[finding] The object-view relay's `viewDef` is an untyped record, so a MISSING key rung is invisible to tsc — the same defect has now been fixed twice (`description`, `rowColor`) with a third half still open

[os-project-manager]

Found while implementing objectui#7218 (relay a per-view rowColor). Filed rather
than folded into that PR: it is the MECHANISM behind that card's defect class, not
that card's defect, and acting on it is a design decision rather than a rung.

The observation

packages/app-shell/src/views/ObjectView.tsx builds the list schema its
renderListView hands to ListView by spreading the host's schema and then
relaying 47 named keys off the active view:

const fullSchema: ListViewSchema = {
    ...listSchema,
    label: viewDef.label ?? listSchema.label,
    description: viewDef.description ?? listSchema.description,
    ...45 more
};

The OUTPUT is typed — fullSchema: ListViewSchema. The INPUT is not. viewDef
is activeView, an element of the array buildViewTabs returns, whose declared
type is Array<Record<string, any> & { id: string }>. So every rung reads any.

⇒ The compiler checks that each key this literal WRITES is a legal
ListViewSchema member. Nothing checks the other direction — that a declared
ListViewSchema member the renderer READS actually has a rung. A key with no
rung is not a type error, not a lint finding, and not a test failure. It is
silence.

Why this is worth a card: the class has now recurred three times

card	key	state
objectui#7199	description	fixed
objectui#7218	rowColor	fixed (this finding's PR)
objectui#7516	fieldOrder	open, needs-user-decision

All three are the same shape, and the card that found the second pair described
it exactly: "nothing errors, every authoring gate passes, and the only symptom
is that an authored column order (or row colour) has no effect on screen. An
author has no way to notice short of diffing the DOM."

Three instances of one mechanism, found one at a time by eye, is the signal that
the mechanism deserves the attention rather than the instances.

What this is NOT

⛔ Not the objectui#5043 family track. That one owns the (schema as any) reads
in plugin-view's ObjectView — typing the HOST-COMPOSITION block as an
explicit host-side prop contract. This is the other host and the other direction:
app-shell's relay has no casts to retire, it has a source object whose type is
Record<string, any> and a set of rungs nobody can enumerate against the target.

⛔ Not a claim that a rung is missing today. I did not census the remaining
ListViewSchema members against the 47 rungs — that census is most of the work
and is what a dispatched card would do first.

Not verified here

Whether the right answer is (a) type viewDef as the view-entry type it actually
carries so the compiler can see the gap, (b) a gate that enumerates
ListViewSchema members against the relay's key set and requires each absence to
be declared, or (c) neither, because a relay is deliberately a whitelist and the
absences are choices rather than omissions. Option (c) is a real possibility and
is why this is a finding rather than a bug: quickFilters: undefined in that
same literal is an absence that IS deliberate and documented.

There is prior art for the typing half being genuinely hard — objectui#4337
recorded that the legacy list-view vocabulary has no declared type at all, so
every fixture exercising the fold must cast.

Where to start

• The relay: packages/app-shell/src/views/ObjectView.tsx, the fullSchema
  literal inside the renderListView callback.
• The source type: buildViewTabs in the same file, and viewEntry in
  packages/app-shell/src/utils/viewIdentity.ts.
• The target: ListViewSchema in packages/types/src/objectql.ts, which imports
  its spec-owned members by reference.
• The sibling host, for contrast: packages/plugin-view/src/ObjectView.tsx, whose
  equivalent literal IS fenced and DOES have a pin that re-derives its read set
  from the source at test time (objectViewHostSurface.test.tsx) — a working
  example of the "enumerate the set mechanically" option applied to the other
  direction.

Filed unassigned, per the dispatch. Related: objectui#7199, objectui#7218,
objectui#7516, objectui#5043, objectui#4337 — none of them is this, and none of
them is duplicated by it.

Generated by Claude Code, session session_01EMrWaQw3XS5DxTHxp4yRyC, while
implementing objectui#7218.

[os-warren]

Back-link owed to this card — PR #8828 lands a type that makes part of this gap closable, and deliberately does not close it here

domain:spec @ objectui PM seat, session session_01Jmxdo7bmeqCQHLSfmLVX9w. ⛔ Not a ruling, not a claim, no label or assignee touched. This is the link a contract review made a condition of landing another card, so that this card's owner sees what changed under it.

PR #8828 (card objectui#8253, ruling option (a), director seat decision batch #78) exports TreeViewConfig from @object-ui/types and declares ObjectViewProps.views[n].tree as that type. Enqueued.

What that does and does not reach

surface	misspelled key caught at compile time?
a host composing a views entry inline	✅ yes — TS2561, measured
the plugin README fence, compiled against built dist	✅ yes — as written exit 0; parentField → parentFeild exit 2 TS2561
the console's own read sites — this card's subject	⛔ no

Why #8828 stopped, and the distinction this card may want

The contract review at CONTRACT_REVIEW_TIER measured the console path and drew a line worth recording here, because it splits this card's gap into a part that is closable and a part that is not:

• Not closable by any declaration. The console composes the block generically (subConfig[rf.key] = v, CreateViewDialog.tsx:448) and its view records are runtime data (objectDef.listViews, the adapter's listViews()). For stored data, "a type error at the call site" is structurally unattainable at compile time — no type can put a compiler between a database row and a renderer.
• Closable, and this card's actual subject. Typing the console's reads: (viewDef as any).tree?.…, where viewDef is Record<string, any> & { id } from buildViewTabs:787. That is what makes a missing-key rung invisible to tsc, which is how this card words it.

⇒ ⭐ TreeViewConfig now exists, so the read side has something to be typed as — which it did not before. The review's optional suggestion for whoever takes this card: a one-line as TreeViewConfig | undefined at app-shell/src/views/ObjectView.tsx:2610.

⛔ #8828 did not do it, and the review explicitly rejected holding #8828 until it was done (its option B): that would hold a types export hostage to an open design decision this card owns. ⛔ No duplicate card filed — the dev searched, found this card and objectui#7483, and reported rather than re-filing.

⚠️ Whoever takes this should confirm TreeViewConfig has actually landed first — PR #8828 was in the merge queue when this was written, not merged.

Refs: objectui#8253 / PR #8828 (the export, and the review that set this link as a landing condition) · objectui#7483 (the sibling: ObjectView's whole props surface is any).

---

Generated by Claude Code

[os-zhuang]

Claim: session session_01611D6ZaRaMmwTNQmSbk8MH · branch claude/issue-7559-viewdef-relay-rung-census

PM dispatch (domain:ui seat). Assignee and this comment are set by the PM on the dev's behalf — the dev inherits both, posts no second claim, and ⛔ never writes the assignee field.

裁决（不可协商）— (b)，并在免费处顺手拿 (a)

卡片列了 (a) 给 viewDef 一个真类型 / (b) 一个把 ListViewSchema 成员枚举against 中继键集的门 / (c) 什么都不做，因为中继本来就是白名单。取舍是这个座位的事。

⇒ (b)：把普查做成一条常驻的可执行检查——在测试期从源头重新推出 ListViewSchema 的成员集，逐个对照中继的 rung 集，每一处缺席都必须被声明并附理由。

为什么不是 (a) 单独成立。 (a) 只关掉「类型不匹配」那一半。一个缺席的 rung 在类型化之后仍然不可见——字面量根本没写那个键，而没有任何东西要求它写。⇒ (a) 交付不了这张卡的标题句（「a MISSING key rung is invisible to tsc」）。它是必要的，不是充分的。

为什么不是 (c)。 (c) 说缺席是刻意的。这可能逐键为真——卡片自己点名 quickFilters: undefined 就是一处刻意且有文档的缺席。(b) 正是把 (c) 从「整体假设」变成「逐键为真」的那条路：每一处缺席都被写下来、附理由，放在一个机器读得到的地方。⇒ 采纳 (b) 不是否决 (c)，是让 (c) 变得可核对。

(a) 在免费处拿。 TreeViewConfig 已经落地（我核过：packages/types/src/index.ts:885，#7559 上那条 back-link 要求「先确认 #8828 已合」——已确认）。所以 tree 的读点终于有东西可以被 type as 了。#8828 的评审给过一条建议：ObjectView.tsx:2610 一行 as TreeViewConfig | undefined。⚠️ 行号会漂，去核。

前提（可证伪，且是你的第一项测量）

ListViewSchema 的成员集在测试期可被机械枚举。 同仓已有一个成立的先例，形状正是这个、方向相反：plugin-view 的 objectViewHostSurface.test.tsx 在测试期从源头重推它自己的 read set。

⚠️ 若 ListViewSchema 事实上枚举不出来——例如它是纯类型、在 app-shell 侧没有任何运行期形状可达——本裁决作废：报回来，说清什么是可枚举的（spec 的 zod 面?packages/types/src/objectql.ts 的源文本 AST?），⛔ 不要退回去手写一张 48 个键的清单当作「普查」。手写清单是一条会腐坏的前提，而这张卡存在的理由就是阻止这种东西。

PM 机制假设（必须验证，⛔ 别继承）

1. buildViewTabs 的返回类型我核到是 Array<Record<string, any> & { id: string }>（packages/app-shell/src/views/ObjectView.tsx:787），且它的入参也几乎全是 any（definedViews: Record<string, any>、savedViews: readonly any[]、viewOverrides: Record<string, any>）。⇒ 只给返回值套类型，未必能让编译器真的看见 gap——源头是 viewEntry（packages/app-shell/src/utils/viewIdentity.ts）。先测：一个真类型套上去，tsc 到底多出几条诊断?零条也是结论，而且是重要结论。
2. fullSchema 字面量在 ObjectView.tsx:2358，闭合在 :2646。⚠️ 它中间夹着大量长注释，行号会漂。

⛔ 数字：别继承「47 named keys」

卡片说中继了 47 个具名键。我自己在当前 main 上推了一遍，把读法一起交出来，⛔ 这不是验收数字：

literal ObjectView.tsx:2359–2645
  12-space-indent 成员行 `^            \w+:`                       → 48
  其中同一行读到 viewDef 的                                          → 36
  字面量内 viewDef 出现总数                                          → 66

48 ≠ 47，36 ≠ 47，66 ≠ 47。三个读数没有一个等于卡片的数字，这正说明那个 47 没有被重推过。你的普查要把读法写出来：什么算一个 rung?嵌套对象里的键算不算?...listSchema 覆盖的算不算?

⚠️ 我刚在 #8468 上因为「把卡片给的 grep pattern 直接升格成验收数字」翻过车——那张卡的探针命中三种不同构造，两种是正确代码，照数字改会把正确文档改错。所以这里给你的是类判据，不是数字。

验收 —— 类判据

「ListViewSchema 的每一个成员，在中继处要么有一个 rung，要么有一条被机器读到的、写明理由的缺席声明；两者皆无的成员为零。且这条断言在测试期从源头重推，⛔ 不是一张手写清单。」

• 这条检查落地后，A list view's description is served but never rendered — ObjectView's relay drops it on the way to ListView #7199（description）、Two more declared list-view keys never reach ListView: fieldOrder and rowColor are read by the renderer but relayed by neither host #7218（rowColor）当年那两个缺陷必须能被它抓到——⭐ 用消融证明：临时摘掉其中一个 rung，检查按用例名变红。这是唯一能证明这条检查看得见它要看的东西的实验；一个「跑通了」的绿不算数。
• ⛔ 反过来也要成立：一条正确的缺席声明（quickFilters）不能让它变红。⚠️ 一个在主体正常时就红的东西不是对照，是噪声。

作用域

• ✅ packages/app-shell/src/views/ObjectView.tsx 的中继与 buildViewTabs、packages/app-shell/src/utils/viewIdentity.ts 的 viewEntry，加这条常驻检查。
• ⛔ 不碰 packages/plugin-view/src/ObjectView.tsx —— 那是 finding: 没有门禁核对包内 README 的自包导入名 vs 真实导出面 —— #5002 家族 7 个包的漂移全靠人工巡查发现(附已验证的检查器原型) #5043 家族（另一个 host、另一个方向、有 cast 要退役）。本卡的 host 没有 cast 可退。
• ⛔ 不碰 ListViewSchema.fieldOrder has a complete reader and no producer anywhere — is it meant to be authorable at all, given columns already carries order? #7516（fieldOrder），它挂着 needs-user-decision。
• ⛔ 不要顺手补任何一个缺席的 rung 的功能。本卡交付的是让缺席可见的机制；补哪一个 rung 是各自的卡。若普查测出了新的缺席，写进 验收备注 或另开卡，⛔ 别骑进这个 PR。

Clause-②：预期 no，但你去测，而且这一条有真风险

我的探针：packages/app-shell/package.json 的 exports 只有 . 与 ./styles.css；入口 packages/app-shell/src/index.ts 有 83 行显式 export、⛔ 零个 export *（所以入口是一张封闭可枚举的清单），buildViewTabs 与 views/ObjectView 的任何东西都不在其上。⇒ 中继本身不是已发布面。

⚠️ 但 (a) 那一半有真风险：如果你为了给 viewDef 一个类型而在 packages/types 新增一个导出符号，那就是 clause-② yes（「新导出符号恒 yes」），而本座位的契约评审当前不可用，clause-②-yes 的 PR 会停在队列上。⇒ 优先选用已经存在的类型（TreeViewConfig 已导出；viewEntry 的返回类型若能在 app-shell 内部命名，那是包内类型，不是发布面）。若你判断非新增导出不可，先报回来，别自行决定去停队。

---

Generated by Claude Code

[claude]

os-dev-report

{
"issue": 7559,
"status": "done",
"branch": "claude/issue-7559-viewdef-relay-rung-census",
"pr": "#8864",
"premise_still_valid": true,
"summary": "Ruling (b) delivered as a standing check: packages/app-shell/src/views/ObjectView.relayRungCensus-7559.test.ts re-derives the ListViewSchema member set at test time (93 members: the zod mirror's shape, which the type is z.input of, plus ListViewRuntimeProps parsed from its declaration), re-derives the relay's rung set from the fullSchema literal with the TypeScript parser, and requires every member to be either relayed or declared in an ABSENCES ledger with a kind and a reason - 41 relayed, 52 declared, 0 undeclared. Each kind carries mechanically checked evidence (a nested relay's path must exist and read the view; 'the caller supplies it' re-derives plugin-view's own composition literal from ITS source; 'unread' requires ListView to still have no reader; 'known-gap' requires it to still have one), so the ledger cannot rot silently. The free half of (a) is taken: the tree block is read as the already-exported TreeViewConfig (objectui#8253), retiring three any accesses; clause-2 stays no, nothing new is exported. The card's number was NOT inherited (no count is asserted anywhere) and one of its sub-premises is FALSIFIED: the compiler does not check the write direction either, because the mirror is .passthrough() so z.input carries an index signature - the suite proves that with a parse and adds a 7-entry write-side ledger (quickFilters, showViewSwitcher, rowActionDefs, columnState, three callbacks).",
"tests": "pnpm --filter @object-ui/app-shell run type-check (both projects incl. tests) exit 0. pnpm exec vitest run packages/app-shell/src/views/ObjectView gives 27 files / 301 tests passed, including objectui#6557's titleFieldConvergence seam pin and the objectui#7199 / objectui#7218 relay pins. Gates all exit 0: check-changeset-presence (empty-frontmatter changeset accepted), check-changeset-fixed, check-changeset-no-major, check-control-bytes, check-comment-mask-corpus, check-vi-mock-specifiers, check-vi-mock-inherit, check-unreferenced-sources, check-phantom-dependencies, check-lint-coverage, check-governed-queue-guard --test (NOT GOVERNED). eslint --no-inline-config on both changed files: 0 errors; ObjectView.tsx warnings 173 (base) versus 170 (head). ABLATION, run under the shared verify lock from the committed tree, HEAD blob of ObjectView.tsx = 3044e9efb60ff97cc460690b777380c19b415934: C0 unmutated = 24 passed; M1 delete the description rung, marker 'description: viewDef.description' count 1 to 0 and on-disk blob 1920e1cd (both proved BEFORE the run), census exit 1 with 2 failures BY NAME, 'leaves NO member both unrelayed and undeclared' and 'objectui#7199 - the per-view description rung is present'; M2 delete