[loopx-agent]

Research and Goal overview can stay blank after native navigation even though their content is present in the accessibility tree. Their entrance animations begin at zero opacity; a suspended animation clock leaves the document and authoritative values invisible. This change keeps those regions and Stop consequences fully visible from the first frame, while retaining the small movement and reduced-motion behavior.

研究内容、真实读数和停止操作后果从首帧起可见；保留轻微位移，不依赖动画结束才显示。设计规范和现有回归同时补上暂停动画时钟的验收，避免截图自动结束动画掩盖空白。

Scope and validation

• Existing panel/value/popover CSS owners; no routing, backend, model execution, permissions or stored-state changes.
• Unchanged production source 61b83ac fails both frozen-clock oracles (research opacity 0 at 1280px/390px, missing Goal values). Those CSS bytes are also unchanged on the PR base 9b8adda.
• Production/test revision d903f26a9afe36fcf0a5898f29fa1ee14a8182eb: TypeScript/production build; packaged research-native-reading, goal-masthead, workspace-view-recovery and research-history all pass. Native measurement scenarios exercise the owning backend; remaining shell fixtures are synthetic. Existing Stop/recovery, reduced motion, precision/unknown/negative results, long research and selection checks remain covered.
• Risk-based premerge: four direct checks and 13 selected catalog/risk/public-boundary checks pass. All six binary-image holds were inspected for pixels, metadata and synthetic provenance. Existing large-chunk build warning remains.
• This qualifies the packaged frontend. Installed native adoption and cross-restart workspace restoration are separate readbacks; this PR does not claim either from DOM presence or screenshots. No new model analysis or benchmark run was launched.

Final head 78356922410052eda006470f444c305226936f31 differs from that tested revision only by naming two unchanged PNGs by their actual 1440px width; all production and test source bytes are identical.

Visual evidence

Actual first viewports of the packaged build with the CSS animation clock paused; synthetic public fixtures, not screenshots of a private workspace. Screenshots deliberately allow animations rather than fast-forwarding them. Baseline 61b83ac; proposed revision above. The populated Goal overview includes long task names, two waiting decisions and unavailable/stale result states. The research fixture has native measurements and zero history rows; populated/expanded research parity is covered separately by research-history.

View	Before	After
Populated Goal overview, 1440px
Research, 1280px
Research, 390px

At 390px the toolbar, scope and primary reading are visible without horizontal overflow; auxiliary readings continue below the fold. This is a visibility repair; it does not certify the entire workbench redesign complete.

Exact-scope change-quality qualification: cqr_138c63eaea1bc4b41317, base 9b8adda, head 7835692; pass with no unresolved blockers. Review/readiness and installed native verification remain independent.

[loopx-agent]

Reviewer: model_agent | gpt-6.1-sol | OpenAI | runtime_reported | reasoning_effort=max

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

没有发现阻塞本次修复的缺陷。评审头为 78356922410052eda006470f444c305226936f31；安装后的原生验收另列，不由打包截图代替。

动机

阅读已有 Goal 研究结果和工作状态的用户。切换到研究页后，标题和控件已变化，正文与读数却仍为空白；修复后暂停动画时钟也能立即看到同一份内容，无需再滚动或切换来碰运气。已验证打包页面在暂停时钟的桌面和窄屏显示真实读数，停止确认仍能读到完整后果；本次不把可访问树有文字当成像素可读。本次不增加数据、分析模型调用或权限，不改变研究排序、来源选择与缓存身份。真实安装后的原生回读与跨重启恢复仍需分别验收，整页精简和报告布局沿既有工作继续。

改动思路

可见性属于现有展示层的正确性，不应另设一个开关、路由补丁或状态 owner；删除透明起始帧就是最小有用修复。本 PR 交付 panel、读数和停止后果首帧可见，并补同一入口的反例；不扩大到会话、持久化或浏览器重启策略。我比较了维持现状、强制滚动/重绘、完全关闭动画和仅删除透明起始帧：现状有可重复空白，重绘补丁增加平台耦合和人工步骤，完全取消动效过宽；当前方案继续使用原 CSS owner 和现有回归。数据来自原 Goal 的既有投影/测量后端，JS 数据和动作处理不变；展开阅读与显式分析仍是原路径。已有明确意图到可读结果无需新增选择、确认或复制信息。

条件	原版本	当前版本
暂停时钟后切换研究	内容透明，AX 仍有文字	原正文与读数立即可见
暂停时钟后打开停止确认	入口使用透明首帧	原后果可读，可原样取消
正常/减少动态效果	既有路径	保留位移与减少动态效果分支

具体改动

以修改前 docs/development/design.md、不可变版本 9b8adda2f04c285d1c13ecc90b3abba93202ebee 为依据：research-workspace 的真实 Goal/只读切换由原 GoalWorkspacePanels 和恢复测试覆盖；populated-page-review 已检查长任务、混合状态、76 条研究记录以及展开正文；motion 保留短暂位移与减少动态效果；accessibility 验证窄屏、键盘与 Escape 返回。新增 Motion 文字本身是有理由的规范修正，不拿它反过来证明自身合格。

关键代码讲解

• goal-overview.css:110 的 goal-panel-enter 删除 opacity=0，仍从 4px 位移开始。goal-workspace-panels.tsx:22 的真实活动面板直接消费；隐藏面板仍通过 display:none 隐藏，未把所有标签同时显示。复用该动画的折叠输入工具也获得相同可见性，不新建状态开关。
• goal-masthead.css:62 的 goal-value-swap 保留 3px 位移，从完整真实值开始。goal-masthead.tsx:148 仍直接显示原读数，不插值任务数、费用或时间，也未把未知变为零。
• goal-masthead.css:61 的 goal-popover-enter 保留轻微缩放与位移。goal-masthead.tsx:124 的停止确认保留全部原后果；取消与执行仍由既有处理器决定，可见性不授予权限。

两个原有浏览器场景在导航前暂停 CSS 动画，逐级相乘检查祖先透明度，并用 animations:allow 截图；结束后移除测试样式继续原有断言。6 张公开合成 PNG 均检查像素、尺寸和元数据，无私有消息、路径或原始运行资料。最终头比运行测试的 d903f26 只多两张 PNG 的尺寸命名修正，生产/测试/设计正文的字节完全一致。

对主干的风险

最强反例是原生可访问树已经改变、动画时钟却不前进：旧截图若自动结束动画会掩盖缺陷。原生产 61b83ac（相关 CSS 在本 PR 的 9b8adda 基线相同）通过相同暂停入口，研究在 1280/390px 均测到 opacity=0，读数也失败；修复后相同入口为 1。早先尝试回卷已经结束的研究动画曾通过，那条保留记录，不被当作有效反例；之后改为导航前暂停，才得到上述失败与修复对照。

TypeScript/生产构建及 packaged research-native-reading、goal-masthead、research-history、workspace-view-recovery 全过；包括真实后端 23 组原生值、未知/负结果/精度与引用，及合成动作路径中的停止、取消、恢复、无隐式 action 和减少动态效果。最终头 premerge 的 4 项直接检查及 13 项选定检查均无失败，CQR cqr_138c63eaea1bc4b41317 有效。6 个二进制图像 manual holds 已逐张人工检查，文本扫描不能替代它。旧的大块 bundle 提示仍是 warning。按本次 wait_for_ci=false，不查询或等待远端 CI，也不宣称 CI 绿。

主要风险是浏览器冻结时钟不能解释全部原生合成器问题；这不会被包装成安装成功。CSS 不涉及持久化、调度、quota、权限或协议，回退此补丁也不迁移数据。原生更新必须走正式路径，跨重启恢复保留原缺口。相关 #7304 报告布局和 #7312 只读预览在其既有 owner 内，不由本补丁接管。

我的整体评价

这是 justified_increment。long_horizon=preserved：原记录、来源、后续读取和动作 owner 不变，未证明长期成本收益；user_experience=improved 仅指已复现的首帧空白修复。已看整页：长任务名、两项待决定与过期/不可用结果的概览不再只剩标签；研究页主读数可见，展开内容仍在原阅读区，390px 辅助值可继续向下读。未验证该新头在已安装原生 App 中的持续重绘和跨重启恢复；合成截图不证明这两项，也不代表整个研究工作台改版完成。

repository_reuse=reused，observable_semantics=intentional_change_validated，change_proportionality=proportionate；机制适配结论为 retain。可见性属于现有展示层的正确性，不应另设一个开关、路由补丁或状态 owner；删除透明起始帧就是最小有用修复。本 PR 交付 panel、读数和停止后果首帧可见，并补同一入口的反例；不扩大到会话、持久化或浏览器重启策略。没有 opt-in/default-off 或新的 actor/authority 协议，相关隔离/命名评估为 not_applicable；也没有新增共享枚举、状态或 prose 分类器。相邻同原因透明首帧一起删去，保留现有动画 owner，不建补偿框架。已验证的收益足以交付这个可逆增量；不据此关闭整体研究工作台验收。作者自审以 COMMENTED 发布批准结论，仍需 unchanged-head readiness 和已有明确合并授权，不能充当 GitHub 自己批准。

English verdict: APPROVE — 78356922410052eda006470f444c305226936f31. The frozen-clock baseline fails while the proposed packaged source preserves full initial visibility and existing read/action semantics. Required local checks and six visual holds are reviewed; installed native repaint/restart qualification remains separate. Author-owned COMMENTED fallback; CI was not consulted.