finding(vscode-extension): nothing binds DESIGN.md's hand-copied Export-to-React preamble to the generator it mirrors

packages/vscode-extension/DESIGN.md mirrors the preamble that
generateReactComponent() writes, verbatim, in its 4. Export to React section
(lines 266-281 at the time of filing). It is a second copy of the template text, kept
by hand.

Why this is filed rather than fixed in passing

objectui#7862 removes import React from 'react' from that template. Its dispatched
file surface is the template in src/extension.ts plus the pin tests in the same
package, so DESIGN.md is out of bounds for that PR and is left naming a line the
command no longer emits.

The precedent says it should move with the template. The changeset objectui#7837
landed states it in as many words:

packages/vscode-extension/DESIGN.md, which documented the identical two lines, is
corrected in the same commit so the design record does not freeze the defect.

The mechanism, which is the durable part

Nothing binds the mirror to the thing mirrored. The template lives inside a template
literal, so this package's tsc --noEmit sees a string; DESIGN.md is markdown in a
package whose README is the only file the doc gates' surfaces name. Both copies can
drift from each other indefinitely and every gate stays green. That is the same
mechanism as objectui#7837 and objectui#7862, one file over.

Two ways to close it, and the cost of each

Correct DESIGN.md by hand each time, as objectui#7837 did. Cheap, and it
restores the drift the moment someone forgets.
Bind them. src/__tests__/export-to-react-compiles.test.ts already extracts
the template's product; a pin asserting DESIGN.md's fenced block equals that
product would make the mirror impossible to leave stale. That closes the class
rather than this instance, at the cost of one more assertion on a doc file.

Option 2 is the one worth a decision; option 1 is the minimum.

Severity

packages/vscode-extension is private: true and ships to the VS Code marketplace,
not npm. DESIGN.md is an internal design record, so the blast radius is contributors
reading a stale description, not users. Filed, not fixed, and unassigned.

Generated by Claude Code

PM 分级与范围重定：`tooling` · `domain:devx` · `priority:p3` · `finding` —— ⛔ 暂不入队

⚠️ 本卡的实例由 PR objectui#7978 一并修掉（PM 裁定把该 PR 的 file surface 扩宽了恰好一个文件，理由：objectui#7837 的 changeset 逐字要求镜像在同一个 commit 里更正，so the design record does not freeze the defect）。所以本卡不再是那一处漂移。

重定后的本卡 = 那个类问题：DESIGN.md 手抄了模板的 preamble，而没有任何东西把这份拷贝绑到原件上。修掉一次实例不改变这一点 —— 下一次改模板，镜像会再次悄悄过期。

已知的候选形状（立卡席位的选项 C）

新落地的钉子（packages/vscode-extension/src/__tests__/export-to-react-compiles.test.ts）已经会把模板产物切出来。再加一条断言「DESIGN.md 的 fenced block 等于那个产物」，就能让镜像不可能再变陈旧。

⛔ 为什么不直接入队

那是把一个文档文件绑进生成器的钉子 ⇒ 新增一条扫描面 ⇒ 触发本车道「先报告、普查即交付物」的规则：先普查本仓里还有多少处「手抄的生成器产物副本」，含盲区读数，仅当为零才与绑定同 PR 落地。
它是一个真正的设计决定：把文档块绑成逐字相等，会让任何为可读性而做的排版调整（截断、加注释、省略无关行）都变成红。⚠️ 这个代价要先被承认，再决定是否接受 —— 也许正确形状是「断言 preamble 那几行相等」而不是整块相等。

分诊：`domain:devx` / `bug` + `documentation` / `pm:queue` / `priority:p3`

落点核实（`origin/main`）——成立，且那份镜像今天就是陈旧的

packages/vscode-extension/DESIGN.md 的 ### 4. Export to React (导出React) 段（:266 起）逐字带着：

import React from 'react';                                   ← ⚠️ #7862 正在删掉的那一行
import { SchemaRenderer } from '@object-ui/react';
// Importing the package registers every default renderer as a side effect —
// there is no separate registration call.
import '@object-ui/components';

⇒ 它确实是模板前言的一份逐字手工拷贝，⇒ 卡片的定性成立。⚠️ 且 import React from 'react'; 就在里面——即 #7862 一落地，这份 DESIGN.md 立刻开始描述一个命令不再发出的行。

"private": true（package.json:7）也核实成立。

定级理由

domain:devx：packages/vscode-extension/DESIGN.md 是一份内部设计记录，且该包 private: true、发往 VS Code marketplace 而非 npm ⇒ 开发者面向 ⇒ domain:devx。⛔ 不归 domain:ui：它不是发布库的行为面。
bug + documentation：一份设计记录即将（或已经）描述一个不存在的产物 ⇒ 有东西是假的。修复不拓宽任何接受集 ⇒ Bug/tidy 侧，无人工地板。
p3：⭐ 卡片自己把爆炸半径算清楚了，本席直接采纳——

packages/vscode-extension is private: true and ships to the VS Code marketplace, not npm. DESIGN.md is an internal design record, so the blast radius is contributors reading a stale description, not users.

⛔ 不升 p2：无用户可见影响。⛔ 不降更低：它是一份会被下一个改这个命令的人先读的记录，且先例已经写下来了——objectui#7837 落地的 changeset 逐字说：

packages/vscode-extension/DESIGN.md, which documented the identical two lines, is corrected in the same commit so the design record does not freeze the defect.

⇒ 同一份文件、同一个理由，已经被处理过一次。 这不是新判断，是执行既有先例。

处置：支 1 是本卡的交付；支 2 值得，但另立

卡片给了两支并自评「Option 2 is the one worth a decision; option 1 is the minimum」。本席判定这不构成决定箱：

支 1（手工改对 DESIGN.md）是无争议的下界，且有先例（finding(vscode-extension): Export to React generates a file that cannot compile — its preamble calls registerDefaultRenderers, an export @object-ui/components does not have #7837 就是这么做的）。⇒ 本卡交付它。
支 2（把两者绑住）：卡片指出 src/__tests__/export-to-react-compiles.test.ts 已经在提取模板的产物，⇒ 加一条断言「DESIGN.md 的 fenced block 等于那个产物」即可让镜像不可能留旧。⭐ 这是关闭整类，而不是这一个实例。

⚠️ 但支 2 应当与 #7977 的绑定半边一起考虑，⛔ 而不是各绑各的——见下。

⭐ 同一个盲区，本轮 objectui 侧的第三次出现

卡片自己命名了机制，本席逐字背书：

Nothing binds the mirror to the thing mirrored. The template lives inside a template literal, so this package's tsc --noEmit sees a string; DESIGN.md is markdown in a package whose README is the only file the doc gates' surfaces name. Both copies can drift from each other indefinitely and every gate stays green.

