{
  "added_at": "2026-10-20",
  "arxiv_id": "2606.30182",
  "arxiv_version_reviewed": "v2",
  "authors": [
    "Tom Adamczewski",
    "David Owen",
    "David Rein",
    "Florian Brand",
    "Giles Edkins",
    "Allen Hart",
    "Daniel O'Connell"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-mirrorcode.bib",
  "code_url": "https://github.com/epoch-research/MirrorCode",
  "date_basis": "First arXiv submission; the reviewed version is identified separately.",
  "date_basis_zh": "采用首次 arXiv 提交日期，另列所复核版本。",
  "date_precision": "day",
  "description_en": "Tests reimplementation of 25 command-line programs using execute-only reference access, documentation and visible examples. Separate scoring environments compare exact outputs on visible and held-out end-to-end tests.",
  "description_zh": "利用仅可执行的参考程序、文档及可见样例，评测 25 个命令行程序的重新实现；在独立评分环境中对可见及留出端到端测试进行精确输出比较。",
  "doi": "10.48550/arxiv.2606.30182",
  "entry_url": "https://awesomerea.org/papers/paper-mirrorcode/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-mirrorcode/",
  "evidence_urls": [
    "https://arxiv.org/abs/2606.30182",
    "https://arxiv.org/html/2606.30182v2",
    "https://api.datacite.org/dois/10.48550/arxiv.2606.30182",
    "https://github.com/epoch-research/MirrorCode/blob/5c9d7b00b0c6d7609003e33896cd41d28f092fcb/README.md",
    "https://github.com/epoch-research/MirrorCode/blob/5c9d7b00b0c6d7609003e33896cd41d28f092fcb/LICENSE"
  ],
  "first_arxiv_date": "2026-06-29",
  "id": "paper-mirrorcode",
  "json_url": "https://awesomerea.org/papers/paper-mirrorcode/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-mirrorcode/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-mirrorcode/index.md",
  "metadata_notes": "June 29 first submission, July 17 v2: a six-month-window candidate. Distinct from ProgramBench, explicitly compared in the paper. The pinned repository has an MIT root license and public harness/task definitions, excluding private tasks; target-program licensing is separate.",
  "metadata_notes_zh": "首投为 6 月 29 日，v2 为 7 月 17 日，属于近六个月补充候选。与 ProgramBench 不同，论文专门比较两者。固定仓库的根许可证为 MIT，包含公开框架与任务定义，不含私有任务；目标程序许可另行适用。",
  "pdf_url": "https://arxiv.org/pdf/2606.30182v2",
  "project_url": null,
  "publication_date": "2026-06-29",
  "publication_status": "preprint",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "Targets are scoped, deterministic, stateless programs with text output. Test agreement is not universal equivalence or code-quality assessment. A memorization screen flags possible exposure for 17 targets. Large inference budgets and scaffold choices limit comparisons with cheaper benchmarks; three targets remain private.",
  "scope_limitations_zh": "目标限于指定范围内、确定性、无状态且输出文本的程序。测试一致不等于普遍等价或代码质量评估；记忆探测发现 17 个目标可能已被接触。较大推理预算及框架选择限制与低成本基准的比较，三个目标仍未公开。",
  "targets": [
    "binary",
    "app"
  ],
  "title": "MirrorCode: AI can rebuild entire programs from behavior alone",
  "title_zh": "MirrorCode：仅凭行为重建完整程序的评测",
  "url": "https://arxiv.org/abs/2606.30182",
  "venue": "arXiv",
  "verification": "source-reviewed",
  "verified_at": "2026-10-20",
  "why_fits_en": "Directly evaluates sustained reconstruction from observable software behavior.",
  "why_fits_zh": "直接评估依据可观察的软件行为开展长期重建的能力。",
  "year": 2026
}