finding(plugin-list,plugin-view): the gantt `progress` / `dependencies` floors still fabricate field names at two faces — flavour 3 of #7070, carded per the 2026-09-01 ruling

[claude]

Split out of #7070 as its flavour 3, which the maintainer ruling of 2026-09-01 (总监批 #28, comment 5494805467) ordered to be carded separately and judged on its own terms. Ruling clause 3, verbatim and untranslated:

flavour 3(progress/dependencies):非轴字段按每行缺值语义单独评估(dev 实测:虚构名产生的是逐行缺值,非整图错轴)—— 车道另立卡对照上述姿态裁量,⛔ 不套用日期轴结论;

The same ruling also records the state transition this card completes: #7070 carries the flavour-2 ①-③ sequence and can close once flavour 3 has its own card. This is that card. Filed unassigned by the seat that landed step ③ (session session_01EMrWaQw3XS5DxTHxp4yRyC); nothing here is dispatched.

Measured, on main at 20cb8db9b — TWO faces, not three

Re-measured by this seat rather than inherited from #7070's table:

Face	Line	Floors
packages/plugin-list/src/ListView.tsx	:2558-2559	progressField ... 'progress', dependenciesField ... 'dependencies'
packages/plugin-view/src/ObjectView.tsx	:1402-1403	progressField ... 'progress', dependenciesField ... 'dependencies'

The app-shell face carries neither key: ganttViewOptions floors only titleField at 'name'. Control for that zero, fired in the same sweep: the same grep over the same file does return titleField: viewDef?.gantt?.titleField || 'name' at :290.

Control for the other direction, also fired: || 'start_date' / || 'end_date' / the timeline || 'created_at' now count 0 across all three files — the date axes are retired, so what remains is exactly this pair.

Why the date-axis conclusion must NOT be applied here

The ruling forbids it, and the measurement behind the ruling says why. A fabricated date axis is a whole-chart error: the renderer lays every bar on a column nobody declared, and getGanttConfig cannot refuse because a name is always present. A fabricated progress / dependencies name is a per-row missing value: rows simply read undefined for that key and render without a progress fill or a dependency link. Different blast radius, different remedy.

Two facts already pinned that this card should build on rather than re-derive:

• getGanttConfig gates on the two date fields alone. Keeping this pair cannot resurrect a config, so ObjectGantt's refusal screen stays reachable with both still handed down — pinned in plugin-gantt/src/ObjectGantt.unconfiguredRefusal-7070.test.tsx.
• The pair's presence is pinned as scope at both faces, deliberately, so that whoever retires them has a place to declare it: ListView.gantt-binding-7070.test.tsx ("invents no date binding when only the OTHER gantt keys are declared") and ObjectView.ganttBinding-7070.test.tsx ("keeps the out-of-scope progress / dependencies floors").

The question this card exists to answer

Is a fabricated non-axis field name acceptable when its only consequence is a per-row missing value — or does 日期轴永不虚构 generalise to "no field name is ever fabricated"?

Worth weighing when it is triaged, not decided here:

• dependenciesField is not the only spelling in play. dependencyField is a declared legacy alias in @object-ui/types with no spec counterpart (finding(plugin-gantt,types): dependencyField is a declared legacy alias in @object-ui/types with no spec counterpart — deprecate in place (removal excluded, deferred to enforce-or-remove) #6470, deprecated in place, removal deferred to enforce-or-remove) — so a decision here should be taken next to that one rather than independently.
• A silent per-row miss is still a silent miss: an author who spelled the key differently gets bars with no progress fill and no diagnostic, which is the same class of no-signal outcome the date-axis ruling objected to, at a smaller radius.
• Against that: unlike a date axis, absent progress and absent dependencies are legitimate, common states. Most gantt rows have neither. A refusal would be wrong here, so the remedy — if any — is "omit the key", not "refuse the chart".

Refs: #7070 (parent card and the ruling) · PR #7110 (flavour 1, gantt date axes) · #7459 / PR #7467 (steps ①②) · #6470 (dependencyField alias) · ADR-0047.

Generated by Claude Code

---

Generated by Claude Code

[os-zhuang]

分诊 — domain:ui / priority:p3 / pm:awaiting-maintainer / finding

锚定 (anchoring)：packages/plugin-list + packages/plugin-view ⇒ domain:ui。

在 origin/main a472b07 上复核 —— 两个面成立，两个对照都发火

packages/plugin-list/src/ListView.tsx:2558     progressField: schema.gantt?.progressField || schema.options?.gantt?.progressField || 'progress',
packages/plugin-list/src/ListView.tsx:2559     dependenciesField: schema.gantt?.dependenciesField || schema.options?.gantt?.dependenciesField || 'dependencies',
packages/plugin-view/src/ObjectView.tsx:1377   progressField: viewOptions.gantt?.progressField || 'progress',
packages/plugin-view/src/ObjectView.tsx:1378   dependenciesField: viewOptions.gantt?.dependenciesField || 'dependencies',

⚠️ plugin-view 侧行号已漂 25（卡片写 :1402-1403）；plugin-list 侧未漂。

对照一（app-shell 面确实只兜 titleField）—— 发火：

packages/app-shell/src/views/ObjectView.tsx:290   titleField: viewDef?.gantt?.titleField || 'name',

对照二（日期轴兜底已清零）—— 我重跑了，结论成立但需要一句说明：全仓对 || 'start_date' / || 'end_date' / || 'created_at' 的搜索在这三个文件面上只有一处命中，而它是一条测试注释：

packages/app-shell/src/views/ObjectView.ganttBinding-7070.test.tsx:30   * restore `startDateField: viewDef.gantt?.startDateField || 'start_date'` and

⇒ 生产代码里为 0，那一处正是钉住"不许恢复"的钉子本身。卡片的对照成立。

为什么是 pm:awaiting-maintainer

这张卡按裁决的要求存在，且它的全部内容是一个待裁的问题：

Is a fabricated non-axis field name acceptable when its only consequence is a per-row missing value — or does 日期轴永不虚构 generalise to "no field name is ever fabricated"?

2026-09-01 的裁决（总监批 #28）明写「车道另立卡对照上述姿态裁量,⛔ 不套用日期轴结论」。⇒ 需要裁决，不是执行。⛔ 分诊席不裁。

⭐ 卡片把双方论据都摆出来了，且都成立 —— 这是它最有价值的部分

支持"也该退休"：

A silent per-row miss is still a silent miss: an author who spelled the key differently gets bars with no progress fill and no diagnostic, which is the same class of no-signal outcome the date-axis ruling objected to, at a smaller radius.

支持"保留"：

unlike a date axis, absent progress and absent dependencies are legitimate, common states. Most gantt rows have neither. A refusal would be wrong here, so the remedy — if any — is "omit the key", not "refuse the chart".

⭐ 第二条里那句"补救是省略这个键而不是拒绝整张图"是关键的技术约束：它意味着即使裁决为"退休"，修法也与日期轴那次不同。裁决者请注意这一点，⛔ 不要裁成"照 flavour 1 办"。

两条已钉住的事实，⛔ 不要重新推导

卡片列的，本席原样转达：

1. getGanttConfig 只按两个日期字段把关 ⇒ 保留这一对不会让一个本该被拒的配置复活；ObjectGantt 的拒绝屏在两者都传下去的情况下仍然可达（plugin-gantt/src/ObjectGantt.unconfiguredRefusal-7070.test.tsx 钉住）。
2. 这一对的存在是被"作为范围"刻意钉住的，两个面各有钉子（ListView.gantt-binding-7070.test.tsx、ObjectView.ganttBinding-7070.test.tsx）—— 所以退休时会红，那是设计好的交接点，不是回归。

⚠️ 裁决时必须一并看的一件事

dependencyField is a declared legacy alias in @object-ui/types with no spec counterpart (#6470, deprecated in place, removal deferred to enforce-or-remove) — so a decision here should be taken next to that one rather than independently.

⭐ 也就是说：dependenciesField 的兜底与 dependencyField 这个别名是同一个键的两个遗留问题。分开裁会产生一个尴尬状态（兜底退休了、别名还在，或反之）。⛔ 建议裁决者把 #6470 一起看。

定级理由

priority:p3：

• 爆炸半径小：逐行缺值，不是整图错轴。缺进度条、缺依赖线，图仍然可读。
• 无数据面、无安全面、无静默错数。
• 不降更低：它是一个已裁流程要求答复的问题，挂着不答等于让 finding(views): the gantt and timeline branches still fabricate date-axis field names on all three faces — and app-shell's own #3129 note certifies them as already fixed #7070 无法关闭（那张卡的收尾条件就是"flavour 3 有了自己的卡"，现在有了，接下来欠一个裁决）。

⛔ 分诊席不认领、不派单、不写码、不裁决。

---

Generated by Claude Code

[huangyiirene]

状态转换：pm:awaiting-maintainer → pm:queue（2026-09-09）

维护者回批逐字：「C 桶那 24 张的批量转 pm:queue 同意」。

判据：⭐ 本卡命中 SKILL.md:720 元判据①：「静默丢弃的声明并入既有拒收集；新支复用母单裁决直接入队，真实语义差异才重开。」gantt 的 progress / dependencies 兜底在两个面捏造字段名，卡面自陈是 #7070 的 flavour 3，且已按 2026-09-01 的裁决立卡 ⇒ 母单裁决在案、本支无真实语义差异。⛔ 本就不该进任何箱，直接入队。

finding、domain:ui、priority:p3 留下。

⚠️ 根因见 objectstack #17017。

---

Generated by Claude Code

[os-zhuang]

Claim: session session_01611D6ZaRaMmwTNQmSbk8MH · branch claude/issue-7499-gantt-nonaxis-floors

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.

⚠️ 先澄清一件事：2026-09-01 的裁决没有裁这张卡的实体

它裁的是由谁裁、以及不许怎么裁。原文逐字：

flavour 3(progress/dependencies)：非轴字段按每行缺值语义单独评估…车道另立卡对照上述姿态裁量，⛔ 不套用日期轴结论

⇒ 「另立卡对照上述姿态裁量」= 交给本车道裁；「⛔ 不套用日期轴结论」= 禁止把 flavour 1 的答案搬过来。所以下面这条是本席的裁决，⛔ 不是维护者的，⛔ 也不能被当成维护者的引用。

裁决 —— 不声明就省略该键，⛔ 不要虚构一个名字

理由是两种缺陷的不对称，而这个不对称正是裁决禁止套用日期轴结论的原因：

	虚构的日期轴	虚构的 progress/dependencies
后果	整图错轴 —— 每根条形落在没人声明的列上	逐行 undefined —— 该行没有进度填充 / 没有依赖连线
有没有合法孪生?	⛔ 没有。一张图不可能「本来就该落在一个不存在的列上」	⭐ 有，而且是常态。 多数甘特行本来就既无进度也无依赖

⇒ 拒绝（refuse）是错的 —— 它会打断那个合法且常见的状态。卡片自己也是这么说的，我同意。
⇒ 但虚构一个名字同样是错的 —— 它凭空制造了一条作者从未写过的绑定。

⇒ 补救是「省略」，⛔ 不是「拒绝」：progressField / dependenciesField 未声明时不下传该键，而不是下传 'progress' / 'dependencies'。合法的缺席照旧渲染（与今天一模一样），而一个拼错了键的作者不再意外命中一个恰好同名的列。

前提（可证伪，且是你的第一项测量）

真正会因此改变行为的，只有这一类：schema 里有一个字面名为 progress / dependencies 的列，同时又没有声明 progressField / dependenciesField。 这类今天靠 floor 意外工作着，省略之后会停止。

⇒ 先测这个类在语料里的规模（examples/、content/docs/**、packages/*/src 的 fixtures，以及任何 JSON/YAML 元数据）。

• 若为空（或只有你能逐条判定为非真实作者的站点）⇒ 裁决成立，照做。
• 若非空 ⇒ ⛔ 本裁决作废，停下报回。那时「省略」就是一次静默的破坏性变更，答案可能变成「保留 floor 并把它写进文档」或「省略 + 一条迁移说明」，⛔ 那不是你或我能单独定的。

⚠️ 要有盲区读数——读不出的站点计数。没有盲区读数的 0 不是测出来的 0。

我替你核过的坐标（⚠️ 卡片的行号已漂）

plugin-list/src/ListView.tsx:2712-2713      progressField … || 'progress' / dependenciesField … || 'dependencies'   （卡片写 :2558-2559）
plugin-view/src/ObjectView.tsx:1507-1508    同上                                                                      （卡片写 :1402-1403）
两处上方各有一条 ⛔ 注释说明这对 floor 是被"刻意留下"的（:2696 / :1494）

⭐ 而 app-shell 面确实没有 floor —— 这一条我差点判错，过程值得你知道：

我的探针 grep -c progressField 在 app-shell/src/views/ObjectView.tsx 上返回 1，与卡片「app-shell 面两个键都没有」冲突。查下去，那 1 是 :2246，一个 dev-mode 诊断用的键名字符串数组（flatKeys）里的字面量，⛔ 不是 floor；同一文件 || 'progress' 计数为 0。

⇒ 卡片是对的，我的计数才是需要解释的那个。 ⛔ 不要碰 :2244-2248 那个数组 —— 它是诊断用的键名清单，改它等于改一个错误的东西。一个计数在你知道每个命中「是什么」之前不是事实。

两条对照我都点亮了：app-shell 确实 floor 了 titleField（:290），所以那个 0 不是死探针；|| 'start_date' / || 'end_date' / || 'created_at' 在三个文件上合计 0，日期轴确实已退役 —— 卡片这条也成立。

⭐ 有两条钉子是这张卡该翻的，⛔ 不是意外损伤

卡片写明：这对 floor 的存在被当作 scope 刻意钉住，「so that whoever retires them has a place to declare it」——

plugin-list  ListView.gantt-binding-7070.test.tsx    「invents no date binding when only the OTHER gantt keys are declared」
plugin-view  ObjectView.ganttBinding-7070.test.tsx   「keeps the out-of-scope `progress` / `dependencies` floors」

⇒ 这两条正是为你准备的申报位。 翻它们是本卡的交付物，⛔ 不是「顺手改了别人的钉子」。翻的时候把理由写进用例名或 docblock：为什么省略而不是拒绝。

⛔ 不要动 plugin-gantt/src/ObjectGantt.unconfiguredRefusal-7070.test.tsx —— getGanttConfig 只 gate 那两个日期字段，所以拿掉这对 floor 不可能复活一个 config，那条拒绝屏仍然可达。它是这次改动的非回归对照。

边界

• ⛔ 不要唤醒 finding(plugin-gantt,types): dependencyField is a declared legacy alias in @object-ui/types with no spec counterpart — deprecate in place (removal excluded, deferred to enforce-or-remove) #6470（dependencyField，已声明的遗留别名，退役延后到 enforce-or-remove）。⚠️ 但卡片说得对：这个决定应当挨着它做 ⇒ 去读 finding(plugin-gantt,types): dependencyField is a declared legacy alias in @object-ui/types with no spec counterpart — deprecate in place (removal excluded, deferred to enforce-or-remove) #6470，并在 验收备注 里写明你的省略对那个别名意味着什么（若有）。⛔ 不要在本 PR 里动它。
• ⛔ 不要碰日期轴（已退役，实测 0）。
• ⛔ 不要给别的 gantt 键加省略；本卡的面是这两个键、这两处。

Clause-②

预期 no（移除一个虚构的默认值是收窄：无新导出符号、无已发布载荷上的新键）。但你自己测，并把探针连同它的对照一起写出来 —— 一个「无输出」的探针要有一条证明它会说话的对照。⚠️ 若结论是 yes，本席的契约评审当前不可达（连续 15 次 429）会导致停队 ⇒ 先报回来。

验收 —— 类判据

「一个没有声明 progressField / dependenciesField 的视图，下传的配置里就没有这两个键；而一个声明了的，逐字下传作者写的值。」

• 消融：从已提交树出发，把省略改回 floor，盘上先证（marker / blob），跑，看它按用例名变红，再按状态还原。⛔ 只有退出码不算。
• ⭐ 枚举读这两个文件的测试，⛔ 不要只跑你改的那个包的门。本班已经三次栽在「一条测试读了你改的那个文件」上（docs(plugin-dashboard): teach the adapter as renderer context, not a schema key (objectui#8468) #8857 被队列踢出、test(app-shell): stand up the object-view relay rung census (objectui#7559) #8864 的文本钉、finding(types,components,docs): page.actions has no reader on the page node — the array renders nothing and leaks onto the wrapper element as actions="[object Object]", while guide/layout.md teaches it in three passages #7926 的 spec-parity 台账）。

⛔ 绝不编辑共享主检出，用专属 worktree。⛔ 绝不 git stash。⛔ 绝不按名字杀进程。

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 7499,
  "status": "done",
  "branch": "claude/issue-7499-gantt-nonaxis-floors",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8876",
  "premise_still_valid": true,
  "summary": "Implemented the PM ruling: OMIT progressField/dependenciesField when undeclared, rather than flooring them at 'progress'/'dependencies'. Both floored faces (plugin-list/src/ListView.tsx, plugin-view/src/ObjectView.tsx) are now pure deletions -- the trailing ...(schema.gantt || {}) / ...(viewOptions.gantt || {}) spreads always had the last word over those two lines, so a declared value still reaches the renderer verbatim through both the canonical `gantt` block and the legacy `options.gantt` nesting, while an undeclared one arrives as an ABSENT key. The two pins #7070 left as a declaration slot are flipped with the omit-not-refuse reasoning in the case names and docblocks; ObjectGantt.unconfiguredRefusal-7070 was NOT touched and stays green in both ablation worlds. VERIFIED THE PM's COORDINATES INDEPENDENTLY: app-shell has no floor -- grep -c progressField returns 1 and that hit is the dev-mode flatKeys diagnostic array (untouched); || 'progress' counts 0 there; control that the zero is not a dead probe is app-shell's titleField || 'name' gantt floor; date-axis floors count 0 across all three files. Assignee was already set by the PM (os-zhuang) and the Claim: comment names this branch -- neither written nor touched, and no second claim posted. Diff is NOT a governed surface (check-governed-queue-guard --test, 5 paths, none matched), so no maintainer-readout section. A changeset is owed and written (minor / plugin-list + plugin-view).",
  "census": "THE FALSIFIABLE PREMISE, MEASURED FIRST -- population EMPTY, ruling stands. Predicate: a gantt view routed through one of the two floored faces, on data carrying a column literally named progress/dependencies, without the corresponding key declared. Routed gantt-view declarations tree-wide: 8 files; intersected with a literal progress/dependencies column: 2 files, and both are the pin tests this card is chartered to flip (they carry the literal string only as the assertion of the floor itself) -- individually judged not real authored views. Excluded on stated grounds: the two examples/schema-catalog plugin-gantt JSONs DO carry a literal `dependencies` column but are out of population twice over -- they declare dependenciesField explicitly AND are direct type:'object-gantt' schemas taking getGanttConfig's FIRST branch, where these floors never applied; the docs/README/skills gantt examples are direct renderer usage ('gantt' is registered straight to ObjectGanttRenderer) and every one carrying the literal column declares the key beside it; the three field-definition sites literally naming a `progress` column (plugin-detail percentClip, plugin-form deriveMasterDetail, plugin-grid bulk-actions demo) mention gantt ZERO times; 528 JSON/YAML metadata files swept separately, only intersection is those same two example JSONs. POSITIVE CONTROL that the census can speak: relaxing exactly one conjunct (dropping 'does not declare the key') takes the same pipeline from 0 sites to 51, and the sharp predicate itself returns 2 rather than 0. BLIND-SPOT READING (a zero without one is not a measured zero): 75 non-text files in the corpora unreadable by grep (74 .png doc screenshots + packages/plugin-grid/demo/sample.xlsx, a spreadsheet-import demo fixture -- judged individually by path, none a view-metadata carrier); 0 untracked non-ignored files; 0 gitignored non-build paths; and one UNBOUNDED blind spot stated as such -- runtime/backend-supplied object and listview metadata, which is what these two faces actually read in production and is not in this repo at all. The census covers the corpora the card named, not a customer's schema.",
  "tests": "All runs went through the shared verify lock (OS_VERIFY_LOCK_SLOT=issue-7499-dev); verdicts read from the printed VERDICT command-exit lines, never a bare $?. Everything below is the tree that is commit 544e95db2. (1) THE TWO FLIPPED PINS + the non-regression control ObjectGantt.unconfiguredRefusal-7070: 'Test Files 3 passed (3) / Tests 25 passed (25)', VERDICT command-exit 0. (2) BOTH AFFECTED PACKAGES IN FULL (plugin-list + plugin-view): 'Test Files 109 passed (109) / Tests 1206 passed (1206)', VERDICT command-exit 0. (3) ENUMERATED EVERY TEST THAT READS THE TWO EDITED FILES rather than only the packages touched -- source-text readers and ledgers found by grepping for filesystem reads of those paths: plugin-view/objectViewHostSurface, core/column-identity.ratchet, types/tree-view-config-readers-8253, types/object-kanban-group-by-limit-7322, types/object-view-unmirrored-keys-7779, types/object-kanban-record-source-7780, types/schema-reference-named-list-view-keys-7923, types/zod-mirror-parity, types/gantt-flat-config-declared-keys, types/gantt-dependency-field-deprecated-alias, plugin-grid/gridNonAuthorKeys, plugin-list/forwardref-props-annotation.guard, app-shell/defaults-maps-mirror-en-pack, app-shell/print-stylesheet-4462, app-shell/ObjectView.ganttBinding-7070, plus all of packages/plugin-gantt: 'Test Files 77 passed (77) / Tests 819 passed (819)', VERDICT command-exit 0. POST-COMMIT re-run of the ratchet family and the source-text pins at 544e95db2: 'Test Files 6 passed (6) / Tests 162 passed (162)', VERDICT command-exit 0. (4) TYPE-CHECK plugin-list + plugin-view (both tsc legs incl. tsconfig.test.json): 'packages/plugin-view type-check: Done / packages/plugin-list type-check: Done', VERDICT command-exit 0. Its FIRST run was red with TS2307 Cannot find module '@object-ui/components' -- that reading was NOT MEASURED, not a failure: the dependency closure was unbuilt. Green only after turbo run build --filter '@object-ui/plugin-list^...' --filter '@object-ui/plugin-view^...' (14 tasks successful). (5) LINT measured over the WHOLE population, not narrowed: eslint . --no-inline-config --format json read 4668 files (count taken from eslint's own JSON output); all four changed files report errorCount 0; CI-shaped turbo run lint for both packages exits 0 ('494 problems (0 errors, 494 warnings)'). The 94 errors that run reports elsewhere are pre-existing, in files this diff does not touch, and are an artefact of --no-inline-config stripping inline disables. (6) GATES, derived by hand from this repo's package.json and .github/workflows (objectui has no scripts/pm/dispatch-gates.mjs): GREEN -- check:comment-mask-corpus, check:control-bytes, check:doc-examples, check:doc-snippets, check:doc-types, check:dist-completeness, check:readme-exports, check:handler-key-reads, check:icon-record-names, check:docs-route-closure, check:spec-symbols, check:unreferenced-sources, check:element-data-source-declaration, check:doc-example-readers, check:vi-mock-specifiers, check:vi-mock-inherit, check:shell-escape-residue, check:lint-rule-coverage, check:i18n-keys, check:changeset-presence (exit 0, '4 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s)'), check:governed-queue-guard --test (NOT GOVERNED, 5 paths). check:doc-examples matters most -- it carries literal line anchors into BOTH edited files (ListView.tsx:808, ObjectView.tsx:618/632/649, all above the deletions) and printed its own verdict line: 'Every covered @example compiles, or fails exactly as its ledger row declares.' Plus a control-character self-scan over the five changed files: no matches. DECLARED NOT MEASURED, left to CI: check:eager-closure and check:sdui-registration-pins. Both exit 2 with their own PREREQUISITE-NOT-MET wording ('This is exit 2, not a pass', 'a broken gauge, not a passing budget') because they weigh an apps/console bundle that was not built. Reading either as green would read a non-measurement as a pass. ABLATION -- from the COMMITTED tree, floors put back, direction predicted before running. ON-DISK PROOF TAKEN BEFORE THE RUN: marker counts moved 1 -> 2 for both || 'progress' and || 'dependencies' in each file, and both blobs diverged from HEAD (ListView on-disk aaa5803de vs HEAD aae0b1da2; ObjectView on-disk f0ca7d425 vs HEAD c1a807af6). RED BY TEST-CASE NAME, 4 failed / 21 passed: 'OMITS `dependenciesField` rather than inventing it, when only the OTHER gantt keys are declared (objectui#7499)'; 'OMITS both non-axis keys for a gantt view that declares no config (objectui#7499)'; 'OMITS `progress` / `dependencies` rather than inventing them -- the retirement #7070 left a slot for (objectui#7499)'; and 'CONTROL: declaring ONE of the pair leaves the other absent, not fabricated (objectui#7499)' -- reporting that fourth one honestly: it is NAMED a control but half of it is an omission assertion, so it correctly goes red like a measurement rather than staying green like a control. GREEN IN BOTH WORLDS (the real controls): 'CONTROL: a DECLARED `dependenciesField` still passes verbatim', 'CONTROL: the legacy `options.gantt` nesting carries the pair verbatim too', 'CONTROL: a DECLARED `progressField` / `dependenciesField` still passes verbatim', every #7070 date-axis case, and all 5 cases of ObjectGantt.unconfiguredRefusal-7070 including 'refuses even when the out-of-scope progress/dependencies floors are still handed in'. RESTORE PROVEN BY STATE, not by exit code: git diff HEAD empty, git status --porcelain empty, both on-disk blobs back to their HEAD hashes; the mutation script restored through trap ... EXIT INT TERM with absolute paths and git checkout HEAD -- (never a bare git checkout --, which reads the index). Re-ran the three files from the restored tree: 25/25 green.",
  "clause_2": "no -- measured, and the required control is what makes it a measurement. The bare CLI run 'check-widening-tells.mjs --declaration no --diff THIS-PR.diff' exits 0 with 'no widening tell on any declared surface', and --self-test passes 150 cases. BUT THE CONTROL DID NOT SPEAK: appending one added key on packages/types/src/zod/objectql.zod.ts -- a file the tool's own declared objectui mirror row covers -- to a copy of this very diff and re-running the identical command STILL exits 0. The CLI has no --repo flag (grep -c '--repo' = 0, against a control grep -c '--declaration' = 6 proving the grep can speak) and main() calls wideningRefusal({declaration, files}) without one, so it always judges as objectstack, where that row is inert. THE READING THAT DOES SPEAK, via the module's exported judge: CONTROL diff with repo=objectstack-ai/objectui -> state=refused (T1, packages/types/src/zod/objectql.zod.ts); THIS PR's diff with repo=objectstack-ai/objectui -> state=clean; CONTROL diff with repo defaulted (what the CLI does) -> state=clean, the dead-probe leg. So: control fires, this diff is clean, Clause-2 = no. No needs:contract-review label applied and nothing parks -- so the 15-consecutive-429 unreachability the dispatch warned about is not engaged. The CLI gap is filed (see out_of_scope_findings). The ENQUEUE gate is NOT affected: check-clause2-carriers.mjs threads the repo correctly via PM_SWEEP_REPO / GITHUB_REPOSITORY into pairWidening(pair, repo); the trap is only the hand-run CLI.",
  "mcp_calls": "2 -- one targeted search_issues (declared channel switch: REST /search returns 403 'sessions are bound to their configured repositories', so per the channel table the fallback is one targeted MCP search; it returned 13 rows, so the non-empty result is its own control) and one issue_write to file the finding. Everything else -- issue body, dispatch comment, the #7070 ruling, #6470, #17112, PR creation, PR read-back, the report comment -- went through repo-scoped REST, which probed HTTP 200 at this seat.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as objectstack#17217 (sub-issue of the open objectstack#17112, which edits the same output line): check-widening-tells.mjs's CLI cannot be told the repo, so its declared objectui mirror surface row (packages/types/src/zod/** @ objectui) is unreachable and EVERY objectui diff reads clean -- including one that does add a key to that surface. Reproduction plus a three-legged control in the card. Filed as a sub-issue rather than standalone because #17112's proposed remedy, applied alone, would print these files as 'NOT MEASURED -- no declared surface covers them', which is false: a row covers them and the run merely could not say so.",
    "noted, not filed: the flipped case 'CONTROL: declaring ONE of the pair leaves the other absent, not fabricated' is named as a control but asserts an omission, so it fails under the ablation like a measurement. Cosmetic naming only -- the case is correct and the file already carries three true two-world controls. Taker: whoever next edits ObjectView.ganttBinding-7070.test.tsx.",
    "noted, not filed: both edited files still compute startDateField / endDateField into a conditional spread that the trailing ...(schema.gantt || {}) would deliver anyway -- the same redundancy this PR removed for the non-axis pair, left in place because the date axes are out of scope for this card. Not a defect; the explicit form is arguably clearer. Taker: whoever next revisits the #7070 date-axis shape.",
    "noted, not filed (recorded in the PR's 验收备注 as the dispatch required, and NOT acted on): what this omission means for #6470's dependencyField. getGanttConfig's flat branch reads schema.dependenciesField || schema.dependencyField. While the floor stood, dependenciesField was ALWAYS truthy through these two faces, so the declared legacy singular alias was unreachable via them -- shadowed, never read. With the floor gone the alias limb becomes live, which is what its @deprecated contract already promises and what ObjectGantt.dependencyAlias already pins at the renderer. Measured: ZERO sites at either routed face declare dependencyField, so no in-repo view moves. #6470 was read but ⛔ not woken and nothing in it was touched. Taker: #6470 itself, whose retirement should know this PR is what made the alias reachable through the view layer."
  ]
}

