{
  "added_at": "2026-10-20",
  "arxiv_id": "2609.05370",
  "arxiv_version_reviewed": "v1",
  "authors": [
    "Chang Liu",
    "Edward Raff",
    "Kristopher Micinski"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-decompile-diverge.bib",
  "code_url": null,
  "date_basis": "First arXiv submission date from the reviewed submission history; no formal publication was verified.",
  "date_basis_zh": "采用已核对的 arXiv 首次提交日期；尚未核实正式出版信息。",
  "date_precision": "day",
  "description_en": "Decompile-Diverge builds function-level behavioral comparisons from source-derived drivers and fuzzed reference inputs. The study compares reconstructed functions on the same inputs and examines divergences and missing crashes that compilation and supplied tests can miss.",
  "description_zh": "Decompile-Diverge 基于源码生成驱动，并对参考实现进行模糊测试，以构建函数级行为比较。研究让重建函数使用相同输入，检查编译及既有测试可能漏掉的行为差异与崩溃消失。",
  "doi": null,
  "entry_url": "https://awesomerea.org/papers/paper-decompile-diverge/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-decompile-diverge/",
  "evidence_urls": [
    "https://arxiv.org/abs/2609.05370v1",
    "https://arxiv.org/html/2609.05370v1"
  ],
  "first_arxiv_date": "2026-09-04",
  "id": "paper-decompile-diverge",
  "json_url": "https://awesomerea.org/papers/paper-decompile-diverge/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-decompile-diverge/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-decompile-diverge/index.md",
  "metadata_notes": "Latest reviewed version is arXiv v1, licensed CC BY 4.0. No author-linked code or dataset URL was verified in the reviewed paper and focused artifact search; availability and artifact licenses remain unverified. Reported measurements.",
  "metadata_notes_zh": "最新核对版本为 arXiv v1，文章采用 CC BY 4.0。已读论文及定向产物检索未核实作者代码或数据集链接；产物可用性与许可仍未确定。文中测量结果。",
  "pdf_url": "https://arxiv.org/pdf/2609.05370v1",
  "project_url": null,
  "publication_date": "2026-09-04",
  "publication_status": "preprint",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "The oracle requires reference source, supported inputs and buildable harnesses. It compares a finite corpus and bounded observable state; Matched does not prove all-input equivalence. Nine of 300 GitHub reference functions failed harness builds, leaving 291. Some types/functions are excluded. DeGPT-Qwen is a substituted-backend configuration, not the originally published DeGPT model.",
  "scope_limitations_zh": "该判定器需要参考源码、受支持输入及可构建的测试框架。比较仅覆盖有限输入语料和有界可观测状态；Matched 不代表所有输入上的等价。300 个 GitHub 参考函数中九个无法构建，最终保留 291 个；部分类型与函数被排除。DeGPT-Qwen 使用替换后的后端，并非原始发表的 DeGPT 模型。",
  "targets": [
    "binary"
  ],
  "title": "When LLM Decompilers Recompile More and Preserve Less",
  "title_zh": "当 LLM 反编译器更易重编译，却更少保留原始行为",
  "url": "https://arxiv.org/abs/2609.05370",
  "venue": "arXiv",
  "verification": "source-reviewed",
  "verified_at": "2026-10-20",
  "why_fits_en": "Offers a concrete check on what recompilation and benchmark pass rates do not establish about recovered code.",
  "why_fits_zh": "具体说明重编译及基准通过率无法证明恢复代码的哪些行为属性。",
  "year": 2026
}