Skip to content

fix(research): search saved translations and export original matches - #7423

Merged
loopx-agent merged 1 commit into
mainfrom
codex/research-translated-search-20261027
Oct 27, 2026
Merged

loopx-agent merged 1 commit into
mainfrom
codex/research-translated-search-20261027

Conversation

@loopx-agent

Copy link
Copy Markdown
Collaborator

Words visible in the Chinese research reader returned no search results because the filter only inspected original source fields. Search now also matches the current qualified display text, including decoded observations and decision lists. Downloads use the same filter and retain the original records.

The existing history reader still owns language, source and snapshot qualification. Original view removes translation aliases; mismatched sources and missing fields fall back through the same owner. Matching stays literal, preserves status filtering and precedes display paging. It reuses the rendering map and adds no model execution or search index.

Validation at 86dbd917a51741899ecb0c74b81f90e0f838f92b:

  • The same Chinese title query fails on the baseline (0/30) and succeeds on this head (1/30). Both new search/export regression cases also fail on baseline source.
  • Nine owner cases and two packaged browser journeys passed against disposable SQLite and the real HTTP owner: Chinese title/body/decision queries, original JSON download, language/source isolation, 79-record paging, stale refresh, download failure and recovery.
  • Additional original-mode comparison retained identical results for 1,292 searches and 80 complete JSON exports. Type checking and desktop/Chat builds passed.
  • Native premerge: four direct checks and eight selected checks passed, no automatic failures. The six PNG manual holds remain explicit; each image was inspected for readable text, hierarchy, narrow layout and public safety. Exact-scope CQR is valid.

This is packaged renderer/backend evidence with synthetic analysis outputs, not installed-App adoption or a claim that every saved translation is semantically correct. Existing translation quality and native adoption work remain separate.

Visual evidence (1440px and 390px; synthetic research with mixed tentative/refuted findings):

Default first view:

Desktop first viewport
Narrow first viewport

The same displayed-language query, before and after:

Before: no match for displayed Chinese text
After: the original record matches its displayed translation
Narrow title match

Expanded negative observation, preserving the original number, uncertainty and source term:

Narrow body match and expanded evidence

Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>

@loopx-agent loopx-agent left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · effort=max

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

动机

在中文研究工作台按眼前正文查找历史记录的用户。 同一条中文问题,旧版搜索返回零条;本版本返回对应的原始记录,下载保留原始正文。 旧实现会迫使读者猜英文或逐条翻阅历史,已有中文展示也无法完成查找。标题、观察和决策正文可按当前展示语言检索,同时仍能按原文和来源标识查找。 本PR不生成或修订译文,不改变研究判断、原始记录或模型执行权限。 已有译文的实质语义质量和正式App采用仍需沿原工作验证。

改动思路

复用原有译文资格检查、显示映射和统一筛选函数,不建立搜索服务、额外索引或新的证据所有者。 本PR交付研究记录搜索与原文下载的一致性,生成质量、版本比较和原生App采用保持各自边界。 原始记录指实际保存的事件版本与正文;展示译文只提供额外检索文字。现有页面先核对来源、语言及完整快照,再把可用译文交给筛选。筛选始终返回原始对象,因此列表与下载共用同一匹配规则,而不是另建翻译后的证据。没有译文时继续原来的原文路径,不增加设置、确认或后台分析。相较只改提示词或只搜标题,本次补足了观察、解读及决策列表这一实际遗漏。

场景 旧版 当前提交
搜索眼前中文标题 0条 对应原始记录1条
搜索正文或决策短语 忽略译文 同时匹配当前有效译文
切回原文或来源不适用 原文筛选 移除译文匹配,保留原文筛选

具体改动

关键代码讲解

  • searchResearchRecords 将已有显示映射及解码段落加入文字匹配;状态仍按原始状态精确筛选,空查询、顺序、原文标识与分页前筛选保持原语义。它返回原始事件,未用模型文本判断结论是否通过。
  • ResearchRecordSearch 复用列表显示所用的映射,让列表和下载拿到相同筛选输入。没有建立持续缓存或第二个索引;输入提示改为“正文、来源引用或 ID”。
  • exportResearchRecords 仍先检查原始Goal,再序列化原始事件与覆盖信息。中文命中改变的是选择范围,不是事件正文、编号、状态或原始证据。
  • GoalResearchHistory 沿现有资格检查同时传递显示事件与段落。原文切换、语言或来源不匹配时两者都撤下,避免残留旧译文搜索结果。

规格依据为 docs/development/design.md,固定版本 99921829f2c5fa2ad35a024c72abb88717003ef3:已接受规范。Owner language first:已实现当前展示语言查找;Research Workspace:复用原版本化owner,保留原文与普通读取零模型;Populated Page Review:已检查桌面、窄屏、中文结果及展开负例。七个文本文件94行新增、11行删除,六张图片为合成数据的可审阅证据,没有提交生成bundle。

对主干的风险

主要风险是来源失效后旧译文仍可命中,或下载把展示译文当成了原始证据。本轮通过原文切换、来源不适用、语言和哈希不可核验等路径检查了资格撤销;又把中文命中的下载对象与原始导出逐项比较。旧版同一中文查询失败,新提交在1440及390宽度通过,正文和决策短语也有独立断言。九项owner用例通过;79条长记录原文旅程覆盖了状态、后续分页、旧版本、刷新失败与下载恢复。额外原文模式对比的1,292次搜索和80份完整JSON导出与基线相同。该扩展只做文字子串查找,不从文字推断状态、权限或科研结论。

类型检查及两种生产构建通过;原生premerge四项直接检查和八项选中检查通过,没有自动失败。六项PNG人工门仍保留,截图已逐张阅读,不将无溢出或字段存在冒充可读性判断。当前配置不查询或等待CI;作者评审不构成维护者合并授权。

我的整体评价

该准确提交是 justified_increment:交付了从中文查找到原文下载的完整现有路径。long_horizon:preserved,历史、状态、恢复和执行边界保留,没有新增反复分析或索引维护;user_experience:improved,读者可直接查找眼前内容,不必补一次翻译或猜英文。已逐张检查合成研究整页、中文搜索和390px展开负例;主要读数、记录与按需来源层次保持清楚,中文检索结果可以直接定位。 正式原生App采用和真实已存译文的全面语义质量未由本次合成渲染验证。 既有展示与解码owner足够,局部映射类型不新增共享语义词表;复用已有筛选与映射是本轮已落实的简化。准确head 86dbd917a51741899ecb0c74b81f90e0f838f92b、CQR和本地反例/集成结果支持 APPROVE;正式App采用及所有真实译文质量不在这个评审结论中被补判通过。

English verdict: APPROVE — 86dbd917a51741899ecb0c74b81f90e0f838f92b repairs displayed-language research search and keeps list/download scope aligned on canonical records. Baseline fails the same Chinese query; nine owner cases, both real SQLite/HTTP packaged journeys, source-mode parity, type/build and all selected automatic premerge checks pass. Six visual holds were manually inspected. Native adoption and complete translation semantics remain outside this consumer slice.

@loopx-agent
loopx-agent merged commit f1d9478 into main Oct 27, 2026
31 of 45 checks passed
@loopx-agent
loopx-agent deleted the codex/research-translated-search-20261027 branch October 27, 2026 03:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant