Document the isolated native UI review entry

docs/3.0/PROGRESS.md

@@ -1,5 +1,11 @@
 # PROGRESS · Clawket 3.0 进度日志
 
+- 2026-10-17 Native 第一轮 UI 初验独立入口（RN 保持默认）。
+  - 原合成peer进程已退出，原Studio当前只有缓存/正在重连；不以新peer冒充原physical恢复。另起专用UI peer、新physical及新saved connection，显示名「Clawket · UI 体验」，已实际通过产品连接表单建立。WebDriver仅发送一次文字与三张已知fixture图片，独立stats为acceptedSends1/completedRuns1/active0/images3/dedupe0；完成窗口App无Alert/Stop，一张新回复图片实际原生QuickLook→完成通过。这是新合成UI窗口，不是真模型或原连接恢复验收。
+  - 原Studio空draft原UTF8及三资源ID/顺序与baseline完全一致，未发held。真实Hermes长回复已通过，Native Stop、OpenClaw认证、Relay/弱网恢复仍未通过；VoiceOver/完整N5与owner接受未完成，不以文档/CI替代UI或模型验收。
+  - 主线程已确认首轮入口：Device Hub → Clawket Chat Craft QA → 当前「Clawket · UI 体验」。当前空draft/三张新的未发草稿图片，实际在聊天table下拉后keyboardVisible由true变false，无Alert；截图已保存并实际查看。新peer累计仍accepted1/completed1/active0/images3/dedupe0，无额外Send；原Studio baseline空draft hash/三有序refs再次全等。此窗口可交互体验模拟回复、聊天/编辑/上传/预览与导航，不是完整功能或真实后端合格证明。
+  - GUI使用已安装的Device Hub，仅选「Clawket Chat Craft QA」模拟器，不触物理手机。先前误按旧Simulator路径判断缺GUI而进行的不必要XIP解压已停止，未修改完全磁盘访问权限。无RN默认、共享服务、既有凭据或发布变更；本独立切片只更新两份验收文档，不使用设备/租约或运行本机测试。16指令对检查与diff通过，独立只读文案复核无阻挡；完整CI另验，不抵UI/模型验收。
+
 - 2026-10-16 Native 有界流开发包真实复验与第一轮 UI 初验准备（RN 保持默认）。
   - #276 exactfe157111/current-main/CLEAN全部11门禁绿后合2d8a7890并finish；仅证据文档，不改产品。新独立acceptance工作树使用#277 productiond5d2e6a3（与重基6d2fd826源码/测试等价）开发App，不打Archive/IPA或改共享服务。
   - 实际UI automation stale导致预检停在Send0/Stop0，临时Studio草稿已恢复，三refs/held保留；随后真saved connection、header/原draft/导航route精确核对。新真实Send1经被动reader绑定receipt/见非空live，Native单assistant15630字节、streaming=false、无failure/Stop，长流完整结束。元素Stop调用未拿到中止证据；另一次新input单Send/单coordinate tap、live截图有Stop，最终亦final非aborted。未证明handler或坐标等于live center，原生取消仍未通过，均未重发。

docs/native-ios/BACKEND-ACCEPTANCE.md

@@ -50,7 +50,19 @@ One fresh real input received a run receipt and independently observed live text
 
 Afterward, there was no failure notice or Stop control. Actual system Paste restored the original 16-character draft with zero resources and Send enabled. An independent read-only history check verified the original owned input, same physical session and no active run. This establishes a completed real long reply and the later idle window; it is not full-network stability, an original-failure typed sample or successful native cancellation.
 
-The first owner review may focus on Studio's chat typography, code Copy, scroll/Latest, multiline editing, ordered three-image tray, preview/return and Work/Menu navigation. Studio is synthetic; use the separate Hermes QA conversation for real text only. OpenClaw authentication, native Stop, Relay/network recovery, VoiceOver and full parity remain pending. React Native remains the default.
+The first owner review targets chat typography, code Copy, scroll/Latest, multiline editing, ordered images, preview/return and Work/Menu navigation. The original Studio is no longer the live review entry; the separate UI connection below has its own synthetic scope. Use the separate Hermes QA conversation for real text only. OpenClaw authentication, native Stop, Relay/network recovery, VoiceOver and full parity remain pending. React Native remains the default.
+
+## First UI review: a separate synthetic connection
+
+The original synthetic-peer processes had exited. Original Studio now shows cached content and reconnecting; a replacement peer must not be described as recovery of its physical session. A dedicated UI peer was started with a new process and physical session, and a new saved connection named **Clawket · UI 体验** was actually established through the product connection form.
+
+WebDriver sent text with three known fixture images once. Independent peer statistics recorded `acceptedSends=1`, `completedRuns=1`, `active=0`, `images=3` and `dedupe=0`. The app showed no Alert or Stop in the completed window. A new reply image actually opened in native QuickLook and returned through Done. This proves that bounded synthetic UI interaction, not model inference, backend cancellation or recovery of the original connection.
+
+Original Studio's empty draft bytes and three resource IDs in their original order exactly matched the baseline; its held input was not sent. The completed real Hermes long-reply window remains separate evidence. Native Stop, OpenClaw authentication, Relay/weak-network recovery, VoiceOver and full parity remain pending.
+
+The main thread confirmed the first review entry: **Device Hub → Clawket Chat Craft QA → Clawket · UI 体验**. That connection has an empty text draft and three new unsent draft images. An actual downward drag in the chat table hid the keyboard, and the saved screenshot was inspected with no Alert present. Peer totals remained one accepted send, one completed run, zero active runs, three images and zero deduplicated sends; no extra Send occurred. Original Studio's baseline draft hash and three ordered resource IDs were checked again and remained identical. This prepared window is an interactive first review of simulated replies, chat, editing, uploads, previews and navigation, not full functional or real-backend acceptance.
+
+The existing Device Hub GUI selected only the **Clawket Chat Craft QA** simulator; no physical phone was operated. An unnecessary XIP expansion attempt based on the old Simulator path was stopped, and Full Disk Access permissions were not changed. React Native remains the default; no shared service, existing backend credential or release configuration changed. This documentation task performs no device operation, native build or local test, and documentation CI does not establish UI or model acceptance.
 
 ## Later authentication diagnostics