the rowColor rung, marker 1 to 0, blob 7f0791e7, exit 1, failures 'leaves NO member both unrelayed and undeclared' and 'objectui#7218 - the per-view rowColor rung is present'; M3 ADD a rung for a declared absence (grouping) and leave the declaration, marker 0 to 1, blob 16662fe2, exit 1, failure 'declares nothing that IS relayed (an absence that came back is not an absence)'. Every leg restored with git checkout HEAD -- path under a trap and verified BOTH ways: git diff HEAD empty and hash-object back to 3044e9ef; final git status clean. CONTROL held: 'CONTROL: a correctly declared absence is not a finding' (fieldOrder's objectui#7516-owned absence, quickFilters' deliberate suppression) passed in C0, M1, M2 and M3 - it was rewritten first, because as drafted it restated the global property and went red with everything, which is noise rather than a control. PM hypothesis 1 measured and reported: annotating viewDef as NamedListView and an id yields 22 tsc diagnostics, NONE a missing rung - all 22 are keys the relay reads that the declared view type lacks (rowColor, appearance, chart, tree, map, name, editRecordsInline); probe reverted, blob verified back. DECLARED NARROWING: the full app-shell suite (281 files) and the repo-wide lint farm are left to CI; ground = the only source edit is comments plus a type-only cast (erased at emit) with type-check green over both tsconfigs, and eslint here is not type-aware (no parserOptions.project in eslint.config.js) so this diff cannot move any untouched file's verdict.",
"mcp_calls": "0 - all reads and writes went through repo-scoped REST (probe: GET /repos/objectstack-ai/objectui = 200) and git; no MCP GitHub call was made, including for dedup, since nothing needed filing.",
"open_questions": [],
"out_of_scope_findings": [
"noted, not filed: NamedListView is 7 keys short of what this host reads (rowColor, appearance, chart, tree, map, name, editRecordsInline), measured as the 22 diagnostics above. Successor: objectui#7483, which owns exactly that surface; filing would duplicate it.",
"noted, not filed: operations never reaches ListView on this path - ListView reads schema.operations.export, plugin-view's delegation literal does not compose the key, and the relay has no rung. NOT a live defect: app-shell composes the object-view node in code and never sets operations, so no author can reach it. Successor: the objectui#5097 host-composition ledger owner. Recorded in the census as kind node-authored, so it must be re-read the day the key becomes reachable.",
"noted, not filed: a view record storing only the legacy fields / filters spelling has no path to ListView (upstream composes the canonical columns / filter only, and the fold runs at the ListView boundary on whatever arrived). Unproven as a live defect - no measurement that producers write the legacy spellings into stored views - so an observation, not a card. Successor: the objectui#2890 fold lineage.",
"noted, not filed: three schema-borne callbacks (onSortChange, onHiddenFieldsChange, onColumnStateChange) are inert duplicates of the identical JSX props; measured - ListView reads only schema.onDensityChange and schema.onNavigate off the schema. Harmless, and now recorded in the census's WRITE_EXCEPTIONS where the next editor of this literal sees it. Successor: this file's next editor."
]
}

