The whole list.tabs[] entry is inert, not just label — 36 of 48 authored icons name something the console cannot render, and three test comments teach the wrong model #1307

Activity

1. hotlong

   added

   metadataDeclarative metadata — schema, security posture, UI surfacesDeclarative metadata — schema, security posture, UI surfaces

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

   and removed

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   on Aug 26, 2026
2. hotlong

   assigned

   huangyiirene

   on Aug 26, 2026
3. hotlong commented on Aug 26, 2026

   

   hotlong

   on Aug 26, 2026 – with Claude

   ContributorAuthor

   More actions

   派发认领（R5）— pm:queue → pm:dispatched

   • Session: dea4cde2-db24-5125-950e-e38371290748
   • Branch: claude/issue-1307-list-tabs-inert

   ⚠️ 卡片的「三个 test/ 文件」已过时 —— 实测是六个

   那份清单是 #1283 那轮的勘察结果，PR #1304 落地后又多出一个。开工前对 origin/main @ cd198f10 重扫 list.tabs（⛔ 不是裸 tabs —— userFilters.tabs[] 是另一把活键）：

   文件	行	读法
   test/view-references.test.ts	420-455	v.list?.tabs → t.view（承载三条假信念里最重的一条）
   test/account-renewal-model.test.ts	79	AccountViews.list?.tabs → t.view
   test/decorative-field-sweep.test.ts	114	ProductViews.list?.tabs → t.view
   test/case-assignment.test.ts	371-374	caseViews?.list?.tabs → 找 unassigned_triage，并断言 tab.pinned === true ⚠️ 卡片未列
   test/docs-search-navigation-views.test.ts	58, 179	list.tabs → t.label(58) 与 t.view(179) ⚠️ 卡片未列
   test/view-tab-label-inert.test.ts	全文	#1304 立的钉子 ⚠️ 卡片未列 —— 立卡时它还不存在

   本席上一轮的教训：围栏按枚举文件名下，不按类别下。卡片这句「三个 test 文件」正是类别式表述失准的实例，本条更正即是它的兑现。

   ⚠️ 卡片列了三条假信念，实测是五条 —— 多出来的两条都在测试里

   卡片已列：view-references.test.ts:420-427、crm.app.ts:19、crm.app.ts:86。再加：

   ④ test/case-assignment.test.ts:371-374 —— 同一个 #760 假信念，只是换了个 object。

   const tab = (caseViews?.list?.tabs ?? []).find((t: AnyRec) => t.view === 'unassigned_triage');
   expect(tab, 'no tab reaches the triage view — an unreachable view is not visibility').toBeTruthy();
   expect(tab.pinned, 'the triage tab is not pinned').toBe(true);

   两条断言都建立在「view 不被 tab 列出就够不到」这个已被证伪的模型上；pinned 与 icon 同样惰性。但被断言的东西本身是真的：unassigned_triage 是 case.view.ts:196 的一个 listViews 条目，切换器逐条枚举 listViews，所以它本来就可达，与 tabs 无关。

   ⇒ 改法：可达性由 triage 在 listViews 里存在直接满足（上面第 369-370 行已经断言了），删掉 tab 查找与 pinned 断言即可。⛔ 不要顺手动 unassigned_triage 视图本身的 filter —— 那是 #1145 的标的，本轮不在你手上。

   ⑤ test/docs-search-navigation-views.test.ts:58 —— #1304 之后已是死代码。

   const tabs = ((list.tabs ?? []) as AnyRec[]).map((t) => t.label);
   return [list.label, ...extra.map((v) => v.label), ...tabs].filter(Boolean) as string[];

   t.label 自 PR #1304 起恒为 undefined，.filter(Boolean) 全数丢弃 ⇒ 这一路今天贡献零个字符串。这是 #1304 落地时无人注意到的衰变，本卡顺手收掉：删掉这一路，或改成读目标 view 的 label（那才是屏幕上真正显示的串）。⚠️ 若改成后者，先确认期望集变大不会让该测试变红。

   ⚠️ view-tab-label-inert.test.ts 的反空洞守卫会因本卡变红 —— 设计如此，不是意外

   它自己写着（68-74 行）：

   The guard above is only meaningful while there are tabs to guard. If a refactor empties every tabs[] array, the assertion passes vacuously

   本卡就是那个 refactor。 删光 tabs 之后「不得重新引入 label」变成空洞真，而 is still measuring a non-empty set of tabs 会失败。

   ⇒ 正确做法是改写这个钉子换靶：从「list.tabs[] 里没有 label」升级成「list.tabs[] 整体不存在」，反空洞面随之改成「确实扫到了 12 个 view 文件的 list 块」。⛔ 不得删除该文件、不得 skip、不得降级成注释 —— 换靶 ≠ 撤钉。