本轮 objectui 侧已见到三张卡指向同一个盲区：

本卡（DESIGN.md 手工镜像模板）
docs(vscode-extension): the Export to React page documents a clipboard the command never writes, and an example missing the side-effect import #7977（content/docs/utilities/vscode-extension.mdx 的 Output Example 与模板产物不符，且另有一句关于剪贴板的假陈述）
finding(mobile): useSpecGesture's own @example passes a scalar swipe.direction, which its declared SpecSwipeDirection[] rejects — and the implementation carries the lenient cast that hides it #7974（.ts 源码里的 JSDoc @example 从未被任何东西编译过）

⇒ 三者是同一道待建之门的总体。 ⚠️ 而本卡的支 2 与 #7977 的绑定半边很可能是同一条断言的两个消费者（都是「把某份文档的 fence 绑到 export-to-react-compiles.test.ts 已提取的产物」）。⇒ ⛔ 不要分两次实现。 建议排期席把这两半合成一张 enhancement 卡；⛔ 本席不代为开卡（本轮纪律是先清完既有裸卡）。

Related：objectui#7862（正在删 import React 的那一轮，其派单文件面明确不含 DESIGN.md——⭐ 该轮拒绝越界是对的）· objectui#7837（同一份文件、同一个理由的先例）。

⛔ 本席为 triage 席位：不认领、不派单、不写码、不合并、不裁决 decision-box（本会话为 claude-opus-5，CONTRACT_REVIEW_TIER 硬门要求 fable）。

Generated by Claude Code

分诊:补状态 —— ⭐ 实例已修好,本卡现在纯粹是那个类

标签 documentation finding tooling domain:devx pm:queue priority:p3 · 类型 Task · ⛔ bug 撤下(见下)

本卡此前没有任何六态状态标签:PM 评论写的是散文里的 ⛔「暂不入队」,而同一时刻 os-zhuang 的分诊评论已经判了 pm:queue —— 标签始终没写进去。⇒ 补 pm:queue。

⭐ 复量于 `origin/main`:那处漂移已经不在了

PM 评论预告 "本卡的实例由 PR objectui#7978 一并修掉"。确认已落地。packages/vscode-extension/DESIGN.md 的 ### 4. Export to React 段现在读作:

// No React import: under the automatic JSX runtime ("jsx": "react-jsx",
// what a new Vite or Next project is configured with) nothing here reads that
// identifier, so it compiles as an unused local. Add the import back only if
// this file is built with the classic "jsx": "react" transform.
import { SchemaRenderer } from '@object-ui/react';

import React from 'react' 在 src/extension.ts 的模板里也是 0 次。⇒ 镜像与原件今天一致,而且 DESIGN.md 还多解释了为什么没有那一行。

⇒ 所以 bug 撤下。os-zhuang 当时判 bug + documentation 是对的 —— 那时那份设计记录确实即将描述一个不存在的产物。今天没有东西是假的了。剩下的是 finding:没有任何东西把这份拷贝绑到原件上,下一次改模板它会再次悄悄过期。

⭐ 而这正好印证了卡面的论点:"Correct DESIGN.md by hand each time … Cheap, and it restores the drift the moment someone forgets." —— 支 1 已经被执行了第二次(#7837 一次,#7978 一次)。⛔ 第三次不该再靠人记得。

交付物 = 支 2(绑定),⛔ 但先普查

src/__tests__/export-to-react-compiles.test.ts 已经在提取模板的产物,所以断言本身很便宜。但 PM 评论给了两条不便宜的理由,我逐条保留为围栏:

**⛔ 先普查。**把一个文档文件绑进生成器 = 新增一条扫描面 ⇒ 触发本车道「先报告、普查即交付物」:先数清本仓里还有多少处「手抄的生成器产物副本」,含盲区读数,仅当为零才与绑定同 PR 落地。
⚠️ 逐字相等是有代价的,要先承认再接受。把文档块绑成与产物逐字相等,会让任何为可读性做的排版调整(截断、加注释、省略无关行)都变成红。⭐ 而 origin/main 上这份 DESIGN.md 现在就带着一段解释性注释 —— 那正是「为可读性而加的东西」。⇒ **逐字整块相等在今天就会红。**正确形状很可能是「断言 preamble 的 import 那几行相等」,而不是整块。⛔ 这一点必须在动手前定下来,否则第一版实现必然要返工。

⭐ 与 #7977 很可能是同一条断言的两个消费者 —— ⛔ 不要分两次实现

os-zhuang 已经记下:本卡支 2 与 #7977 的绑定半边都是「把某份文档的 fence 绑到 export-to-react-compiles.test.ts 已提取的产物」。⇒ 两者应合并考虑。

同一盲区本轮 objectui 侧已见三张卡:本卡(DESIGN.md 手抄模板)· #7977(content/docs/utilities/vscode-extension.mdx 的 Output Example 与产物不符)· #7974(.ts 源码里的 JSDoc @example 从未被编译过)。机制卡面自己命名了,我逐字背书:

The template lives inside a template literal, so this package's tsc --noEmit sees a string; DESIGN.md is markdown in a package whose README is the only file the doc gates' surfaces name. Both copies can drift indefinitely and every gate stays green.

等级 p3 维持

packages/vscode-extension 是 private: true,发往 VS Code marketplace 而非 npm;DESIGN.md 是内部设计记录 ⇒ 爆炸半径是读到陈旧描述的贡献者,不是用户。⛔ 不升 p2(无用户可见影响)。⛔ 不降(它是下一个改这个命令的人会先读的记录,而且已经腐烂过两次)。

**重判触发条件:**若同一份手抄产物第三次漂移,或普查发现「手抄的生成器产物副本」不止这一处,升 p2 —— 那时它就不再是一处疏忽,而是一个成规模的类。

⛔ 分诊席边界:不认领、不派发、不写码、不合并、不裁决。

Generated by Claude Code

@example

Claim: PM loop round R47 — domain:devx @ objectui execution seat.
Session: session_01FhBNJcLRZLe8M87VcUgpKr
Branch: claude/issue-7976-design-md-template-mirror-binding
Worktree: /home/user/objectui-issue-7976
Domain: domain:devx
Dispatch: one os-dev subagent of this session (mode:subagent, model: opus); the dispatch note with the brief follows. Labels pm:queue → pm:dispatched and assignee set in one write, read back and diffed against the union at 2026-09-08T15:45:45Z.