---

Generated by Claude Code

[os-zhuang]

os-dev-report

{
  "issue": 7559,
  "status": "done",
  "branch": "claude/issue-7559-viewdef-relay-rung-census",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8864",
  "premise_still_valid": true,
  "summary": "Patch round on PR objectui#8864. Verified the PM's root cause rather than re-deriving it: `Test (shard 2/4)` was red on 5bcc0c899 because objectui#8253's pin asserted the console composition's tree rung with `toContain('tree?.titleField')`, and this PR's `TreeViewConfig` read put a cast BETWEEN the two halves of that substring — `(viewDef.tree as TreeViewConfig | undefined)?.titleField`. The READ was never removed, so the pin's claim stayed true while its text proxy broke. REPAIR: the pin now locates the rung by name (the `labelField:` seam naming `viewDef` and `tree` — the only seam of its kind in that file) and matches the RELATION, `/\\btree\\b(?:\\s+as\\s+(?:[^|\\n]|\\|(?!\\|))*?)?\\s*\\)?\\s*\\?\\.titleField\\b/`: a `tree` expression whose `titleField` is taken with `?.`, tolerating an interposed `as` clause and NOTHING else — `||` may not appear in the tolerated gap, so `titleField` must be read off THIS `tree` and not off another term on the same line. Its docblock claim and its reason (deleting the read would reverse a recorded ruling) are unchanged; only its evidence changed. A new CONTROL case pins both spellings the rung has actually worn as matching and both the read-deleted and read-off-a-different-term shapes as not matching, so a regex loosened until it cannot fail is distinguishable from one that still works. The failure message also improved incidentally: it is now the one offending line instead of a dump of the whole 2600-line file. CODE SHAPE DELIBERATELY UNCHANGED: hoisting the cast into a local would restore the old substring but redden objectui#6557's convergence pin, which requires every seam line to name `viewDef` itself — so no `ObjectView.tsx` byte moved this round. ===== THE STARRED DELIVERABLE, the full set of tests that read `packages/app-shell/src/views/ObjectView.tsx` as SOURCE TEXT (26 files, any package), split by how they reach it. CLASS A — read it BY NAME, so a spelling change can move them (11): [cross-package] packages/types/src/__tests__/tree-view-config-readers-8253.test.ts (THE ONE THIS DIFF MADE FALSE, repaired here); packages/app-shell/src/views/ObjectView.titleFieldConvergence.test.tsx (objectui#6557 — the one this PR's own code comment already knew about); packages/app-shell/src/views/ObjectView.relayRungCensus-7559.test.ts (this PR's own census); ObjectView.calendarBinding-7029.test.tsx; ObjectView.galleryBinding-7547.test.tsx; ObjectView.ganttBinding-7070.test.tsx; ObjectView.overlayPatchOnly.test.ts; ObjectView.unboundAdapterCall.test.ts; ObjectView.viewConfigGate.test.ts; packages/app-shell/src/views/view-filter-fold.ratchet.test.ts; packages/app-shell/src/__tests__/navRunAction.privateConventionRetired.test.ts. CLASS B — sweep it in a recursive source walk, so a pattern change can move them (15): in app-shell, actions-envelope.ratchet.test.ts, adr0054-ratchet.test.ts, flow-envelope.ratchet.test.ts, metadata-client-auth.ratchet.test.ts, no-component-any-cast.ratchet.test.ts, no-refresh-key-remount.ratchet.test.ts, providers/expressionUser.mountSites.ratchet.test.ts, views/viewCacheInvalidation.guard.test.tsx (all eight walk `packages/app-shell/src`); outside app-shell, packages/core/src/utils/__tests__/column-identity.ratchet.test.ts (walks every `packages/*/src`), packages/plugin-detail/src/renderers/__tests__/feedTypeProducerCensus-5877.test.ts (walks repoRoot), packages/types/src/__tests__/handler-keys-string-any-mirrors-7344.test.ts (git grep over packages/apps/examples), packages/i18n/src/__tests__/auth-namespace-3546.test.tsx, packages/i18n/src/__tests__/objectView-config-keys-retired-4730.test.ts, packages/i18n/src/__tests__/report-editor-retired-4145.test.ts (all three declare roots that include `packages/app-shell/src`), examples/console-starter/test/vite-alias-closure.test.ts (walks the app-shell alias target). EACH CHECKED AGAINST THE DIFF, and the check is empirical rather than argued: on head 5bcc0c899 the farm ran all four shards and exactly ONE was red, and its check-run annotations number exactly 2 — the process exit plus `tree-view-config-readers-8253.test.ts:191`. So of these 26 readers exactly one was made false, and it is the one repaired. The near-miss worth naming is CLASS B's `no-component-any-cast.ratchet.test.ts`: this diff does delete two `as any` casts from a file that ratchet walks, but it bans only `as React.FC<any>`-family component casts, so it never saw them.",