   前提复核（origin/main @ cd198f10，本席实测）

   • tabs: [ 出现在 12 个 view 文件（lead.view.ts / event_attendee.view.ts 无）。
   • 块内条目 60、icon: 48 —— 与卡片数字逐字相符。
   • listViews 条目内无 tabs（三处 grep 命中全是散文/注释）⇒ 面就是这 12 个 list.tabs。
   • label: 在 src/views 全仓 160（ADR-0049 enforce-or-remove: all 60 list.tabs[].label strings across the 12 view files are dead configuration — the console renders each view's own label #1283 前 220，PR refactor(views): remove the inert label key from all 60 list.tabs[] entries (ADR-0049) #1304 削掉 60）—— 与上一轮收尾读数一致。
   • crm.app.ts 第 19 / 86 行两句假信念都还在，原文照旧。

   「先重测渲染器」这条验收 —— 本席已代做半边

   • @objectstack/console 不在 package.json，是传递依赖；pnpm-lock.yaml 解析为 @objectstack/console@17.1.0（720 / 3697 / 3855 行）⇒ pin 未移动，卡片的 viewTypeIcons 读数取自 17.1.0，锁文件今天仍是 17.1.0。
   • 剩下半边归你：pnpm install 后到 node_modules/@objectstack/console 里亲眼确认 viewTypeIcons 这张硬编码表，以及「切换器从 listViews + list 造条、不读 tabs」。
   • ⚠️ 共享 checkout 的 node_modules 是空的 —— 本席在那里读到 viewTypeIcons 零命中，但控制词 listViews 同为零，⇒ 是树没装，不是符号不存在。零命中在控制词证明搜到了代码之前不构成结论。你在自己的 worktree 里装完再读。
   • ⛔ 若实测与卡片不符：停手回报，不要按卡片的模型改。这张卡整个前提是一次测量，本车道已被「把测量当耐久事实」咬过四次。

   围栏（枚举，非类别）

   ✅ 你的面，仅此 19 个文件：
   src/views/{account,campaign,case,contact,contract,event,forecast,knowledge_article,opportunity,product,quote,task}.view.ts、src/apps/crm.app.ts、test/{view-references,account-renewal-model,decorative-field-sweep,case-assignment,docs-search-navigation-views,view-tab-label-inert}.test.ts。

   ⛔ 不得动：

   • src/views/lead.view.ts、src/views/event_attendee.view.ts —— 无 tabs，本卡无事可做
   • test/forecast-seeds.test.ts —— forecast-seeds.test.ts says "these eight authored records" where the seed holds seven — in the one file whose job is to state the seed's invariants exactly #1282 本轮并行
   • src/objects/event_attendee.object.ts、test/event-attendee-cascade.test.ts —— The hand-written "15 lookups on sys_user" figure in two event_attendee comments has no guard, and had already drifted to 16 #1188 本轮并行
   • scripts/check-source-hygiene.mjs、playwright.config.ts —— The three root .ts files are outside the TODO/FIXME and copyright-header checks, and playwright.config.ts has no header #1236 本轮并行
   • unassigned_triage 视图的 filter —— A resolved unowned case sits in Unassigned — triage forever: the tab filters is_closed, which only flips on closed #1145 的标的
   • content/docs/releases/** —— 任何代码 PR 恒禁
   • ⛔ 不得删 ViewTabSchema.label 或 schema 字段 —— userFilters.tabs[]（ADR-0047）复用同一 schema，那里 label 是活的，删错就删掉一个真实面
   • ⛔ 不得重开 enforce vs remove 之争 —— ADR-0049 enforce-or-remove: all 60 list.tabs[].label strings across the 12 view files are dead configuration — the console renders each view's own label #1283 已裁 remove，轴③对 icon 逐字适用

   交付

   worktree → 分支 → 改 → pnpm test 全绿 → changeset（本仓 changeset-check.yml 门住每个 PR，判据是本 PR 新增的 .changeset/*.md；本卡有作者可见面的变化，写一份真的）→ push → draft PR。

