{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.00148", "kind": "paper", "url": "https://arxiv.org/abs/2606.00148"}, {"id": "gh:hexixiang/stembind", "kind": "repo", "url": "https://github.com/hexixiang/StemBind"}, {"kind": "website", "url": "https://hexixiang.github.io/StemBind"}], "authors": ["Xixiang He", "Baiqi Wu", "Xingming Li", "Ao Cheng", "Qiyao Sun", "Xuanyu Ji", "Qingyong Hu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.00148", "source_url": "https://arxiv.org/abs/2606.00148"}, "categories": ["General AI", "Reasoning", "Vision & 3D"], "description": {"en": "StemBind is a diagnostic benchmark with shared-stem questions to attribute failures in abstract visual reasoning. It includes perception, rule, and full tasks with stage annotations but no official code or dataset release."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.00148", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.00148", "title": "StemBind"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["National University of Defense Technology", "Zhejiang University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.00148", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.00148", "source_url": "https://openalex.org/W7163192299"}, "key": "claire-radar:2606.00148", "modality": null, "name": "StemBind", "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_stembind_bdce22aa", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "a5a90ff981ebfaed1e4a941bec78edebcc57d9fdf35d4e789a958f1980e3195f", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.00148"}, "publisher": null, "released": "2026-05-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.00148", "source_url": "https://arxiv.org/abs/2606.00148"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-00148", "source": "claire_radar", "source_benchmark_id": "2606.00148", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/hexixiang/StemBind", "githubScope": "benchmark_repo", "githubStars": 3, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.00148", "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": ["Reasoning"], "confidence": "High", "construction": "Transform Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "36262cddaf9bdc858bb3fda3ebf8046138fb7d7e81302c6a1639a6ade468db6e", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "StemBind is a diagnostic benchmark with shared-stem questions to attribute failures in abstract visual reasoning. It includes perception, rule, and full tasks with stage annotations but no official code or dataset release.", "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", "public artifact URL"], "snippet": "We introduce StemBind, a shared-stem diagnostic benchmark that probes the same visual stem with three aligned questions: Perception (what is in the image), Rule (what pattern governs it), and Full (which option completes it), so a final-answer error can be attributed to a specific sub-step on the same evidence."}, "familyId": "bmf_140387d0197e", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 1, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "linked-from-project", "linkedUrl": "https://github.com/hexixiang/StemBind", "url": "https://hexixiang.github.io/StemBind"}], "scope": "benchmark_repo", "url": "https://github.com/hexixiang/StemBind", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_stembind_bdce22aa", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/hexixiang/StemBind", "data": null, "hfPaper": "https://huggingface.co/papers/2606.00148", "paper": "https://arxiv.org/abs/2606.00148", "pdf": "https://arxiv.org/pdf/2606.00148", "project": "https://hexixiang.github.io/StemBind", "report": "https://arxiv.org/abs/2606.00148"}, "metrics": [], "motivation": "Multimodal large language models (MLLMs) often know the rule but pick the wrong answer: on abstract visual reasoning (AVR) tasks, a model can describe what it sees and name the underlying pattern, yet still fail to choose the matching candidate.", "name": "StemBind", "oneLine": "StemBind is a diagnostic benchmark with shared-stem questions to attribute failures in abstract visual reasoning. It includes perception, rule, and full tasks with stage annotations but no official code or dataset release.", "paperTitle": "StemBind: When MLLMs Get Lost Between Rules and Instances in Abstract Visual Reasoning", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.00148", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-29", "source": {"authors": ["Xixiang He", "Baiqi Wu", "Xingming Li", "Ao Cheng", "Qiyao Sun", "Xuanyu Ji", "Qingyong Hu"], "categories": ["cs.CV", "cs.AI"], "id": "2606.00148", "title": "StemBind: When MLLMs Get Lost Between Rules and Instances in Abstract Visual Reasoning", "type": "arxiv", "url": "https://arxiv.org/abs/2606.00148"}, "sourceUpdatedAt": "2026-05-29T03:20:05Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "It aims to localize reasoning failures to specific sub-steps, potentially guiding improvements in multimodal models, but lacks a public path for independent verification."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