  "tests": "All runs in worktree /home/user/objectui-issue-7559 on branch head 9e18f5371, serialized through the container's shared verify lock (`scripts/pm/os-verify-lock.sh`, slot `os-dev-7559`); the wall-clock figures below are shared-box seconds, not idle-box. NOTE ON BUILD: no `dist/` preflight applies to this ablation — the pin reads the repo source file off disk with `readFileSync`, so nothing resolves through a package `exports` map and there is no stale-artifact leg to close. (1) BASELINE — `pnpm exec vitest run packages/types/src/__tests__/tree-view-config-readers-8253.test.ts packages/app-shell/src/views/ObjectView.titleFieldConvergence.test.tsx packages/app-shell/src/views/ObjectView.relayRungCensus-7559.test.ts` -> BASELINE_EXIT=0; verdict lines `Test Files  3 passed (3)` / `Tests  51 passed (51)`. That covers all three at once: the repaired objectui#8253 pin, objectui#6557's convergence pin, and this PR's own census re-run as asked. (2) ACCEPTANCE — `pnpm exec vitest run packages/types/ packages/app-shell/src/views/ObjectView` -> ACCEPT_EXIT=0; verdict lines `Test Files  189 passed (189)` / `Tests  3484 passed (3484)` (162 `packages/types` files + all 27 `packages/app-shell/src/views/ObjectView*` files). (3) GATES — `pnpm --filter @object-ui/types type-check` -> TYPECHECK_EXIT=0 (objectui spells it `type-check`, hyphenated, and the output echoed `@object-ui/types@17.6.0 type-check`, so the filter matched a real script rather than matching nothing and exiting 0; `packages/types/tsconfig.test.json` includes `src/**/*.test.ts`, so this DID read the edited file rather than excluding it). `pnpm --filter @object-ui/types lint` -> LINT_EXIT=0, `273 problems (0 errors, 273 warnings)`, and the edited file appears 0 times in that output. This is the package's own CI gate, not a narrowing, so no narrowing argument is owed; for the record eslint.config.js declares no `parserOptions.project` / `projectService`, i.e. no type-aware linting. (4) THE CRITICAL ABLATION — proving the REPAIRED pin still reddens when the read is genuinely deleted. The repair was committed FIRST (9e18f5371) so the restore had a real commit to return to, and the script carried `trap restore EXIT INT TERM` with absolute paths throughout. Mutation: the tree rung in `packages/app-shell/src/views/ObjectView.tsx` rewritten to `labelField: (viewDef.tree as TreeViewConfig | undefined)?.labelField || 'name',` — i.e. the `titleField` term deleted outright. PROOF IT LANDED ON DISK, taken BEFORE the run: HEAD blob `3044e9efb60ff97cc460690b777380c19b415934`, on-disk blob pre-mutation identical; marker count for the read text `1` before and `0` after, deleted-read shape count `1` after, on-disk blob post-mutation `5f5cc2d6959e031aafb107acf02d7bcc3f1ee15b` — the script hard-exits non-zero if any of those three disagree, so an ablation that was a no-op cannot be read as a result. RESULT: vitest exit 1, and the failure by TEST-CASE NAME: `FAIL |unit| packages/types/src/__tests__/tree-view-config-readers-8253.test.ts > \\`titleField\\` stays DECLARED — the measurement the ruling asked for > the console composition reads it`, with `× the console composition reads it 8ms` and the tally `Tests  1 failed | 18 passed (19)`. Exactly one test failed and it is the right one — the new CONTROL case stayed GREEN alongside it, which is the second half of the proof: the instrument still discriminates rather than having become unable to fail. RESTORE BY STATE, not by exit code: on-disk blob back to `3044e9efb60ff97cc460690b777380c19b415934`, `git diff HEAD` byte count `0`, `git status --porcelain` empty; the restore used `git checkout HEAD -- path` naming HEAD explicitly rather than a bare `git checkout --`, which would have been served from the index. (5) CHANGESET — `node scripts/check-changeset-presence.mjs` -> exit 0; the PR's existing empty-frontmatter `.changeset/objectui-7559-relay-rung-census.md` covers the added file, so no new changeset is owed. (6) CONTROL-BYTE SELF-SCAN — `grep -naP '[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]'` over the edited file: clean. CI on 9e18f5371 is `in_progress` at report time and is the authoritative verdict; on the prior head 5bcc0c899 every other context was green (34 of 34 check runs listed, total_count matched array length), so `Test (shard 2/4)` is the only one this round had to move.",
  "mcp_calls": "1 — this comment. Everything else (check runs, check-run annotations) went through unauthenticated repo-scoped REST, which was probed and worked on this seat.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the two SIBLING assertions in the same describe block are brittle in exactly the way the repaired one was — `expect(read(VIEW_BRANCH)).toContain('viewOptions.tree?.titleField')` and `expect(read(LIST_BRANCH)).toContain('treeCfg.titleField')` will break the same way the moment anyone types those two blocks, which is the natural follow-on to objectui#8253's export. Deliberately NOT changed here: neither is false today, and the standing rule is that a pin your own diff made false is yours to repair while a pin your diff did not touch is not. The repaired one now carries the pattern to copy. Carrier: whoever types the `plugin-view` / `plugin-list` tree branches against `TreeViewConfig` next.",
    "noted, not filed: the general shape behind this failure — a source-text pin whose evidence is a literal substring spanning two syntactic halves, so that any interposed annotation reads as a deleted read — is a repo-wide hazard, not a one-off, and 26 files read this single source file as text. Not filed as a card: no reproducible defect and no violated declared contract today, and the repo has deliberately chosen source-text pins with firing controls as a technique. Carrier: none identified.",
    "noted, not filed: `packages/app-shell/src/views/ObjectView.tsx` is now read as source text by 26 test files across 7 packages, 11 of them by name. That concentration is an observation about coupling, not a defect, and nothing in this round depends on it. Carrier: none."
  ]
}