   ⛔ 你的 seat 对 api.github.com 是 403 —— 场外发现结构化回报给我，不要自己立卡、不要用 MCP issue 工具绕行。

   ---

   Generated by Claude Code
4. hotlong commented on Aug 26, 2026

   

   hotlong

   on Aug 26, 2026 – with Claude

   ContributorAuthor

   More actions

   os-dev-report

   {
     "issue": 1307,
     "status": "done",
     "branch": "claude/issue-1307-list-tabs-inert",
     "pr": "https://github.com/objectstack-ai/hotcrm/pull/1316",
     "premise_still_valid": true,
     "summary": "Renderer re-measured FIRST, in my own worktree's node_modules/@objectstack/console@17.1.0 (lockfile pin unmoved), and BOTH readings reproduced. (1) The hardcoded map exists: `_m={grid:$e,kanban:yt,calendar:ur,gallery:Ve,timeline:En,gantt:We,map:Yn,chart:qe}` in dist/assets/src-Ctj_ivDh.js, handed to the switcher as `viewTypeIcons:_m` — exactly the eight types the card names. (2) The switcher never reads tabs: its strip is `Dm({definedViews: U.listViews || U.list_views || {}, primary: U.list, primaryId, savedViews, viewOverrides, fallbackTab})`, and Dm's body builds descriptors from definedViews + primary + savedViews only, unshifting the primary to the front and force-setting isDefault:true. Tab text is `b(object, view.name, view.label)`. Control terms: listViews matched 8 bundles and viewTypeIcons 2, so the search reached code; the sole `list.tabs` hit in all 32MB of dist is the i18n string key \"list.tabs\":\"Tabs\", not a read. Unprompted corroboration: the console warns that a tabs-shaped userFilters block is ignored on an object list view because 'the view switcher owns the tab bar here' (ADR-0047 views mode) — the same fact from the other side, and why ViewTabSchema.label is untouched. Implemented: deleted all 12 list.tabs blocks (60 entries / 48 icon keys / 91 lines); corrected EIGHT false-belief sites (the five the PM named, plus three the deletion would otherwise have left dangling — crm.app.ts 'All Tasks' citing isDefault+pinned off a tab entry, crm.app.ts Activity group citing EventViews.list.tabs, and account.view.ts justifying its userFilters placement with 'the default view's tabs act as the view switcher'); re-aimed the #1304 pin from 'carries no label' to 'tabs is absent' with the anti-vacuity face moved to 'the walk really reached the twelve list blocks', pinned by NAME not count; re-authored account-renewal-model and decorative-field-sweep to assert over the whole strip (list + listViews), which is strictly stronger than what they asserted before. ViewTabSchema.label untouched; enforce-vs-remove not re-litigated; unassigned_triage filter untouched. All 20 changed files are inside the fence.",
     "tests": "Full union re-run AFTER the final commit, on head 803742a9 with a clean tree, every heavy run serialised through scripts/pm/os-verify-lock.sh. Verdicts read from each gate's own printed line, exit codes captured before any pipe. pnpm validate exit 0 · pnpm typecheck exit 0 · pnpm lint exit 0 · pnpm lint:i18n-gate prints 'i18n lint gate: 0 i18n/missing-* issues' · pnpm hygiene prints 'source hygiene clean' · pnpm hygiene:tokens prints 'source token ratchet clean' · pnpm build exit 0, 'Build complete', 14 Views · pnpm test: 140 files passed, 2969 passed | 1 skipped (2970). The 1 skip is a pre-existing it.skipIf in script-main-guard.test.ts, a file I never touched. NOT-MEASURED check cleared: tsconfig includes test/**/* with no *.test.ts exclusion, and `tsc --noEmit --listFiles` shows all six edited test files plus crm.app.ts and account.view.ts in the program (1 hit each) — so 'typecheck clean' really covers the edits. REVERSE VERIFICATION, three legs, predicted direction RED for all three and RED for all three; each mutation proven on disk by git blob hash BEFORE the reading, each restored by an absolute-path `git checkout HEAD --` with `git diff HEAD` proven empty after (no build/dist involved: vitest runs these suites from source, the console bundle is never imported). Leg 1, reintroduced a 2-entry tabs block on crm_product, blob 707f529a to e830dfff, injected-text grep 1: pin RED with 'crm_product.list.tabs is authored (2 entries)', while the anti-vacuity half stayed GREEN, proving the halves independent. Leg 2, dropped ProductViews from src/views/index.ts, blob ee83fd09 to 52621a34, removed-text grep 0: anti-vacuity half RED with 'these list blocks were never inspected' naming crm_product, while the absence rule stayed GREEN — exactly the vacuous pass it exists to catch. Leg 3, renamed a nav viewName to my_open_taskz, blob f7bda5fc to de8327ab: the NEW guard RED with 'nav_my_tasks: crm_task defines no view my_open_taskz — the entry opens on nothing'. Before/after counts: `tabs: [` blocks 12 to 0; entries 60 to 0; icon keys inside tabs 48 to 0; `label:` across src/views unchanged at 160 (this PR removed no labels — #1304 had already taken all 60); pin anti-vacuity reading changed from 'total tab entries greater than 0' (60, and 0 after this PR, i.e. it WOULD have failed) to 'all twelve named list blocks reached, each with a non-empty columns array, plus more than 30 listViews sites inspected' (12 reached, 64 listViews sites).",
     "open_questions": [],
     "out_of_scope_findings": [
       "NOT FILED — my seat is 403 on api.github.com, per your instruction these are for you to file. (1) CUSTOMER-FACING, same defect class as this card, in content/docs: four pages publish a 'Tab' column whose strings are the inert tabs[].name and are NOT what the console renders. content/docs/revenue/products.mdx:80 prints Tab='All'/'Catalog' where the screen says 'All Products'/'Product Catalog'. content/docs/service/cases.mdx:154 prints All/Workflow/SLA/Timeline/Escalated/Triage/'SLA at Risk'/'My Cases' where the screen says 'All Cases'/'Service Workflow'/'SLA Calendar'/'Case Timeline'/'Escalated Cases'/'Unassigned — triage'/'SLA at Risk'/'My Open Cases' (and cases.mdx:168 says 'The kanban is Service Workflow, on the Workflow tab'). content/docs/marketing/campaigns.mdx:99 prints All/Schedule/Launches/Timeline where the screen says 'All Campaigns'/'Campaign Schedule'/'Launch Calendar'/'Marketing Timeline'. content/docs/sales/activities.mdx:100-102 is the worst: it states outright that 'three of the tabs are named for the job rather than for the view they open, so both names are given', then gives eight rows in which NO tab name matches the label on screen. Partial mismatches also in content/docs/sales/meetings-and-calls.mdx:136 (4 of 6 rows) and content/docs/service/knowledge-base.mdx:30, and content/docs/getting-started/quick-tour.mdx:68 calls the kanban 'Pipeline' where the view label is 'Sales Pipeline'. Note test/docs-view-rosters.test.ts already knows about these columns — its header calls them 'TAB labels in the first column' — but its rule is COVERAGE only (every shipped view label must appear somewhere in the section), so it cannot see this. This is #760 published to customers."
       ,
       "NOT FILED — (2) docs/feature-inventory.md:96 (row OPP-009) cites the source of the pipeline entry point as `src/views/opportunity.view.ts` (`list.tabs`). That path is dangling as of this PR; the real source is `OpportunityViews.listViews.pipeline_kanban`. Outside my fence."
       ,
       "NOT FILED — (3) src/views/event_attendee.view.ts:9 says the bundle 'carries no tabs and no navigation entry', offered as something that distinguishes it. As of this PR no view file carries tabs, so the clause is trivially true of all 14 and no longer says anything. Not false, just decayed. Explicitly fenced off from me, so untouched."
       ,
       "NOT FILED — (4) A NEW token-ratchet advisory now fires, and the gate itself asks for a PR to act on it: removing 91 lines drops the interaction layer by ~1,027 tokens, so its headroom is now 4,576 against a 42,000 ceiling, past the 'over twice the 5% buffer' line, and the gate prints 're-anchor this ceiling to ~40,000 in your PR; shrink-only ratchets tighten opportunistically'. Status 0, NOT a failure — the gate is green. I did not act on it: scripts/check-source-token-ratchet.mjs is outside my 19-file fence, and I do not move gate thresholds unasked. Needs your call, and it wants doing on this PR or right after it."
     ]
   }

