{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.02449", "kind": "paper", "url": "https://arxiv.org/abs/2606.02449"}, {"id": "gh:xinhaos0101/hll", "kind": "repo", "url": "https://github.com/XinhaoS0101/HLL"}], "authors": ["Xinhao Song", "Su Su", "Sirui Song", "Hongliang Wu", "Wen Shen", "Zhihua Wei", "Gongshen Liu", "Linfeng Zhang", "Dongrui Liu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.02449", "source_url": "https://arxiv.org/abs/2606.02449"}, "categories": ["Agents & Tool Use", "General AI", "cs.AI"], "description": {"en": "HLL is a benchmark that evaluates multimodal agents on interactive CAPTCHA verification in a closed-loop GUI environment, covering diverse task types such as text transcription, image selection, sliders, jigsaw puzzles, and logic-based challenges. Scoring is based on task completion and trace-conditioned validation."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.02449", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.02449", "title": "HLL"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Shanghai Jiao Tong University", "Shandong University", "Tongji University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.02449", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.02449", "source_url": "https://openalex.org/W7163282726"}, "key": "claire-radar:2606.02449", "modality": null, "name": "HLL", "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_hll_52c829e2", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "a267bd87b8ef79465894acb40973702b4bc92cf8d3caf2dc272e23ba5137b77c", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.02449"}, "publisher": null, "released": "2026-06-01", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.02449", "source_url": "https://arxiv.org/abs/2606.02449"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-02449", "source": "claire_radar", "source_benchmark_id": "2606.02449", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/XinhaoS0101/HLL", "githubScope": "benchmark_repo", "githubStars": 3, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.02449", "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": [], "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": "3d31bcb2387de283f6c8469fb538a05a6a19fbf95c8790d3508dab3d7df214c1", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "HLL is a benchmark that evaluates multimodal agents on interactive CAPTCHA verification in a closed-loop GUI environment, covering diverse task types such as text transcription, image selection, sliders, jigsaw puzzles, and logic-based challenges. Scoring is based on task completion and trace-conditioned validation.", "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 \\textbf{Humanity's Last Line of Verification (HLL)}, a controlled benchmark that uses interactive CAPTCHA verification to evaluate whether agents can cross this boundary through grounded, human-like interaction rather than recognition alone."}, "familyId": "bmf_f97ab7d41b89", "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/XinhaoS0101/HLL", "url": "https://github.com/XinhaoS0101/HLL"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/XinhaoS0101/HLL", "url": "https://huggingface.co/papers/2606.02449"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/XinhaoS0101/HLL", "url": "https://arxiv.org/abs/2606.02449"}], "scope": "project_repo", "url": "https://github.com/XinhaoS0101/HLL", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_hll_52c829e2", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/XinhaoS0101/HLL", "data": null, "hfPaper": "https://huggingface.co/papers/2606.02449", "paper": "https://arxiv.org/abs/2606.02449", "pdf": "https://arxiv.org/pdf/2606.02449", "project": null, "report": "https://arxiv.org/abs/2606.02449"}, "metrics": [], "motivation": "Multimodal agents are increasingly expected to operate interfaces on behalf of users, raising a central deployment question: can they truly substitute for humans in workflows that services deliberately protect against automation?", "name": "HLL", "oneLine": "HLL is a benchmark that evaluates multimodal agents on interactive CAPTCHA verification in a closed-loop GUI environment, covering diverse task types such as text transcription, image selection, sliders, jigsaw puzzles, and logic-based challenges. Scoring is based on task completion and trace-conditioned validation.", "paperTitle": "HLL: Can Agents Cross Humanity's Last Line of Verification?", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.02449", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Xinhao S0101", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/XinhaoS0101/HLL"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-01", "source": {"authors": ["Xinhao Song", "Su Su", "Sirui Song", "Hongliang Wu", "Wen Shen", "Zhihua Wei", "Gongshen Liu", "Linfeng Zhang", "Dongrui Liu"], "categories": ["cs.AI", "cs.CL", "cs.CV", "cs.LG", "cs.MM"], "id": "2606.02449", "title": "HLL: Can Agents Cross Humanity's Last Line of Verification?", "type": "arxiv", "url": "https://arxiv.org/abs/2606.02449"}, "sourceUpdatedAt": "2026-06-01T16:20:45Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "HLL addresses the gap in measuring agent capability at automation-protected workflows, providing a testbed for comparing progress in human-like interaction and process consistency."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