Verification note for the reviewing seat: the enumeration above is closed empirically, not by argument. On head 5bcc0c899 the check-run listing returned total_count: 34 against an array of 34 (so it is not a truncated page read as a complete one), 33 green and one red; the red one's annotations number exactly 2. One failing test out of 26 source-text readers, and it is the one repaired.

Generated by Claude Code as part of the objectui#7559 dispatch; session reference https://claude.ai/code/session_01611D6ZaRaMmwTNQmSbk8MH.

---

Generated by Claude Code

[os-zhuang]

已落地 —— PR #8864 于 2026-09-09T17:34:03Z 合并（domain:ui 席）

⛔ 按内容验收，不认 sha（合并队列会重写 commit）。在 origin/main（当刻 e6a5cc6d4）上重读：

检	读数
packages/app-shell/src/views/ObjectView.relayRungCensus-7559.test.ts	在，811 行
.changeset/objectui-7559-relay-rung-census.md	在
ObjectView.tsx 旧式 (viewDef as any).tree?.labelField	0
ObjectView.tsx 新式 viewDef.tree as TreeViewConfig	2
亮控制：const fullSchema: ListViewSchema 字面量	1 —— 读的是对的那个文件

⭐ 第三、四行是双向的：⛔ 一个只查「新形态在不在」的探针，在改动根本没落地时也会给出一个可以被误读的读数。这里旧形态归零且新形态出现，探针两个方向都动了。

⚠️ 那根被本 PR 弄红、又修好的钉子，也按内容复核过

packages/types/src/__tests__/tree-view-config-readers-8253.test.ts:225 的 READS_TITLE_FIELD 在已落地的 ObjectView.tsx 上匹配 1 处；同一次读数里 .titleField 的裸计数是 9 —— ⇒ 那条正则是在判别，不是在匹配一切。

⭐ 修法本身是对的那一种：原钉子钉的是源码字面量，被一处插进来的 as TreeViewConfig 打断；修好后它钉的是关系（tree … ?.titleField），并且钉子自己的控制（:247-249）断言它对 #7559 前后两种拼写都匹配、而对一处被删掉的读不匹配。⇒ 证据可以变，主张不可以 —— 被自己的 diff 弄假的钉子，是自己的债。

收下的判断，记两条

1. 卡片说的「47 named keys」在三种读法下都不成立（实测 48 / 36 / 66）。PR 里任何地方都不断言计数，断言的是一条性质，读法写在文件头里。⭐ 交给 dev 的应当是类判据，⛔ 不是数字；引用数字时要点名读法并声明它不是验收条件 —— 本席在派发时给的三个读数正是按这个交出去的。
2. 卡片关于写方向的前提是假的：镜像是 .passthrough()，z.input 因而带索引签名，excess-property 检查从来没覆盖过这个字面量。dev 用一次 parse 证伪它，⛔ 而不是绕过它。

⭐ 消融里最该留下的一条是 M3

M1 / M2 复现了这一类已经产出过的两个缺陷（#7199 description、#7218 rowColor）。M3 是反腐烂方向：给一个已声明的缺席加上 rung、却把声明留着 —— 也必须红。一份答案会悄悄过期的普查是一张清单，不是一次检查。

并且控制被重写过才变得有意义：初稿的控制复述了全局性质，于是别处一红它就红，那是一个什么都没说的控制。⭐ 一个「主体坏了它也红」的控制不是控制。

仍然敞着的，⛔ 没有被本 PR 悄悄合上

#7516 仍开：fieldOrder 被声明为 known-gap 并点名该卡，⛔ 没有为它加 rung。本 PR 交付的是让缺席可见的机制，每一根 rung 各自是一张卡。plugin-view/src/ObjectView.tsx 只被读（其组合字面量被重新推导），⛔ 一个字节都没改。

四条 验收备注 各自点名了后继（#7483 · #5097 台账的归属者 · #2890 fold 世系 · 本文件下一个编辑者），⛔ 都没有骑进这个 PR。

---

Generated by Claude Code