{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.26094", "kind": "paper", "url": "https://arxiv.org/abs/2606.26094"}], "authors": ["Babak Rahmani", "Sebastian Dziadzio", "Joschka Strüber", "Sergio Hernández-Gutiérrez", "Matthias Bethge"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.26094", "source_url": "https://arxiv.org/abs/2606.26094"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "RevengeBench is a benchmark for recovering code-space policies from behavioral traces. It includes 75 LLM-generated policies across five game environments, where a learner designs behavioral probes and submits executable hypotheses, evaluated using continuous action-distance metrics."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.26094", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.26094", "title": "RevengeBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.26094", "modality": null, "name": "RevengeBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "Medium", "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_revengebench_23ac692d", "origin_source": "arxiv", "recognition_confidence": "0.85", "record_sha256": "7c9b3740baf32dc7cbda7f7be17e4f60a68ef9f13d6c8d3078ed1e37f8f87410", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.26094"}, "publisher": null, "released": "2026-06-24", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.26094", "source_url": "https://arxiv.org/abs/2606.26094"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-26094", "source": "claire_radar", "source_benchmark_id": "2606.26094", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General 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": 1, "hfPaperUrl": "https://huggingface.co/papers/2606.26094", "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", "lastSuccessfulDate": "2026-10-04", "state": "stale"}}, "status": "stale"}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "42b247ad81d5bd6e9df43d160eb22e5216533c64ab9133092e848e1d86399656", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "RevengeBench is a benchmark for recovering code-space policies from behavioral traces. It includes 75 LLM-generated policies across five game environments, where a learner designs behavioral probes and submits executable hypotheses, evaluated using continuous action-distance metrics.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence"], "snippet": "We introduce RevengeBench, a benchmark of 75 LLM generated, Elo-calibrated policies across five game environments, drawn from CodeClash tournament trajectories."}, "familyId": "bmf_928072ade2b4", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_revengebench_23ac692d", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.26094", "paper": "https://arxiv.org/abs/2606.26094", "pdf": "https://arxiv.org/pdf/2606.26094", "project": null, "report": "https://arxiv.org/abs/2606.26094"}, "metrics": [], "motivation": "For most of scientific history, researchers studying behavior could only infer hidden mechanisms from outward actions: an inverse problem that becomes more tractable when observation is augmented by targeted intervention.", "name": "RevengeBench", "oneLine": "RevengeBench is a benchmark for recovering code-space policies from behavioral traces. It includes 75 LLM-generated policies across five game environments, where a learner designs behavioral probes and submits executable hypotheses, evaluated using continuous action-distance metrics.", "paperTitle": "RevengeBench: Reverse Engineering Code-Space Policies from Behavioral Experiments", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.26094", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "arXiv", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2606.26094"}], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 0.85, "relation": "introduces", "releasedAt": "2026-06-24", "source": {"authors": ["Babak Rahmani", "Sebastian Dziadzio", "Joschka Str\\\"uber", "Sergio Hern\\'andez-Guti\\'errez", "Matthias Bethge"], "categories": ["cs.LG"], "id": "2606.26094", "title": "RevengeBench: Reverse Engineering Code-Space Policies from Behavioral Experiments", "type": "arxiv", "url": "https://arxiv.org/abs/2606.26094"}, "sourceUpdatedAt": "2026-06-24T17:59:46Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "Addresses the inverse problem of inferring hidden decision programs from observations, relevant to opponent modeling and policy interpretability. It provides a tractable testbed for studying how controlled experiments improve code-space recovery."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
