{
  "added_at": "2026-10-17",
  "arxiv_id": "2605.03546",
  "arxiv_version_reviewed": "v1",
  "authors": [
    "John Yang",
    "Kilian Lieret",
    "Jeffrey Ma",
    "Parth Thakkar",
    "Dmitrii Pedchenko",
    "Sten Sootla",
    "Emily McMilin",
    "Pengcheng Yin",
    "Rui Hou",
    "Gabriel Synnaeve",
    "Diyi Yang",
    "Ofir Press"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-programbench.bib",
  "code_url": "https://github.com/facebookresearch/ProgramBench",
  "date_basis": "The official NeurIPS 2026 poster listing verifies conference acceptance; no proceedings publication date is asserted.",
  "date_basis_zh": "NeurIPS 2026 官方海报列表确认会议接收；不推定论文集的出版日期。",
  "date_precision": "year",
  "description_en": "Evaluates black-box program reconstruction across 200 tasks: agents query an executable and read its documentation, then independently implement a codebase whose behavior is checked by generated tests.",
  "description_zh": "通过 200 项任务评估黑盒程序重建：智能体运行可执行文件进行探测并阅读文档，再独立实现代码库，由生成的测试检查其行为。",
  "doi": "10.48550/arXiv.2605.03546",
  "entry_url": "https://awesomerea.org/papers/paper-programbench/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-programbench/",
  "evidence_urls": [
    "https://arxiv.org/abs/2605.03546",
    "https://arxiv.org/html/2605.03546v1",
    "https://neurips.cc/virtual/2026/poster/139090",
    "https://neurips.cc/Downloads/2026",
    "https://dl4c.github.io/papers/",
    "https://dl4c.github.io/callforpapers/",
    "https://programbench.com/",
    "https://github.com/facebookresearch/ProgramBench/blob/895f65916bc5ba34a915d36da321fde0301125ad/README.md",
    "https://github.com/facebookresearch/ProgramBench/blob/895f65916bc5ba34a915d36da321fde0301125ad/docs/README.md",
    "https://github.com/facebookresearch/ProgramBench/blob/895f65916bc5ba34a915d36da321fde0301125ad/pyproject.toml",
    "https://github.com/facebookresearch/ProgramBench/blob/895f65916bc5ba34a915d36da321fde0301125ad/LICENSE",
    "https://huggingface.co/datasets/programbench/ProgramBench-Tests/blob/de0ddfb637590c7ecb54fa0b5301f6dc7dfbcee5/README.md"
  ],
  "first_arxiv_date": "2026-05-05",
  "id": "paper-programbench",
  "json_url": "https://awesomerea.org/papers/paper-programbench/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-programbench/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-programbench/index.md",
  "metadata_notes": "NeurIPS lists a December 2026 poster; this records acceptance, not proceedings publication. The July 2026 DL4C oral was a separate non-archival workshop appearance. The DOI identifies the arXiv preprint. The pinned harness is version 1.2.5 under MIT; its generated-test dataset preserves per-task attribution and upstream license terms, including copyleft obligations for derivative portions.",
  "metadata_notes_zh": "NeurIPS 列出 2026 年 12 月的海报报告；本条记录接收状态，不视为论文集已出版。2026 年 7 月的 DL4C 口头报告属于另一次非存档研讨会展示。DOI 指向 arXiv 预印本。固定版本的评测程序为 1.2.5，采用 MIT 许可；生成测试数据集保留逐任务署名与上游许可条款，包括衍生部分的 copyleft 义务。",
  "pdf_url": "https://arxiv.org/pdf/2605.03546v1",
  "project_url": "https://programbench.com/",
  "publication_date": null,
  "publication_status": "conference-accepted",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "Binary inspection, decompilation, and internet access are forbidden during inference. Passing finite tests does not prove complete behavioral equivalence or recovery of the original source. Later live-leaderboard runs are separate from the reviewed arXiv results.",
  "scope_limitations_zh": "推理阶段禁止检查二进制内容、反编译和访问互联网。通过有限测试不能证明完整行为等价或恢复了原始源码。实时榜单中的后续运行结果应与所审阅 arXiv 版本的结果区分。",
  "targets": [
    "binary",
    "app"
  ],
  "title": "ProgramBench: Can Language Models Rebuild Programs From Scratch?",
  "title_zh": "ProgramBench：语言模型能否从零重建程序？",
  "url": "https://arxiv.org/abs/2605.03546",
  "venue": "NeurIPS 2026",
  "verification": "source-reviewed",
  "verified_at": "2026-10-16",
  "why_fits_en": "Provides an executable benchmark for recovering software behavior through experiments, complementing benchmarks that permit static binary analysis.",
  "why_fits_zh": "为通过实验恢复软件行为提供可执行基准，与允许静态二进制分析的评测互为补充。",
  "year": 2026
}