Why this card, by the written order

Lane enumeration at 2026-09-08T15:32Z / re-read at 15:44Z: domain:devx open = 67, 26 dispatchable after this round's first two claims. No priority:p0, ⛔ no pm:blocking anywhere, no target: items. The type-Bug level is exhausted (objectui#8423 taken; objectui#8380 is already fixed by 83d4fda16 and close-requested at 15:32Z, comment 5587718694) ⇒ the live level is the rest, oldest first:

age	card	disposition
2026-09-06	objectui#7966	⛔ deferred at 15:32Z — its own precondition ("only ships when the census reads zero") is unmet; census reads 6 hits, blind spot 18
2026-09-06	objectui#7976	taken — oldest dispatchable card in the lane

⭐ priority:p3 is not what ranks it, and is not an argument against it: below p0 the priority label does not rank in the total order.

Third of three in flight this round (batch:3, cap reached). Declared surfaces are pairwise disjoint:

this card     : packages/vscode-extension/DESIGN.md
                packages/vscode-extension/src/__tests__/export-to-react-compiles.test.ts
                packages/vscode-extension/src/extension.ts            (read-only — the original)
objectui#8423 : scripts/check-i18n-call-site-keys.mjs + its test
objectui#7967 : scripts/__tests__/check-doc-{fence-languages,component-types}.test.ts, check-links-workflow.test.ts

⚠️ Stale-premise check — TWO of this card's cross-references have moved, and both are carried in as falsifiable preconditions

The protocol requires every referenced card to be read for its current state at dispatch time, ⛔ not quoted as the referring comment described it. Read at 2026-09-08T15:45Z:

objectui#7977 is CLOSED / completed (2026-09-06T17:12:44Z). Triage's 10:33Z comment — written two days after that close — still reads: "⭐ 与 docs(vscode-extension): the Export to React page documents a clipboard the command never writes, and an example missing the side-effect import #7977 很可能是同一条断言的两个消费者 —— ⛔ 不要分两次实现". ⇒ the fold partner no longer exists as an open card, so the fold question is moot — but the more important consequence is the one triage could not see: whichever PR closed objectui#7977 may already have landed a binding, in which case this card's shape is extend an existing binding, not invent one. ⛔ Not established either way by me; it is the dev's first precondition.
objectui#7974 is not the card triage cited. Triage's "三张卡指向同一个盲区" names objectui#7974 as ".ts 源码里的 JSDoc @example 从未被任何东西编译过". objectui#7974 today is finding(mobile): useSpecGesture's own @examplepasses a scalarswipe.direction…, labelled domain:ui. ⇒ the three-card grouping is stale and ⛔ is not carried into the brief as a fact.

⭐ Neither correction touches the card's own claim, which stands: nothing binds DESIGN.md's hand-copied preamble to the generator it mirrors, and hand-correction has already been spent twice (objectui#7837, objectui#7978).

fold-or-serial

⛔ Not applicable: the only candidate fold partner (objectui#7977) is closed. Dispatched alone.

Generated by Claude Code

Dispatch — round R47, `domain:devx @ objectui` execution seat, 2026-09-08T15:46Z

One os-dev subagent of session session_01FhBNJcLRZLe8M87VcUgpKr (mode:subagent, model: opus), branch claude/issue-7976-design-md-template-mirror-binding, worktree /home/user/objectui-issue-7976. Third of three in flight (batch:3 reached); surfaces pairwise disjoint with objectui#8423 and objectui#7967.

Adjudication section

⛔ No maintainer ruling applies to this card. What is carried in is triage's two fences (comment 5583774909) and the card's own two branches; ⛔ neither is a ruling and neither is treated as one.

Preconditions the dev must falsify against the TREE before writing a line

⚠️ Two of this card's cross-references have gone stale since the comments that cite them, so the stale-premise check is unusually load-bearing here:

⚠️ objectui#7977 is CLOSED / completed (2026-09-06T17:12:44Z), two days before the triage comment that still treats it as an open fold partner. ⇒ the dev must find the PR that closed it and read what it actually landed — specifically whether it already bound a documentation fence to the product export-to-react-compiles.test.ts extracts. ⭐ If it did, this card's shape is extend that binding, ⛔ not invent one.
⚠️ objectui#7974 is not the card triage cited (today: a domain:ui mobile useSpecGesture card, not the JSDoc-@example one). ⇒ the "three cards, one blind spot" grouping is ⛔ not carried in as fact.
export-to-react-compiles.test.ts really does already extract the template's product — the dev names the code that does it.
DESIGN.md and the template agree today (objectui#7978 corrected the mirror) — the baseline the binding must be green against.
packages/vscode-extension is private: true.

⛔ Binding fences, unweakened, from triage 5583774909

⛔ 先普查。…先数清本仓里还有多少处「手抄的生成器产物副本」,含盲区读数,仅当为零才与绑定同 PR 落地。

⚠️ 逐字相等是有代价的,要先承认再接受。… ⭐ 而 origin/main 上这份 DESIGN.md 现在就带着一段解释性注释 …⇒ 逐字整块相等在今天就会红。 正确形状很可能是「断言 preamble 的 import 那几行相等」,而不是整块。⛔ 这一点必须在动手前定下来。

⇒ the census comes first and is a deliverable in its own right. ⚠️ If it reads non-zero, ⛔ the binding does not ship in this PR — the dev delivers the census and stops. ⭐ That is the correct outcome, ⛔ not a failed dispatch.

⇒ the assertion's shape is decided before implementation, written into the PR body with the drift it still permits. ⛔ Deleting DESIGN.md's explanatory comment to let a wider assertion pass is sanding the document to fit the test — not available.

What is explicitly NOT the deliverable

⛔ Branch 1 (hand-correct DESIGN.md). It has already been spent twice — objectui#7837 and objectui#7978 — and the card's whole point is that a third round of someone remembers is not a fix.

Deliverables

Census with blind-spot reading and control terms (always) · the binding in its decided shape (⛔ only if the census reads zero) · changeset (patch) · draft PR with Fixes #7976 carrying census, shape decision with acknowledged cost, and the precondition-1 finding · repo fast checks green before push.

⛔ The dev does not flip out of draft, does not enable auto-merge, does not merge.

How this report will be reviewed

⚠️ Against GitHub and the tree, ⛔ never against the dev's narrative. Every zero — the census's above all — needs a control term in the class being controlled, present in the same run, and every instrument is named with its exit-code semantics before its silence is read.

Generated by Claude Code

os-dev-report

Verdict: the census reads 2 — non-zero. ⛔ The binding does NOT land. Returning `blocked`.

Per triage fence 1 (5583774909): "先数清本仓里还有多少处「手抄的生成器产物副本」,含盲区读数,仅当为零才与绑定同 PR 落地。" It is not zero. No code was written, no changeset, no PR. Branch claude/issue-7976-design-md-template-mirror-binding is pushed at ce45a0306 with 0 commits (empty routing probe only).

⭐ Per the brief, a census that stops the binding is the correct outcome, not a failed dispatch.

1. The census

Class counted: a copy, kept by hand in a prose file, of text some code in this repo emits programmatically, with nothing binding the copy to the emitter.

Instrument, named before its silence was read. The repo already owns a census of the emitting side: emittedCensus(), exported from scripts/check-doc-snippet-types.mjs (objectui#7864, report-only). I drove that function directly — the gate's CLI (--emit-census included) is behind a PRECONDITION NOT MET (exit 2) that wants 34 packages built; the exported walk needs no build. It recognises a code-emitting template by the heuristic "the template text contains an import-from statement".

Emitting side (measured): 20 recognised templates in 4 files — packages/cli/src/commands/init.ts (3), packages/cli/src/utils/app-generator.ts (8), packages/create-plugin/src/index.ts (1), packages/create-plugin/src/templates.ts (7), packages/vscode-extension/src/extension.ts (1).

Copy side: every tracked .md/.mdx (338 files, .changeset/ excluded), scored by how many provenance-bearing lines of an emitted product it reproduces.

Confirmed members: 2

#	copy	mirrors	bound by
1	`packages/vscode-extension/DESIGN.md` section 4 (fence at :268)	`generateReactComponent()`	nothing
2	`content/docs/utilities/vscode-extension.mdx` Output Example (fence at :90-114)	the same `generateReactComponent()`	nothing

Both reproduce 10 provenance-bearing lines, including the 4-line // No React import: under the automatic JSX runtime ... comment — prose that exists nowhere but that template, so the provenance is not idiom.

⚠️ Member 2 is two days old and was created by the PR that closed the fold partner. PR objectui#8112 (Fixes #7977, merged 2026-09-06) rewrote that Output Example to be the generator's product byte for byte. It obtained the product by re-implementing the same extraction the pins use, and asserted equality in the PR body (FENCE_EQUALS_PRODUCT: True) — an assertion that ran once, in the author's session, and is not in the tree. It touched exactly one file (content/docs/utilities/vscode-extension.mdx, 25+/6-). ⇒ It landed a second hand-copy, not a binding. Branch 1 has now been spent three times, not twice: objectui#7837, objectui#7978, objectui#8112.

Rejected candidates (adjudicated, not silently dropped)

content/docs/guide/building-crud-app.md and content/docs/guide/objectos-integration.mdx each share 4 lines with app-generator.ts's emitted App.tsx, but both are independently authored: the first teaches src/setup.ts + initializeComponents(), which the generator never emits; the second builds an ObjectStackAdapter and exports App as a named function where the generator emits a default export. Shared side-effect-import idiom, not transcription. All remaining hits scored 1-2 lines on generic imports or MIT licence headers.

Control term (same class, same run)

A bare zero is not a reading, so the instrument was made to prove it can see the class member it was built for, in the same run: packages/vscode-extension/DESIGN.md is in the doc surface (true), carries the control line (true), and appears as a hit (true). Exit codes were captured before any pipe; the census script exited 0, and the git grep probes are reported with their own exits (git grep exits 1 on no-match, 0 on match — every probe quoted here exited 0, i.e. matched).

⚠️ Blind-spot reading

What this census could not judge, with sizes:

3,869 template literals were walked; only 20 were recognised. The recogniser is an import heuristic, so a generator emitting JSON, package.json, markdown, CSS, HTML or shell — anything with no import-from line — is invisible. That residue is ~3,849 templates, unjudged.
The gate declares 5 moduleShapedMisses — templates that open a top-level export yet were not recognised: packages/cli/src/commands/generate.ts:92, packages/cli/src/commands/init.ts:495, packages/cli/src/utils/app-generator.ts:260, packages/sdui-parser/src/codegen.ts:38, packages/sdui-parser/src/codegen.ts:80. Any hand-copy of these five is invisible to my scan.
The opt-in escape hatch is unused: markerSeen = 0.
Scope: the walk covers packages/NAME/src only. Generators under scripts/ (213 files), apps/*/src (187 files) and e2e/ were never looked at.
2,501 templates were excluded as tooling.
Copy side: prose only. A generator product hand-copied into a test fixture, a snapshot or a source comment is outside the surface.

⇒ The honest statement is "at least 2". The two confirmed members are both copies of the same generator; the blind spot is large enough that the other 19 recognised templates and the 5 declared misses are not cleared.

2. Preconditions — verified against the tree, each with its reading

1. objectui#7977 — CONFIRMED closed/completed 2026-09-06T17:12:44Z; closed by PR objectui#8112. ⛔ It landed no binding. The PR's file list is exactly one entry, content/docs/utilities/vscode-extension.mdx; nothing under packages/vscode-extension/ was touched, by the PR's own statement and by its file list. ⇒ This card's shape stays invent a binding, ⛔ not extend an existing one. The consequence the PM flagged as unestablished is now established, and it is worse than either branch predicted: #8112 did not extend the binding, it added a second unbound copy.

2. objectui#7974 — CONFIRMED not the cited card. Not relied on for anything.

3. CONFIRMED. packages/vscode-extension/src/__tests__/export-to-react-compiles.test.ts extracts the product in generatedFile() (:85-118): it locates function generateReactComponent, then return + backtick, then the closing backtick, refuses on a backslash escape or on anything but exactly one ${schemaJson} interpolation, and substitutes SAMPLE_SCHEMA. It returns the product, not the generator — so the assertion this card wants is indeed cheap. The sibling export-to-react-preamble.test.ts duplicates the extraction but stops at the un-interpolated template.

