{
  "added_at": "2026-10-17",
  "arxiv_id": "2609.18805",
  "arxiv_version_reviewed": "v1",
  "authors": [
    "Jeonghye Kim",
    "Minseon Kim",
    "Young Jin Kim",
    "Matheus Pereira",
    "Marc-Alexandre Côté",
    "Alessandro Sordoni",
    "Xingdi Yuan",
    "Zhengyan Shi"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-programdistill.bib",
  "code_url": "https://huggingface.co/datasets/microsoft/ProgramDistill",
  "date_basis": "arXiv v1 submission history; the earlier author blog date is separate.",
  "date_basis_zh": "采用 arXiv v1 的首次提交日期；作者博客的较早发布日期另行区分。",
  "date_precision": "day",
  "description_en": "Evaluates recovery of web-app behavior from a live reference with hidden source. It provides 4,063 feature-restoration tasks across 26 applications and a separate 12-application reconstruction setting, scored by replaying dependent browser workflows.",
  "description_zh": "评测智能体从源码不可见的运行参考中恢复 Web 应用行为的能力，提供覆盖 26 个应用的 4,063 项功能恢复任务，以及独立的 12 应用重建设置，通过重放具有依赖关系的浏览器工作流评分。",
  "doi": "10.48550/arXiv.2609.18805",
  "entry_url": "https://awesomerea.org/papers/paper-programdistill/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-programdistill/",
  "evidence_urls": [
    "https://arxiv.org/abs/2609.18805",
    "https://arxiv.org/html/2609.18805v1",
    "https://microsoft.github.io/debug-gym/blog/2026/09/programdistill/",
    "https://huggingface.co/datasets/microsoft/ProgramDistill/blob/dc782f7003ae00738260996d02d77920d08a5db1/README.md",
    "https://huggingface.co/datasets/microsoft/ProgramDistill/blob/dc782f7003ae00738260996d02d77920d08a5db1/tasks/full-reconstruction/README.md",
    "https://huggingface.co/datasets/microsoft/ProgramDistill/blob/dc782f7003ae00738260996d02d77920d08a5db1/tasks/full-reconstruction/baserow_baserow/tests/scorer/trace_schema.py",
    "https://huggingface.co/datasets/microsoft/ProgramDistill/blob/dc782f7003ae00738260996d02d77920d08a5db1/tasks/full-reconstruction/baserow_baserow/tests/scorer/trace_replay_validator.py",
    "https://api.datacite.org/dois/10.48550/arXiv.2609.18805",
    "https://openreview.net/forum?id=FbKJFN0R5L",
    "https://github.com/iab-agents/iab-agents.github.io/blob/0437015a568d8d541e277833c68ffd213b117c53/src/App.jsx",
    "https://github.com/iab-agents/iab-agents.github.io/blob/0437015a568d8d541e277833c68ffd213b117c53/src/data/siteData.js"
  ],
  "first_arxiv_date": "2026-09-16",
  "id": "paper-programdistill",
  "json_url": "https://awesomerea.org/papers/paper-programdistill/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-programdistill/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-programdistill/index.md",
  "metadata_notes": "The DOI resolves and DataCite records registration on 2026-09-17, despite arXiv’s stale pending banner. Separately, the paper was accepted to the non-archival Workshop on Interpreting Agent Behavior (IAB) at NeurIPS 2026, scheduled for December 12, 2026; its OpenReview record was published on 2026-10-04. That record and the workshop listing put Xingdi Yuan before Alessandro Sordoni; the authors here follow arXiv v1. The public release contains executable benchmark packages, not the mining, task-generation or orchestration pipeline. ProgramDistill-authored artifacts declare MIT; included applications and assets retain upstream terms.",
  "metadata_notes_zh": "DOI 已可解析，DataCite 记载其于 2026-09-17 注册，尽管 arXiv 仍显示过时的待注册提示。此外，该论文已被 NeurIPS 2026 的 Interpreting Agent Behavior（IAB）非存档研讨会接收，研讨会定于 2026 年 12 月 12 日举行；其 OpenReview 记录于 2026-10-04 发布。该记录与研讨会论文列表将 Xingdi Yuan 排在 Alessandro Sordoni 之前，本条目作者顺序以 arXiv v1 为准。公开版本包含可执行基准任务包，但不包含行为挖掘、任务生成或编排流水线。ProgramDistill 自有产物声明 MIT 许可，所含应用与资源仍适用上游条款。",
  "pdf_url": "https://arxiv.org/pdf/2609.18805v1",
  "project_url": "https://microsoft.github.io/debug-gym/blog/2026/09/programdistill/",
  "publication_date": "2026-09-16",
  "publication_status": "preprint",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "Tests cover finite, model-mined behaviors in self-contained apps, including SaaS clones. Persistence failures can be detected through replayed workflows, but passing does not establish all backend or database invariants. Evaluated agents receive structured browser observations without screenshots; visual fidelity is not measured. Public verifiers and reference solutions are not a contamination-resistant test set.",
  "scope_limitations_zh": "测试限于自包含应用（包括 SaaS 克隆）中由模型挖掘的有限行为。重放工作流可发现持久化失败，但通过测试不能证明全部后端或数据库不变量成立。被评测智能体接收不含截图的结构化浏览器观察，未衡量视觉保真度。公开验证器与参考解答也不构成抗污染测试集。",
  "targets": [
    "website"
  ],
  "title": "ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks",
  "title_zh": "ProgramDistill：将交互式 Web 应用转化为可验证的参考驱动软件工程任务",
  "url": "https://arxiv.org/abs/2609.18805",
  "venue": "arXiv",
  "verification": "source-reviewed",
  "verified_at": "2026-10-17",
  "why_fits_en": "Directly studies reference-guided behavioral reconstruction beyond screenshot-to-HTML, including dependent state changes and fresh verification of submitted implementations.",
  "why_fits_zh": "直接研究超越截图到 HTML 的参考驱动行为重建，涵盖有依赖关系的状态变化，并在全新验证环境中检查提交的实现。",
  "year": 2026
}