Skip to content

fix(research): refresh analysis and clarify contextual quantities - #7217

Merged
loopx-agent merged 1 commit into
mainfrom
codex/research-context-readback-20261024
Oct 24, 2026
Merged

loopx-agent merged 1 commit into
mainfrom
codex/research-context-readback-20261024

Conversation

@loopx-agent

Copy link
Copy Markdown
Collaborator

Refreshing Research reread history but left its analysis catalog stale, so a newly registered version required opening an older analysis to refresh it. The existing refresh now reads both through their existing owners, preserving the selected scope, URL and closed analysis state without executing a model.

An explicit source_text_aliases_with_literal_values_v2 encoding also asks the existing analysis to explain source-context-supported quantities in ordinary language while retaining exact literals. Earlier encodings keep their complete prompts, identities and saved results. This is a model instruction, not a semantic verifier; assessment eligibility, metrics, source authority and execution permissions are unchanged.

Validation: 23 TypeScript assessment tests; 55 store/HTTP/observer tests; 24 complete prior-encoding base/head preparation comparisons; control-plane typecheck; full frontend/Chat build; packaged production-owner browser cases at 1440/390px. The stale-catalog case fails against the baseline and passes here, including no execution, unchanged selection, partial/foreign translations, failure recovery and citation keyboard return. Exact-scope CQR is valid. The standard local premerge gate is recorded separately before merge; CI is not consulted under the managed Goal policy.

A bounded actual-model diagnostic resolved the demonstrated compressed-quantity sentence. Lexical completeness and selected qualitative examples do not establish full semantic accuracy, calibrated utility or causal improvement. Repeated long questions in the complete analysis pane and broader source coverage remain in the existing research-workspace work. No whole-page redesign or installed adoption is claimed by this proposal.

Visual evidence

Tested head: 5c139d767d7a636b8c54bbd4f836748a730fa07c. Fresh captures use built assets and the real disposable SQLite/history/job HTTP owner with synthetic long bilingual content and mixed negative/unknown states. These current captures are byte-identical to the existing committed assets below, which are reused rather than duplicated. Baseline presentation is c37ab4df9b00f86cd502f0d355b4d1a52bc97760. No live/private operating state is shown; these are packaged browser captures, not native installed acceptance.

The default page retains one Research entry, compact six-metric readings, shared collapsed source/usage details and the timeline. Expanded provenance wraps at normal text size and retains history and unknowns. Appearance is unchanged; the useful delta is refresh behavior and explicitly versioned generated prose. Functional folding/overflow checks are separate from this visual judgment.

Desktop before Desktop after
Research default before Research default after
Narrow before Narrow after
Research narrow default before Research narrow default after
Expanded source, desktop Same detail, narrow
Expanded source desktop Expanded source narrow

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

研究页读者刷新记录后仍看不到刚登记的分析版本,阅读压缩数量记录时也可能只看到难懂的原始英文记号。 原来一次刷新只更新历史,还需展开旧分析再刷新目录;现在原按钮同时读取两者,保留所选版本。新编码仅在明确分析时要求解释已有来源支持的数量,旧结果不重写。 同一真实后端浏览器反例在基线无法发现新版本,在本提交一次刷新即可发现,且没有模型执行。限定真实模型诊断中的数量句也变得可读,来源词仍保留。 本 PR 不改研究历史、指标、执行权限或评分,不声称全部译文语义通过、原生收益提升或整页优化完成。 完整分析中的重复长问题、其他案例来源覆盖及安装后的采用仍需原研究工作继续验收。

改动思路

刷新复用已有带作用域和错误状态的 reader,数量策略留在现有轨迹分析 owner;明确选择 v2 才改变提示,保留旧记录身份,避免另建配置、翻译服务或状态缓存。 本次边界是一次刷新发现分析版本,以及有来源支持数量的显式新分析;原始记录、科学判定与执行权限不变,整页重复问题继续留在既有任务。

较小的修复已足够:直接复用已有 reader,而非增加缓存、另一颗刷新按钮或新的翻译服务。旧提示原位修改会改变历史结果身份;新模式沿用准确字面量检查,仅增加一句明确的数量表达约束。上下文仍是证据,不能补成原记录的新行动,模型不理解时应保留未知。浏览不执行模型;已有意图和授权下不增加必需用户轮次,版本选择和引用回看是自愿探索。无需另建设置页或让用户填写内部标识。