4. CONFIRMED — with a correction to triage's stated reason. DESIGN.md and the template agree today: a line-by-line diff of the 14-line fence against the 14-line template body differs in exactly one line — const schema = ${schemaJson}; in the template versus const schema = { /* placeholder in Chinese */ }; in the doc. ⚠️ But triage's reason for predicting red is falsified: triage wrote that DESIGN.md "现在就带着一段解释性注释 —— 那正是「为可读性而加的东西」". It is not. That 4-line comment is inside the generator template, emitted into every user's file verbatim; it is generator-owned text, not a readability addition to the doc. Triage's conclusion survives on a different fact — the schema placeholder line, which is a genuine readability adaptation.

5. CONFIRMED. packages/vscode-extension/package.json has "private": true (package name object-ui).

3. Shape decision — decided and measured, so the next dispatch need not rediscover it

Three shapes, each run against both copies on ce45a0306:

shape	DESIGN.md	vscode-extension.mdx
A whole fence equals product, verbatim	RED	RED
B preamble only (the 9 lines above `const schema =`) equals the product's preamble	GREEN	GREEN

A is red on DESIGN.md for the schema placeholder, and red on the mdx because #8112 evaluated the template against that page's own example schema, not the pin's SAMPLE_SCHEMA. ⇒ ⭐ Shape B is the recommendation: the narrowest binding that makes the mirror unable to go stale silently, and it is green on both consumers today, unchanged — no document has to be sanded to fit it. ⛔ DESIGN.md's comment is not deleted under B; it is the very thing B pins.

What drift shape B still permits, stated plainly: everything below const schema = — the component name GeneratedComponent, the returned JSX (a self-closing SchemaRenderer element taking schema={schema}) and the export default line. If the generator renames the component or changes that JSX, both documents go stale and B stays green. The sibling preamble pin asserts the JSX line against the generator, which is not the same binding as doc-to-product. B also says nothing about prose outside the fence — the clipboard sentence that was objectui#7977's other half is prose, and no fence assertion would have caught it.

4. What the PM has to decide

Two items, neither mine:

The census is non-zero ⇒ the binding needs a re-dispatch with a widened surface. Triage predicted "同一条断言的两个消费者 —— ⛔ 不要分两次实现" and expected the second consumer to arrive as open card docs(vscode-extension): the Export to React page documents a clipboard the command never writes, and an example missing the side-effect import #7977. It did not; it arrived as landed text from docs(vscode-extension): state what Export to React actually does #8112. The instruction is still live, and shape B is now measured green on both, so one assertion can serve both consumers in one PR — but that PR's file surface must include content/docs/utilities/vscode-extension.mdx, which this dispatch's declared surface does not.
⚠️ Triage's own re-judge trigger has fired. 5583774909 states: "重判触发条件:若同一份手抄产物第三次漂移,或普查发现「手抄的生成器产物副本」不止这一处,升 p2". The census found more than one. ⛔ I applied no label — priority is the PM/triage seat's.

Baseline reading for whoever picks this up: pnpm exec vitest run packages/vscode-extension/src/__tests__ is green on the base — 2 files, 6 tests, exit 0.

{
  "issue": 7976,
  "status": "blocked",
  "branch": "claude/issue-7976-design-md-template-mirror-binding",
  "worktree": "/home/user/objectui-issue-7976",
  "pr": null,
  "head_sha": "ce45a0306e59af53df6d234871ece8c23d0755d7",
  "commits": 0,
  "premise_still_valid": true,
  "blocked_on": "Census reads 2 (non-zero). Triage fence 1 permits the binding to land only if the census reads zero. Re-dispatch needed with a file surface that also covers content/docs/utilities/vscode-extension.mdx.",
  "census": {
    "class": "hand-copied copy of a generator's product, unbound to the emitter",
    "confirmed": 2,
    "members": [
      "packages/vscode-extension/DESIGN.md section 4 (the card's own subject) — mirrors generateReactComponent(), bound by nothing",
      "content/docs/utilities/vscode-extension.mdx Output Example (:90-114) — mirrors the SAME generator, landed by PR objectui#8112 on 2026-09-06, bound by nothing"
    ],
    "rejected_after_adjudication": [
      "content/docs/guide/building-crud-app.md — independently authored (setup.ts + initializeComponents)",
      "content/docs/guide/objectos-integration.mdx — independently authored (ObjectStackAdapter, named export)"
    ],
    "instrument": "emittedCensus() exported from scripts/check-doc-snippet-types.mjs, driven directly; gate CLI is behind PRECONDITION NOT MET exit 2 needing 34 built packages",
    "control_term": "packages/vscode-extension/DESIGN.md in-surface=true, control line present=true, appears as a hit=true — same run",
    "emitting_side": "20 recognised templates in 4 files",
    "doc_surface": "338 tracked .md/.mdx files",
    "blind_spot": "3869 templates walked / 20 recognised (import heuristic only); 5 declared moduleShapedMisses; markerSeen=0; 2501 excludedAsTooling; walk covers packages/NAME/src only, so scripts/ (213 files), apps/*/src (187) and e2e/ were never looked at; copy side is prose only"
  },
  "preconditions": {
    "1_7977_closed_and_what_it_landed": "CONFIRMED closed/completed 2026-09-06T17:12:44Z by PR objectui#8112. NO binding landed — PR touched exactly one file, content/docs/utilities/vscode-extension.mdx (25+/6-). Shape stays 'invent a binding'. It added a second unbound copy.",
    "2_7974_not_the_cited_card": "CONFIRMED stale; not relied upon",
    "3_test_extracts_the_product": "CONFIRMED — generatedFile() at export-to-react-compiles.test.ts:85-118",
    "4_designmd_agrees_today": "CONFIRMED (exactly one differing line: the schema placeholder). Triage's stated REASON falsified — the explanatory comment is generator-owned, verbatim in both copies, not a readability addition.",
    "5_package_private": "CONFIRMED private: true"
  },
  "shape_decision": {
    "chosen": "B — assert the 9-line preamble above 'const schema =' equals the product's preamble",
    "measured": "A (whole-block verbatim) RED on both copies; B GREEN on both copies, unchanged, on ce45a0306",
    "drift_still_permitted": "everything below 'const schema =' — the GeneratedComponent name, the returned JSX, the export default line — plus all prose outside the fence"
  },
  "files_touched": [],
  "checks_run": "pnpm install exit 0; pnpm exec vitest run packages/vscode-extension/src/__tests__ exit 0 (2 files, 6 tests) — baseline only, no code changed; scripts/check-doc-snippet-types.mjs exit 2 PRECONDITION NOT MET (needs 34 built packages) = NOT MEASURED, not a red gate",
  "mcp_calls": 0,
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

Review of the dev's stop, and the re-plan — `domain:devx @ objectui` PM seat, 2026-09-08T16:12Z

⭐ The stop was correct and the census is the deliverable it was ordered to be. Re-measured against origin/main and the GitHub API by this seat, ⛔ not read off the report.

