{"record": {"aliases": ["Internalizing Geometric Law: Learning from Solver Residuals for Precision-Critical Generation"], "artifacts": [{"id": "arxiv:2606.09278", "kind": "paper", "url": "https://arxiv.org/abs/2606.09278"}, {"id": "gh:huawei-ai4math/pygeox", "kind": "repo", "url": "https://github.com/Huawei-AI4Math/PyGeoX"}], "authors": ["Rafael Cabral", "Pang Zixi", "Ziyi Shou", "Shen Xin"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.09278", "source_url": "https://arxiv.org/abs/2606.09278"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "A 300-problem stratified benchmark for generating precise geometric constructions from natural-language descriptions."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.09278", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.09278", "title": "PyGeoX-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Huawei Technologies (China)", "Huawei Technologies (United Kingdom)"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.09278", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.09278", "source_url": "https://openalex.org/W7164012504"}, "key": "claire-radar:2606.09278", "modality": null, "name": "PyGeoX-Bench", "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_internalizing-geometric-law-learning-from-_a1ab8873", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "c1caf405057157aea8e4cd3f925b17e87b4547dfea971f2e418cb68300350596", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:05:39.521708Z", "source_url": "https://arxiv.org/abs/2606.09278"}, "publisher": null, "released": "2026-06-08", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.09278", "source_url": "https://arxiv.org/abs/2606.09278"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-09278", "source": "claire_radar", "source_benchmark_id": "2606.09278", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Internalizing Geometric Law: Learning from Solver Residuals for Precision-Critical Generation"], "annotation": "Mixed", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "Verifiable rewards and precision-critical geometric generation may receive broad near-term attention across reasoning and RL research.", "score": 4}, "capabilities": [], "confidence": "Medium", "construction": "Aggregate Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:05:39.521708Z", "inputHash": "b4097e833f096d71c473cc59795114c3e164ca8ab02f364fa2ff57e9e48938c8", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "PyGeoX-Bench, a stratified suite of 300 problems", "canonicalNameSource": "abstract", "decisionReason": "The abstract formally names the suite, specifies its task and solving-rate evaluation, and states that the benchmark, engine, and data are publicly released.", "evaluationEvidence": {"reuse": "We release the engine, benchmark, and data at https://github.com/Huawei-AI4Math/PyGeoX.", "scoring": "hard-tier solving rate", "task": "open-ended geometric synthesis from natural language"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:05:39.521708Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A 300-problem stratified benchmark for generating precise geometric constructions from natural-language descriptions.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "To make this tractable, we release PyGeoX, a programmable geometric DSL that compiles declarative constraints into a differentiable loss, and PyGeoX-Bench, a stratified suite of 300 problems with per-constraint verifiable rewards."}, "familyId": "bmf_12f0c4ecdf4a", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_internalizing-geometric-law-learning-from-_a1ab8873", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/Huawei-AI4Math/PyGeoX", "data": null, "paper": "https://arxiv.org/abs/2606.09278", "pdf": "https://arxiv.org/pdf/2606.09278", "project": null, "report": "https://arxiv.org/abs/2606.09278"}, "metrics": [], "motivation": "Large Language Models frequently hallucinate in precision-critical domains such as technical diagramming and mechanical design, where outputs must satisfy strict geometric constraints.", "name": "PyGeoX-Bench", "oneLine": "A 300-problem stratified benchmark for generating precise geometric constructions from natural-language descriptions.", "paperTitle": "Internalizing Geometric Law: Learning from Solver Residuals for Precision-Critical Generation", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.09278", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "publishers": [{"name": "Huawei AI4Math", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/Huawei-AI4Math/PyGeoX"}], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-06-08", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Rafael Cabral", "Pang Zixi", "Ziyi Shou", "Shen Xin"], "categories": ["cs.LG", "cs.AI"], "id": "2606.09278", "title": "Internalizing Geometric Law: Learning from Solver Residuals for Precision-Critical Generation", "type": "arxiv", "url": "https://arxiv.org/abs/2606.09278"}, "sourceUpdatedAt": "2026-06-08T09:44:31Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "It tests whether generated constructions satisfy interacting geometric constraints using verifiable, per-constraint rewards."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
