{
  "added_at": "2026-11-02",
  "additional_evidence_urls": [
    "https://github.com/morluto/rea/blob/e8da7e1edd075feec0e6beedec93d5645fed8125/website/public/blog/reverse-engineering-models/index.html",
    "https://github.com/N0zoM1z0/th08/commit/a393f400786e4cc6cbb1000af0dad3276da12c78"
  ],
  "category": "official-case-studies",
  "description_en": "An essay on REA’s site by N0zoM1z0 (13 October 2026) uses dated Touhou reconstruction commits to ask whether exact function matching needs a larger general model, a compiler-specialist model or better retrieval. It walks through TH09, TH095, TH04 and TH08 cases where later GPT-6-series sessions closed stubborn matches, and the TH08 audit that found accepted functions with wrong floating constants. Specialist and reinforcement-learning comparisons are proposed as future experiments; no trained specialist was evaluated.",
  "description_zh": "REA 网站上 N0zoM1z0 的一篇文章（2026 年 10 月 13 日）依据注明日期的东方 Project 重建提交，探讨精确的函数匹配究竟需要更大的通用模型、针对编译器的专用模型，还是更好的检索。文章逐一介绍 TH09、TH095、TH04 与 TH08 中较晚的 GPT-6 系列会话攻克难以匹配函数的案例，以及 TH08 一次审计发现部分已接受函数含有错误浮点常量的经过。专用模型与强化学习的对比被作为未来实验提出，并未评估任何经过训练的专用模型。",
  "entry_url": "https://awesomerea.org/resources/rea-bigger-model-essay-case/",
  "entry_url_zh": "https://awesomerea.org/zh/resources/rea-bigger-model-essay-case/",
  "evidence_url": "https://github.com/morluto/rea/blob/e8da7e1edd075feec0e6beedec93d5645fed8125/website/evidence/reverse-engineering-models.md",
  "group": "rea",
  "id": "rea-bigger-model-essay-case",
  "json_url": "https://awesomerea.org/resources/rea-bigger-model-essay-case/record.json",
  "kind": "case-study",
  "markdown_url": "https://awesomerea.org/resources/rea-bigger-model-essay-case/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/resources/rea-bigger-model-essay-case/index.md",
  "name": "Does reverse engineering need a bigger model? General and specialist models in Touhou reconstruction",
  "name_zh": "逆向工程需要更大的模型吗？东方 Project 重建中的通用模型与专用模型",
  "reconstruction_method": "other",
  "record_type": "resource",
  "relationship": "official-case-study",
  "requirement_facets": [
    "prerequisites",
    "scope",
    "limits",
    "testing",
    "license"
  ],
  "requirements": [
    "The essay and its notes supply no game files, compilers or builds; each linked Touhou repository documents its own requirements.",
    "The essay is dated 13 October 2026. Its examples are function-level comparisons recorded in linked TH09, TH095, TH04 and TH08 commits.",
    "Later sessions inherited earlier source, diagnostics and failed experiments, so the before-and-after closures are not controlled estimates of model-size effects. The notes treat gpt-web as a changing label that does not identify a run’s model, per-run reasoning effort and budgets are not uniformly recorded, and the GPT-6-Luna-max closure is presented as a discovery across model tiers, not a measured smaller-model advantage.",
    "The TH08 example comes from commit a393f400 (2 September), which target-checks every VC7 floating-literal relocation; the follow-up audits 1,548 relocations and corrects 12 stale rows in five accepted units. The essay’s concern that a reward built on the old check could reinforce wrong source is stated as an inference: the notes record no actual RL exploit or trained model doing so.",
    "Game rights stay with their owners."
  ],
  "requirements_zh": [
    "文章与证据说明不提供游戏文件、编译器或构建产物；所链接的各东方 Project 仓库自行说明其要求。",
    "文章注明日期为 2026 年 10 月 13 日。文中示例是链接的 TH09、TH095、TH04 与 TH08 提交中记录的函数级比对。",
    "较晚的会话继承了此前的源码、诊断和失败实验，因此前后对比的攻克结果并非对模型规模效应的受控估计。证据说明指出 gpt-web 是随时间变化的标签，无法据此确定单次运行所用模型；各次运行的推理强度和预算并未统一记录；GPT-6-Luna-max 的那次攻克被表述为跨模型档位的发现，而非经测量的小模型优势。",
    "TH08 的例子来自提交 a393f400（9 月 2 日），该提交对每个 VC7 浮点字面量重定位都对照目标进行检查；后续审计共检查 1,548 处重定位，修正了五个已接受单元中的 12 条过时记录。文章担心以旧检查为基础的奖励可能强化错误源码，但这被明确表述为推断：证据说明并未记录任何实际的强化学习利用或训练模型这样做。",
    "游戏权利仍归原权利人。"
  ],
  "targets": [
    "binary",
    "game"
  ],
  "url": "https://rea.tools/blog/reverse-engineering-models/",
  "verification": "source-reviewed",
  "verified_at": "2026-10-19"
}