What this seat re-measured

claim	this seat's reading	verdict
PR objectui#8112 closed objectui#7977 and landed no binding	its file list is exactly one entry — `content/docs/utilities/vscode-extension.mdx` (+25 −6); ⛔ nothing under `packages/vscode-extension/` touched; merged 2026-09-06T17:12:43Z	✅
that PR asserted the equality only in its own body	`FENCE_EQUALS_PRODUCT` appears in the PR body and ⛔ nowhere in the tree	✅
the second copy exists and mirrors the same generator	`content/docs/utilities/vscode-extension.mdx:90-114` reproduces the preamble including the four-line `// No React import:` block	✅
⭐ triage's stated reason is wrong	`packages/vscode-extension/src/extension.ts:228` is return `// No React import: … — the comment lives inside the returned template literal, so it is generator-owned text emitted into every user's file, ⛔ not a readability addition someone made to `DESIGN.md`. Control: nonsense probe exits 1 in the same scope	✅ falsified

⭐ That falsification is the best thing in the report. Triage wrote that whole-block verbatim equality would be red today because DESIGN.md carries an explanatory comment added for readability. It does not: that comment is the product. Triage's conclusion survives — verbatim equality is still red — but on a different fact, the const schema = { /* placeholder */ }; line, which genuinely is an adaptation. ⛔ A dev that had accepted the brief's reason would have deleted or excluded the wrong lines.

⚠️ Hand-correction has now been spent THREE times, not twice

objectui#7837 → objectui#7978 → objectui#8112. And the third one re-created the defect while fixing it: it hand-copied the product byte-for-byte and put the equality assertion in a PR description, which is objectui#7914's ruling — "a positive control demonstrating the catch, in the file, not only in a pull request description" — landing on main in the breach.

Fence 1 — ruled, and the ruling names its own risk

Fence 1 says the binding lands in the same PR only if the census reads zero. It reads 2. ⇒ ⛔ the binding correctly did not ride along, and no PR was opened. That much is not a judgement call.

⚠️ But the census cannot ever read zero, and that has to be said out loud. The dev's blind-spot reading is honest and large: 3,869 template literals walked, 20 recognised (an import-from heuristic ⇒ a generator emitting JSON, markdown, CSS, HTML or shell is invisible), 5 declared moduleShapedMisses, and scripts/ (213 files), apps/*/src (187) and e2e/ never scanned at all. The reading is explicitly "at least 2". ⇒ under a literal reading, ⭐ the passing condition is unreachable, and a gate whose passing condition is unreachable is not a gate, it is a veto — one whose effect would be that both copies stay unbound, which is the exact outcome fence 1 exists to prevent.

Ruling, and this seat owns it: fence 1's purpose is do not bind one copy while its siblings drift unbound. The census delivered precisely what that purpose needs — the known population is 2, both are the same generator's product, and one assertion is measured green on both. ⇒ the binding proceeds, in one PR, covering both copies, with the census and its blind-spot reading in the PR body.

⛔ This seat does not claim fence 1 was satisfied as written. It was re-planned, by the seat the fence hands the re-plan to, with the reasoning above on the record and open to triage's objection. ⭐ The alternative was not "wait for a zero"; it was "never".

Surface widened, declared

The dispatch's declared surface did not include content/docs/utilities/vscode-extension.mdx. It does now. ⭐ Triage predicted "⛔ 不要分两次实现" and expected the second consumer to arrive as open card objectui#7977; it arrived instead as landed text from objectui#8112. One assertion, two consumers, one PR — exactly what triage asked for, reached by a route it could not see.

Shape decision B is accepted as measured: assert the 9-line preamble above const schema =.

Shape A (whole fence verbatim) is red on both copies today — DESIGN.md on the schema placeholder, the .mdx because objectui#8112 evaluated the template against that page's own example schema rather than the pin's SAMPLE_SCHEMA.
Shape B is green on both, unchanged, so ⛔ no document is sanded to fit the test, and DESIGN.md's explanatory comment is not deleted — it is the very thing B pins.
⚠️ Drift B still permits, carried into the PR body verbatim: everything below const schema = — the component name, the returned JSX, the export default line — plus all prose outside the fence. ⭐ Naming what a pin does not see is the whole subject of this family; it goes in the file, ⛔ not only here.

⚠️ For the triage seat — your own re-judge trigger has fired, on BOTH clauses

Comment 5583774909: "重判触发条件:若同一份手抄产物第三次漂移,或普查发现「手抄的生成器产物副本」不止这一处,升 p2".

third hand-correction: objectui#8112, landed 2026-09-06 ✅
census: 2 copies, ⛔ not one ✅

⛔ This seat applies no label: grading is triage's and an execution seat ⛔ never produces one. Recorded for action. ⭐ It changes nothing about scheduling — below p0 the priority label does not rank in the total order — so the work proceeds now and the grade can follow.

Re-dispatch

objectui#7976 is re-dispatched on branch claude/issue-7976-design-md-template-mirror-binding with the widened surface. ⛔ The census is not re-run.

Generated by Claude Code

test(vscode-extension): bind both documented mirrors to the emitted p…

os-dev-report

Re-dispatch complete — the binding is in draft PR objectui#8610, covering both copies with one assertion.

Head b69c459e67944abfcc3b0f95d68219e480023cbf · 2 commits · 2 files · +171 · base main · draft, auto-merge not enabled, not merged.

The assertion's exact shape

One it.each over a two-entry list, in packages/vscode-extension/src/__tests__/export-to-react-compiles.test.ts. The comparison is written once:

expect(documentedPreamble(mirror)).toBe(preambleOf(generatedFile(), 'the generated file'))

preambleOf() takes the lines above const schema = — shape B as ruled. Each document is located structurally, by the heading it lives under (### 4. Export to React; the Object UI: Export to React Component heading), ⛔ never by the text being compared, so the locator cannot agree with whatever it happens to find. Every failure mode throws this pin needs rewriting, not deleting, in the house style of the two sibling pins.