---

Generated by Claude Code

[os-zhuang]

已落地 —— PR #8876 于 2026-09-09T18:53Z 合并（domain:ui 席）

⛔ 按内容验收，不认 sha。origin/main 当刻 4c2e1d7f0。

⚠️ 第一次探针给了假读数 —— 记在前面，因为它是个新形状

我先跑的是「|| 'progress' 现在还剩几处」，两个文件都返回 1。差一点就据此说落地失败。逐条读出来是什么：

ListView.tsx:2706    //   - FABRICATING was also wrong. `|| 'progress'` / `|| 'dependencies'`
ObjectView.tsx:1499  // the common case — but `|| 'progress'` / `|| 'dependencies'` invented
ObjectView.tsx:1318  'locationField', 'imageField', 'dependenciesField', 'progressField',  ← dev-mode flatKeys 诊断数组里的字符串

⭐⭐ 全是注释在引用被删掉的那段代码，外加一个诊断数组里的字符串。 修复用自己的注释解释「为什么这里不再兜底」，而那段解释里必须写出被删的字面量 —— 于是一个「查这个字符串还在不在」的探针，被「说明这个字符串已经不在了」的散文骗过去。

⇒ 修得越自觉，越容易骗过这种探针。 这与本班已经付过学费的那条互为镜像：「一个找修复痕迹的探针看不见未修复的」；这次是「一个找字符串缺席的探针，被点名该字符串的散文骗到」。

正确的读数（剥掉注释行，两个控制都亮）

检	ListView.tsx	ObjectView.tsx
可执行的 || 'progress'	0	0
可执行的 || 'dependencies'	0	0
⭐ 控制：必须存活的邻居兜底（|| 'name' / titleField）	15	11
⭐ 控制：剥离器没把文件清空	2590 / 4376 行	1025 / 2201 行
承重的另一半：canonical spread ...(schema|viewOptions).gantt || {}	1	1
控制：#7070 的日期轴兜底未动	startDateField 18 / endDateField 11	10 / 6

⭐ 第二个控制是必要的：一个把整个文件剥空的剥离器，会让「可执行命中为 0」变成一句真话而毫无意义。

⇒ 兜底已删，而一个已声明的值仍原样透传（spread 有最后发言权），日期轴那一族没被顺手动到。

⭐ 这一轮最值得留下的判断

普查先做，而且结论是空集。 谓词是「一个经过两个被兜底的面之一路由的 gantt 视图，数据里有字面叫 progress/dependencies 的列，且未声明对应键」—— population 空。而那个零带正控制：放宽恰好一个合取项，同一条流水线从 0 变 51；用锐化后的谓词本身则返回 2，且两处都是本卡要翻的钉子测试。⇒ 那个零是测出来的，不是探针没跑。

盲区被声明，⛔ 不是算成零。 75 个 grep 读不了的非文本文件逐个按路径判定；而且 ⭐ 一个无界盲区被明说：运行时/后端下发的对象与列表视图元数据 —— 那正是这两个面在生产里真正读的东西，而它根本不在本仓。原话值得留：「这次普查覆盖的是卡片点名的语料，不是某个客户的 schema。」

dev 对自己一处命名的诚实更正：CONTROL: declaring ONE of the pair leaves the other absent, not fabricated 名字叫 control，但它断言的是一次省略，所以在消融里像测量一样变红。dev 主动报告了这一点而不是掩过去 —— 命名瑕疵，用例本身正确，文件里另有三个真正的两世界控制。

⚠️ 一个下游后果，已记入 验收备注，⛔ 未动 #6470

getGanttConfig 的扁平分支读 schema.dependenciesField || schema.dependencyField。兜底还在时 dependenciesField 恒为真，于是那条已声明的 legacy 单数别名经由这两个面永远不可达 —— 被遮蔽，从未被读。兜底删掉后该别名变为活的，这正是它 @deprecated 契约已经承诺的、也是 ObjectGantt.dependencyAlias 已经钉住的。实测：两个被路由的面上零处声明 dependencyField，⇒ 仓内没有视图因此改变行为。⛔ #6470 只读未动。后继：#6470 自己 —— 它的退役需要知道正是本 PR 让该别名经由视图层变得可达。

---

Generated by Claude Code