{"record": {"aliases": ["OOD-RL-Bench: A Benchmark Framework for Out-of-Distribution Detection in Reinforcement Learning"], "artifacts": [{"id": "arxiv:2607.12523", "kind": "paper", "url": "https://arxiv.org/abs/2607.12523"}, {"id": "gh:ood-rl-bench/ood-rl-bench", "kind": "repo", "url": "https://github.com/ood-rl-bench/ood-rl-bench"}], "categories": ["General AI", "Vision & 3D", "cs.LG"], "description": {"en": "OOD-RL-Bench is an extensible framework for testing out-of-distribution detectors on anomalies injected into RL trajectories. It scores detectors with matched-time AUROC, AUPRC, false-positive rate, and detection delay."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.12523", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.12523", "title": "OOD-RL-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.12523", "modality": null, "name": "OOD-RL-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_ood-rl-bench-a-benchmark-framework-for-out_c3fcfd5f", "origin_source": "arxiv", "recognition_confidence": "0.55", "record_sha256": "0e9fcd120073396e1c072065fdd20eacc81116af258fb26aaab735214a2bb609", "relation": "evaluates_only", "review_model": "claude-code-default", "review_state": "source-reviewed", "reviewed_at": "2026-09-25", "source_url": "https://arxiv.org/abs/2607.12523"}, "publisher": null, "released": "2026-07-14", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.12523", "source_url": "https://arxiv.org/abs/2607.12523"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-12523", "source": "claire_radar", "source_benchmark_id": "2607.12523", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["OOD-RL-Bench: A Benchmark Framework for Out-of-Distribution Detection in Reinforcement Learning"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "This is a niche safety-evaluation area, and the only demonstration is one environment.", "score": 35}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T20:39:58.972873Z", "inputHash": "5ae269ebec7216bc2e78331b701917c96230cd75af9e9d03bbdcd77ed9f7547e", "model": "claude-code-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "OOD-RL-Bench: A Benchmark Framework for Out-of-Distribution Detection in Reinforcement Learning", "canonicalNameSource": "paper_title", "decisionReason": "The benchmark is formally named, has defined anomaly-injection tasks and concrete metrics, and the paper states that the framework and results are publicly available. The only demonstration uses a single LunarLander-v3 environment.", "evaluationEvidence": {"reuse": "We make the framework, trained policy checkpoint, and complete results publicly available as a reproducible artefact.", "scoring": "matched-time AUROC, matched-time AUPRC, matched-time false-positive rate, detection delay, and segmented-onset metrics", "task": "evaluate OOD detectors against categories of anomalies injected into RL trajectories"}, "model": "claude-code-default", "reviewedAt": "2026-09-25", "sources": ["https://arxiv.org/abs/2607.12523", "https://github.com/ood-rl-bench/ood-rl-bench"], "state": "source-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "OOD-RL-Bench is an extensible framework for testing out-of-distribution detectors on anomalies injected into RL trajectories. It scores detectors with matched-time AUROC, AUPRC, false-positive rate, and detection delay.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "named entity is a model, method, framework, or dataset used in evaluation"], "snippet": "To address this gap, we present OOD-RL-Bench, a comprehensive and extensible framework designed to evaluate OOD detectors against categories of anomalies injected into RL trajectories."}, "familyId": "bmf_d413bbd12d76", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_ood-rl-bench-a-benchmark-framework-for-out_c3fcfd5f", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/ood-rl-bench/ood-rl-bench", "data": null, "paper": "https://arxiv.org/abs/2607.12523", "pdf": "https://arxiv.org/pdf/2607.12523", "project": null, "report": "https://arxiv.org/abs/2607.12523"}, "metrics": [], "motivation": "Reliable reinforcement learning (RL) agents must maintain operational integrity amidst sensor malfunctions, dynamic disturbances, and slow environmental shifts.", "name": "OOD-RL-Bench", "oneLine": "OOD-RL-Bench is an extensible framework for testing out-of-distribution detectors on anomalies injected into RL trajectories. It scores detectors with matched-time AUROC, AUPRC, false-positive rate, and detection delay.", "paperTitle": "OOD-RL-Bench: A Benchmark Framework for Out-of-Distribution Detection in Reinforcement Learning", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.12523", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Paper only", "recognitionConfidence": 0.55, "relation": "evaluates_only", "releasedAt": "2026-07-14", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Emil Mittag", "Richard Dazeley", "Peter Vamplew"], "categories": ["cs.LG", "cs.AI"], "id": "2607.12523", "title": "OOD-RL-Bench: A Benchmark Framework for Out-of-Distribution Detection in Reinforcement Learning", "type": "arxiv", "url": "https://arxiv.org/abs/2607.12523"}, "sourceUpdatedAt": "2026-07-14T08:57:04Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "Most OOD benchmarks evaluate image classifiers or static datasets. This one tests detection on temporal, action-dependent RL trajectories, which matters for deploying reliable agents."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