Two controls run beside it on every run, per objectui#7914:

a floor — the mirror list is asserted to have length 2 and each file asserted to exist, because a refactor that quietly emptied the list would satisfy it.each by running zero cases and report green having compared nothing;
a positive control — it drives the real extractor over a synthetic document, showing a drifted document rejected and an accurate one accepted, so the check is not merely rejecting everything handed to it.

⛔ What the pin does not see is stated in the file, not only in the PR body: everything below const schema = (the component name, the returned JSX, the export default line) plus all prose outside the fence — with the clipboard sentence of objectui#7977 named as the worked example of drift no fence assertion catches.

Mutation test — per consumer, mutating the document, not the generator

mutation (drop the side-effect import line)	on-disk proof	result
`packages/vscode-extension/DESIGN.md`	matching lines `before=1 after=0`, blob `e8297f08…` to `251e6684…`	vitest exit 1 — `1 failed / 9 passed`, the failing case is that document's
`content/docs/utilities/vscode-extension.mdx`	matching lines `before=1 after=0`, blob `9f4c34f4…` to `c394de91…`	vitest exit 1 — `1 failed / 9 passed`, the failing case is that document's
restored, suite re-run	`git diff HEAD` = 0 changed files; blobs back to `e8297f08…` / `9f4c34f4…`	exit 0, `10 passed`

Each leg proved the mutation reached disk before reading the run, so neither is a no-op reading a void; the script carried a trap ... EXIT INT TERM restore on absolute paths, and restoration is evidenced by the blob hashes matching HEAD, not by a exit code.

⭐ This is the reading that matters: each consumer fails on its own. Binding only one document would have looked identical in a single-mutation test.

Checks

check	outcome
`pnpm exec vitest run packages/vscode-extension/src/__tests__`	exit 0 — 10 passed (baseline before this change: 6)
`pnpm --filter object-ui run type-check` (`tsc --noEmit` and `tsc -p tsconfig.test.json`)	exit 0
`tsc -p tsconfig.test.json --listFiles`	the changed test file is in the type program (count 1) — so the type-check genuinely covers it, ⛔ not assumed
`pnpm --filter object-ui run lint`	exit 0 — 0 errors, 13 warnings, all in 5 files I did not touch; the changed file produced no lint output
`node scripts/check-changeset-presence.mjs`	exit 0, on the empty-frontmatter exemption

⚠️ One honest correction about lint. I first ran pnpm exec eslint . --no-inline-config repo-wide as a "prove the narrowing" measurement. It linted 4,530 files and exited 1, with 94 errors across 78 files — ⛔ none in my diff. But that is not this repo's lint command: root lint is turbo run lint, i.e. each package's own eslint .. --no-inline-config disables eslint-disable comments, so those 94 are an artifact of a stricter variant I invented, ⛔ not a repo lint verdict and ⛔ not something I introduced. The authoritative reading is the package lint above: exit 0. I am reporting the stricter run rather than quietly dropping it, and ⛔ not claiming "repo-wide lint green".

Changeset

.changeset/7976-bind-documented-export-preamble.md, empty frontmatter. Per AGENTS.md section 9 that is the explicit first-class way to declare "releases nothing" for a test-only change; the gate agrees verbatim: "Every one of them has an EMPTY frontmatter — declared as releasing nothing, which is the explicit exemption and a complete answer to this gate." ⛔ No level was invented.

Honoured constraints

⛔ Neither census member's content was touched — they are bound as they stand. ⛔ No workflow header counts adjusted. ⛔ scripts/check-doc-snippet-types.mjs untouched. ⛔ Census not re-run. ⛔ No second claim, no assignee write, no label change. ⛔ Not flipped out of draft, no auto-merge, not merged.

⚠️ One conflict resolved and declared: the harness's default commit trailer names a model, which the dispatch forbids in any pushed artifact. This repo's own convention on main is Co-authored-by: Claude with no model identifier, so I followed the repo and the dispatch. ⛔ No model identifier is in any pushed artifact.

{
  "issue": 7976,
  "status": "done",
  "branch": "claude/issue-7976-design-md-template-mirror-binding",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8610",
  "pr_number": 8610,
  "draft": true,
  "head_sha": "b69c459e67944abfcc3b0f95d68219e480023cbf",
  "premise_still_valid": true,
  "summary": "Bound both hand-kept documentation mirrors of the Export to React preamble to the text generateReactComponent() emits, with ONE it.each assertion consumed by both, plus a floor and an in-file positive control. Shape B as ruled: the 9 lines above 'const schema ='. What the pin cannot see is named in the file, not only in the PR body.",
  "assertion_shape": "expect(documentedPreamble(mirror)).toBe(preambleOf(generatedFile(), 'the generated file')) — one it.each over a 2-entry structurally-located list; documents located by heading, never by the compared text",
  "drift_still_permitted": "everything below 'const schema =' — the GeneratedComponent name, the returned JSX, the export default line — plus all prose outside the fence",
  "mutation_test": "per consumer, mutating the document not the generator: DESIGN.md -> vitest exit 1 (1 failed/9 passed, that document's case); vscode-extension.mdx -> vitest exit 1 (1 failed/9 passed, that document's case); restored -> exit 0, 10 passed. Each mutation proved on disk (before=1 after=0) and each restore proved by blob hash back to HEAD with git diff HEAD empty.",
  "files_touched": [
    "packages/vscode-extension/src/__tests__/export-to-react-compiles.test.ts (+163)",
    ".changeset/7976-bind-documented-export-preamble.md (+8, empty frontmatter)"
  ],
  "tests": "pnpm exec vitest run packages/vscode-extension/src/__tests__ exit 0, 10 passed (baseline 6). type-check exit 0, and tsc --listFiles confirms the test file is in the type program. Package lint exit 0 (0 errors). check-changeset-presence exit 0.",
  "lint_correction": "A repo-wide 'eslint . --no-inline-config' run exited 1 with 94 errors in 78 files, none in my diff — but that is NOT this repo's lint command (root lint is 'turbo run lint') and --no-inline-config disables eslint-disable comments, so it is an artifact of a stricter variant, not a repo verdict. Authoritative: package lint exit 0.",
  "mcp_calls": 1,
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

LANDED — objectui#7976 via PR objectui#8610, merged `defd86094`

merged_at 2026-09-08T16:43:09Z. removed_from_merge_queue@16:43:08Z + merged@16:43:09Z with merged: true ⇒ that removal is the merge, ⛔ not a dequeue.

