venueAttempts": [{"decisionRaw": "Accepted at NeurIPS 2026 (Evaluations & Datasets track). v2: camera-ready version with corrected induction scoring and new analyses.\"", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "Accepted at NeurIPS 2026 (Evaluations & Datasets track). v2: camera-ready version with corrected induction scoring and new analyses.\"", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.13020"}], "reviewStatus": "accepted", "venueName": "NeurIPS 2026 (Evaluations & Datasets track)"}], "whyItMatters": "Existing benchmarks either lack mechanistic ground truth or do not resemble real scientific documents. SciR provides a controllable protocol for isolating extraction vs. inference failures, which is valuable for diagnosing model capabilities in scientific reasoning."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