具体改动

关键代码讲解

  • GoalResearchHistory 刷新 仍更新既有历史 revision,同时调用已有分析 reader;错误归属于原来的 readState,异步请求仍受 Goal、来源、语言、所选包和 generation 限制,失败可沿现有入口重试。
  • prepareDisplay 让显式 v2 复用 v1 的准确数量保护,不改变冻结历史、可评片段、六项指标或原始证据。类型定义只扩展既有 display encoding,不新增权威状态。
  • prepareTrajectoryAssessment 仅在 v2 且已有引用语境时加入数量说明,并计入分析身份。必须保留原始标记,解释有证据的数值和所指,禁止把准备步骤移入原行动;模糊含义不能猜测。新模式不能复用旧模式的输出,旧模式的完整提示与身份仍逐字相同。

依据修改前 docs/development/design.md,spec_revision 为 659c1fd。research-read-effects、faithful-history-language、optional-evidence-detail 在此 head 实现并由真实后端浏览器、契约及原文读取回归覆盖;whole-page-readability 的重复长问题部分仍保留在既有研究工作中,没有改写规范来宣称完成。参考文档同步解释 v2 的执行和语义边界;现有浏览器用例增加“初次读取后出现新目录项”的反例,未增加临时测试框架。

对主干的风险

最强反例不是字段缺失,而是字面检查全过后仍无法理解数量,甚至引入支持材料中的动作。第一轮真实输出确实出现了该问题,历史结果保留;v2 用相同来源的新显式执行验证目标句,预先选定的定性字段检查也保留负结果、未知与不采用语义。这不认证全部模型译文,更不构成科学指标收益。另一个反例是刷新悄悄执行或重置选择;原始基线在新版本可见性断言上失败,此 head 在桌面和窄屏同时通过,模型 mutation 增量为零,选择、URL 和分析关闭状态保持。

23 项 TS、55 项 store/HTTP/进程、24 份旧编码完整准备结果对照、类型检查、完整构建及打包浏览器通过。初次 premerge 因旧 CQR 未被当前合格结果替代而拒绝,保留该失败;修正结果字段长度及引用范围后重新登记准确作用域,再跑完整标准门,13 个选定检查及 4 个直接检查通过,零失败、警告和人工 hold。现有构建分块大小提示保留。相关 #7201/#7208 只涉及摘要 matcher 和另一历史场景,未替代本修复。没有查询或等待远程 CI。

语义与 CI 对齐

本变更扩展已有 typed display vocabulary。旧编码、未指定语言以及有/无上下文的完整准备结果保持;新编码故意形成新 key,未改变 Core 调度、权限、评分或错误分类。数量可读性是模型指令,不能伪称机器语义验证;数值、来源范围和原始状态仍由既有 owner 检查。差异语义探针不覆盖 TS union,零候选不当作无语义变化;已单独审查该新增值,完整仓库语义检查通过。

我的整体评价

long_horizon 为 preserved:普通读取不重复执行,旧结果及失败来源保留,新分析单独记账。user_experience 为 improved:原位刷新消除展开旧分析再刷新的绕路,限定数量句也有实际模型证据;这是 justified_increment。默认桌面和窄屏仍以研究页、指标及历程为主;来源与用量保持一处折叠。展开来源后正文正常换行,引用、历史与未知分组可读;这次没有新增说明文字或宣称视觉改版。 所检为合成数据的打包页面;真实模型只有限定字段的定性核验,不能推广为全部语义正确。完整深度分析的重复长问题、其他来源覆盖与原生白屏稳定性仍由既有研究工作继续处理。

刷新复用已有带作用域和错误状态的 reader,数量策略留在现有轨迹分析 owner;明确选择 v2 才改变提示,保留旧记录身份,避免另建配置、翻译服务或状态缓存。 本次边界是一次刷新发现分析版本,以及有来源支持数量的显式新分析;原始记录、科学判定与执行权限不变,整页重复问题继续留在既有任务。 六个既有文件的范围与问题相称,未发现需要阻塞的具体问题。当前 head 5c139d7 的本地资格和完整评审支持 APPROVE;安装采用、整页剩余体验和更广模型质量分别继续验收,评审结论不自行授予合并权限。

English verdict: APPROVE — 5c139d7. One existing refresh now reads both owners without execution or selection reset; explicit quantity-rendering v2 preserves prior prompts and results. Base/head regression, old-mode parity, backend/browser checks and the standard local gate pass. Semantic diagnostics are bounded; complete model quality and installed adoption are not certified.

@loopx-agent
loopx-agent merged commit 10d23a8 into main Oct 24, 2026
10 of 11 checks passed
@loopx-agent
loopx-agent deleted the codex/research-context-readback-20261024 branch October 24, 2026 05:37
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