{
  "added_at": "2026-10-16",
  "arxiv_id": "2606.26094",
  "arxiv_version_reviewed": "v1",
  "authors": [
    "Babak Rahmani",
    "Sebastian Dziadzio",
    "Joschka Strüber",
    "Sergio Hernández-Gutiérrez",
    "Matthias Bethge"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-revengebench.bib",
  "code_url": "https://github.com/bethgelab/revenge-bench",
  "date_basis": "arXiv v1 submission history; the non-archival AIWILD workshop acceptance does not create a separate publication date.",
  "date_basis_zh": "采用 arXiv v1 的首次提交日期；AIWILD 非存档研讨会接收不构成另一出版日期。",
  "date_precision": "day",
  "description_en": "Tests whether agents can recover executable decision policies from behavioral traces and self-designed opponent probes, using 75 synthetic policies across five game arenas and measuring held-out action agreement and downstream competitive value.",
  "description_zh": "使用五种游戏环境中的 75 个合成策略，测试智能体能否通过行为轨迹和自行设计的对手探测恢复可执行决策策略，并衡量留出状态上的动作一致性及后续对抗价值。",
  "doi": "10.48550/arXiv.2606.26094",
  "entry_url": "https://awesomerea.org/papers/paper-revengebench/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-revengebench/",
  "evidence_urls": [
    "https://arxiv.org/abs/2606.26094",
    "https://arxiv.org/html/2606.26094v1",
    "https://github.com/bethgelab/revenge-bench",
    "https://github.com/bethgelab/revenge-bench/blob/090048696b4c7287e4cd05a728d4bba10fc8d421/README.md",
    "https://joschkastrueber.github.io/publications/",
    "https://openreview.net/forum?id=KTnZYPgaQz",
    "https://icml.cc/virtual/2026/workshop/54082",
    "https://agentwild-workshop.github.io/icml2026/"
  ],
  "first_arxiv_date": "2026-06-24",
  "id": "paper-revengebench",
  "json_url": "https://awesomerea.org/papers/paper-revengebench/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-revengebench/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-revengebench/index.md",
  "metadata_notes": "Accepted to the non-archival Second Workshop on Agents in the Wild (AIWILD) at ICML 2026: its OpenReview record was published on 2026-05-23 and is linked from the ICML page for the July 11, 2026 workshop. The workshop call states a non-archival policy, so this entry keeps the arXiv preprint and first-submission date. The repository and an author’s publication page also report the acceptance. The artifact is official.",
  "metadata_notes_zh": "已被 ICML 2026 的第二届 Agents in the Wild（AIWILD）非存档研讨会接收：其 OpenReview 记录于 2026-05-23 发布，并由 2026 年 7 月 11 日研讨会的 ICML 页面链接。研讨会征稿说明采用非存档政策，因此本条目保留 arXiv 预印本及首次提交日期。仓库与一位作者的发表列表也记载了该接收信息。产物属于官方项目。",
  "pdf_url": "https://arxiv.org/pdf/2606.26094v1",
  "project_url": "https://revengebench.com/",
  "publication_date": "2026-06-24",
  "publication_status": "preprint",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "The targets are fixed synthetic policies, not full games or adaptive adversaries. Agreement on finitely evaluated states does not uniquely identify source code, and some arenas have substantial variance. The public package omits some paper-only plotting and ablation suites; reproduction also needs CodeClash, a container runtime, and model access.",
  "scope_limitations_zh": "目标是固定的合成策略，不是完整游戏或自适应对手。有限评测状态上的动作一致性不能唯一确定源码，部分环境的方差较大。公开包未包含部分仅用于论文的绘图与消融实验；复现还需要 CodeClash、容器运行时和模型访问。",
  "targets": [
    "game"
  ],
  "title": "RevengeBench: Reverse Engineering Code-Space Policies from Behavioral Experiments",
  "title_zh": "RevengeBench：通过行为实验逆向恢复代码形式的策略",
  "url": "https://arxiv.org/abs/2606.26094",
  "venue": "arXiv",
  "verification": "source-reviewed",
  "verified_at": "2026-10-17",
  "why_fits_en": "Provides a game-domain benchmark for agent-assisted behavioral reverse engineering and active experimental design.",
  "why_fits_zh": "为智能体辅助的行为逆向工程与主动实验设计提供游戏领域基准。",
  "year": 2026
}