{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.21830", "kind": "paper", "url": "https://arxiv.org/abs/2606.21830"}], "authors": ["Sarrah R. Mikhail Leung", "Taehan Kim", "Jeongbin Park"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.21830", "source_url": "https://arxiv.org/abs/2606.21830"}, "categories": ["General AI", "Language & Knowledge", "Reasoning"], "description": {"en": "Mat-Pref evaluates compositional reasoning in inorganic materials via 10,837 ionic-substitution questions across 11 structure families, with splits for in-distribution performance, held-out families, and cross-property transfer."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.21830", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.21830", "title": "Mat-Pref"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.21830", "modality": null, "name": "Mat-Pref", "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_mat-pref_c4647f88", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "1d0bf9b6c45e00f436b8dfbed64bcee6bc66c3df2ea8e039149b9173ade86a18", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.21830"}, "publisher": null, "released": "2026-06-20", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.21830", "source_url": "https://arxiv.org/abs/2606.21830"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-21830", "source": "claire_radar", "source_benchmark_id": "2606.21830", "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": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2606.21830", "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", "state": "unavailable"}}, "status": "unavailable"}, "capabilities": ["Reasoning"], "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": "3bac324eb215eab8e185a43d43fa52900eea9cd33d52800fdb374ce972795b22", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Mat-Pref evaluates compositional reasoning in inorganic materials via 10,837 ionic-substitution questions across 11 structure families, with splits for in-distribution performance, held-out families, and cross-property transfer.", "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 Mat-Pref, a benchmark of 10,837 ionic-substitution questions across 11 inorganic structure families, grounded in density functional theory calculations from the Materials Project, with three evaluation splits that isolate in-distribution performance, generalization to entirely held-out structure families, and cross-property transfer: applying band-gap reasoning to hosts seen during training only through formation-energy supervision."}, "familyId": "bmf_fe32cc058aa2", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_mat-pref_c4647f88", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.21830", "paper": "https://arxiv.org/abs/2606.21830", "pdf": "https://arxiv.org/pdf/2606.21830", "project": null, "report": "https://arxiv.org/abs/2606.21830"}, "metrics": [], "motivation": "Reinforcement learning from verifiable rewards (RLVR) has driven rapid progress in mathematical and code reasoning, but when extended to science, existing benchmarks do not decompose what generalizes: do gains reflect structural transfer, property transfer, or memorization?", "name": "Mat-Pref", "oneLine": "Mat-Pref evaluates compositional reasoning in inorganic materials via 10,837 ionic-substitution questions across 11 structure families, with splits for in-distribution performance, held-out families, and cross-property transfer.", "paperTitle": "Mat-Pref: Verifiable-Reward Training Improves Compositional Reasoning in Inorganic Materials", "primaryDomain": "General AI", "publication": {"evidence": "10 pages, 4 figures, Accepted at ICML AI4Physics 2026 Workshop", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.21830", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "ICML AI4Physics 2026 Workshop", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Mat-Pref team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2606.21830"}], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-20", "source": {"authors": ["Sarrah R. Mikhail Leung", "Taehan Kim", "Jeongbin Park"], "categories": ["cs.LG"], "id": "2606.21830", "title": "Mat-Pref: Verifiable-Reward Training Improves Compositional Reasoning in Inorganic Materials", "type": "arxiv", "url": "https://arxiv.org/abs/2606.21830"}, "sourceUpdatedAt": "2026-06-20T01:46:41Z", "topics": ["Reasoning"], "venueAttempts": [{"decisionRaw": "10 pages, 4 figures, Accepted at ICML AI4Physics 2026 Workshop", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "10 pages, 4 figures, Accepted at ICML AI4Physics 2026 Workshop", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.21830"}], "reviewStatus": "accepted", "venueName": "ICML AI4Physics 2026 Workshop"}], "whyItMatters": "It isolates generalization types (structural transfer, property transfer, memorization) in scientific reasoning, helping to identify when RLVR improves reasoning over memorization."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
