{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.29360", "kind": "paper", "url": "https://arxiv.org/abs/2605.29360"}], "authors": ["Tianzhuo Yang", "Zihan Shen", "Zirui Mi", "Zhaoyi Zhang", "Jiayi Zhou", "Jiaming Ji", "Juntao Dai", "Jiawei Chen", "Boyuan Chen", "Yaodong Yang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.29360", "source_url": "https://arxiv.org/abs/2605.29360"}, "categories": ["Language & Knowledge", "Robotics & Embodied AI", "cs.AI"], "description": {"en": "MiraBench evaluates action-conditioned reliability in robotic world models through three levels: Physics Adherence, Action-Following Fidelity, and Optimism Bias Detection, using a human-annotated corpus of over 16,000 judgments."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.29360", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.29360", "title": "MiraBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2605.29360", "modality": null, "name": "MiraBench", "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_mirabench_80851376", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "6f856702f946e61a6616548a1e2e9e31aef83725edf272cbd1c144648ee206a6", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.29360"}, "publisher": null, "released": "2026-05-28", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.29360", "source_url": "https://arxiv.org/abs/2605.29360"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-29360", "source": "claire_radar", "source_benchmark_id": "2605.29360", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Robotics & Embodied 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/2605.29360", "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": [], "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": "6dc4925abdaa7ae18e4c4cd4f0d0499171b6bf8baddf6dac5dcdc0439e8bbf69", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "MiraBench evaluates action-conditioned reliability in robotic world models through three levels: Physics Adherence, Action-Following Fidelity, and Optimism Bias Detection, using a human-annotated corpus of over 16,000 judgments.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce \\textsc{MiraBench}, a hierarchical benchmark that defines \\emph{action-conditioned reliability} as a core evaluation target for robotic world models."}, "familyId": "bmf_46ceeefc922f", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_mirabench_80851376", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Robotics"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2605.29360", "paper": "https://arxiv.org/abs/2605.29360", "pdf": "https://arxiv.org/pdf/2605.29360", "project": null, "report": "https://arxiv.org/abs/2605.29360"}, "metrics": [], "motivation": "Action-conditioned world models are increasingly used as scalable simulators for robot learning, yet current evaluations provide limited evidence that their predictions are reliable under the actions they condition on.", "name": "MiraBench", "oneLine": "MiraBench evaluates action-conditioned reliability in robotic world models through three levels: Physics Adherence, Action-Following Fidelity, and Optimism Bias Detection, using a human-annotated corpus of over 16,000 judgments.", "paperTitle": "MiraBench: Evaluating Action-Conditioned Reliability in Robotic World Models", "primaryDomain": "Robotics & Embodied AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.29360", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-28", "source": {"authors": ["Tianzhuo Yang", "Zihan Shen", "Zirui Mi", "Zhaoyi Zhang", "Jiayi Zhou", "Jiaming Ji", "Juntao Dai", "Jiawei Chen", "Boyuan Chen", "and Yaodong Yang"], "categories": ["cs.AI"], "id": "2605.29360", "title": "MiraBench: Evaluating Action-Conditioned Reliability in Robotic World Models", "type": "arxiv", "url": "https://arxiv.org/abs/2605.29360"}, "sourceUpdatedAt": "2026-05-28T04:58:15Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Existing benchmarks focus on visual fidelity, leaving unclear whether predicted futures are physically plausible, faithful to actions, and calibrated to failure. MiraBench aims to provide a diagnostic foundation for assessing world models as simulators, but its public evaluation protocol and release details are not yet specified."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