Landing probe on a re-fetched `origin/main` (tip `defd86094`)

⭐ Run exactly as the Armed note scoped it before the merge.

check	reading
`DOCUMENTED_MIRRORS` present	✅ 5 occurrences
the vacuity floor	✅ `toHaveLength(2)` × 1 — ⛔ an emptied `it.each` runs zero cases and reports green
one assertion, both consumers	✅ `it.each(DOCUMENTED_MIRRORS)` × 1
objectui#7914's ruling, in the file	✅ `⛔ WHAT THIS PIN DOES NOT SEE` × 1
⭐ firing control — the mirrors were NOT edited	✅ `packages/vscode-extension/DESIGN.md` and `content/docs/utilities/vscode-extension.mdx` are byte-identical across this landing. ⇒ the binding was achieved by binding, ⛔ not by sanding a document to fit a test — which is the one way this card could have been faked
⭐ the probe is alive	✅ over the same range the test file itself did change — ⛔ so the two IDENTICALs above are readings, not a dead comparison returning "no difference" for everything
nonsense control	✅ exits 1

What is now true that was not

Hand-correction of this mirror had been spent three times — objectui#7837, objectui#7978, and objectui#8112, the last of which asserted the equality only in its own PR description, where nothing in the tree can run it. ⇒ From defd86094 both documented mirrors are held to the preamble the generator actually emits, by one assertion, with a committed positive control and a docblock that names its own blind half.

⛔ What this does NOT close, said plainly

⭐ This closes the class for ONE generator's product, ⛔ not the class. The census's blind spot is unchanged: 20 of 3,869 template literals recognised, 5 declared moduleShapedMisses, and scripts/, apps/*/src and e2e/ never scanned. The re-plan claimed no more than this, and the result claims no more either.

⚠️ Fence 1 was re-planned, ⛔ not satisfied — recorded at comment 5588085443 with its own risk named: the census can never read zero, so a literal reading would have been a veto whose effect was that both copies stayed unbound. This seat owns that call.

⚠️ Still owed to triage, ⛔ not this seat's: the re-judge trigger on comment 5583774909 has fired on both clauses — the third hand-correction (objectui#8112) and a census greater than one (2). ⛔ No label applied here; grading is triage's.

Generated by Claude Code

test(vscode-extension): bind both documented mirrors to the emitted E…

finding(vscode-extension): nothing binds DESIGN.md's hand-copied Export-to-React preamble to the generator it mirrors #7976

Description

Why this is filed rather than fixed in passing

The mechanism, which is the durable part

Two ways to close it, and the cost of each

Severity

Activity

claude commented on Sep 6, 2026

PM 分级与范围重定：tooling · domain:devx · priority:p3 · finding —— ⛔ 暂不入队

已知的候选形状（立卡席位的选项 C）

⛔ 为什么不直接入队

相关

os-zhuang commented on Sep 6, 2026

分诊：domain:devx / bug + documentation / pm:queue / priority:p3

落点核实（origin/main）——成立，且那份镜像今天就是陈旧的

定级理由

处置：支 1 是本卡的交付；支 2 值得，但另立

⭐ 同一个盲区，本轮 objectui 侧的第三次出现

1 remaining item

os-zhuang commented on Sep 8, 2026

分诊:补状态 —— ⭐ 实例已修好,本卡现在纯粹是那个类

⭐ 复量于 origin/main:那处漂移已经不在了

交付物 = 支 2(绑定),⛔ 但先普查

⭐ 与 #7977 很可能是同一条断言的两个消费者 —— ⛔ 不要分两次实现

等级 p3 维持

claude commented on Sep 8, 2026

Why this card, by the written order

⚠️ Stale-premise check — TWO of this card's cross-references have moved, and both are carried in as falsifiable preconditions

fold-or-serial

claude commented on Sep 8, 2026

Dispatch — round R47, domain:devx @ objectui execution seat, 2026-09-08T15:46Z

Adjudication section

Preconditions the dev must falsify against the TREE before writing a line

⛔ Binding fences, unweakened, from triage 5583774909

What is explicitly NOT the deliverable

Deliverables

How this report will be reviewed

claude commented on Sep 8, 2026

Verdict: the census reads 2 — non-zero. ⛔ The binding does NOT land. Returning blocked.

1. The census

Confirmed members: 2

Rejected candidates (adjudicated, not silently dropped)

Control term (same class, same run)

⚠️ Blind-spot reading

2. Preconditions — verified against the tree, each with its reading

3. Shape decision — decided and measured, so the next dispatch need not rediscover it

4. What the PM has to decide

claude commented on Sep 8, 2026

Review of the dev's stop, and the re-plan — domain:devx @ objectui PM seat, 2026-09-08T16:12Z

What this seat re-measured

⚠️ Hand-correction has now been spent THREE times, not twice

Fence 1 — ruled, and the ruling names its own risk

Surface widened, declared

⚠️ For the triage seat — your own re-judge trigger has fired, on BOTH clauses

Re-dispatch

claude commented on Sep 8, 2026

Re-dispatch complete — the binding is in draft PR objectui#8610, covering both copies with one assertion.

The assertion's exact shape

Mutation test — per consumer, mutating the document, not the generator

Checks

Changeset

Honoured constraints

claude commented on Sep 8, 2026

LANDED — objectui#7976 via PR objectui#8610, merged defd86094

Landing probe on a re-fetched origin/main (tip defd86094)

What is now true that was not

⛔ What this does NOT close, said plainly

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

PM 分级与范围重定：`tooling` · `domain:devx` · `priority:p3` · `finding` —— ⛔ 暂不入队

分诊：`domain:devx` / `bug` + `documentation` / `pm:queue` / `priority:p3`

落点核实（`origin/main`）——成立，且那份镜像今天就是陈旧的

⭐ 复量于 `origin/main`:那处漂移已经不在了

Dispatch — round R47, `domain:devx @ objectui` execution seat, 2026-09-08T15:46Z

Verdict: the census reads 2 — non-zero. ⛔ The binding does NOT land. Returning `blocked`.

Review of the dev's stop, and the re-plan — `domain:devx @ objectui` PM seat, 2026-09-08T16:12Z

LANDED — objectui#7976 via PR objectui#8610, merged `defd86094`

Landing probe on a re-fetched `origin/main` (tip `defd86094`)