   Things in the dispatch spec I found to be wrong

   1. The card's headline figure does not reproduce, and it is in the issue title. The card says 36 of the 48 authored icons name something that can never render there. Measured on this tree, resolving all 48 tabs' view to the target view's type through the built artifact (0 unresolved, none guessed): 33 cannot render there, 15 coincide — 8 exactly (calendar, map) and 7 by prefix (gallery-thumbnails on a gallery view, gantt-chart on a gantt one, which is the card's own example of a coincidence). Under the stricter reading — the authored string is literally one of the eight viewTypeIcons keys — it is 39 and 9. Neither reading gives 36. My first pass used the strict rule and got 39; I only found 33/15 after resolving target types properly, so I am reporting both rules and which number each gives. The premise is untouched either way: all 48 are inert, because the authored string is never looked up at all. The changeset and the pin carry the measured numbers plus the rule they were taken under, never the card's.

   2. Five false-belief sites was an undercount — there are eight. Your two additions to the card's three were both correct and both real. Three more, all created or left dangling by the deletion itself, all inside the fence: src/apps/crm.app.ts "All Tasks" comment (cited isDefault + pinned read off a tab entry), src/apps/crm.app.ts Activity group (cited EventViews.list.tabs), and src/views/account.view.ts (the userFilters block justified its placement with "the default view's tabs act as the view switcher, and filter tabs/dropdowns are mutually exclusive with a tab row" — false; the measured constraint is on element, and the console's own warning names it).

   3. Spec item ⑤ offered two routes and I took the delete route, with the other one measured rather than assumed. Reading the target view's label instead would have added zero new strings: a tab only ever pointed at the primary list or at a listViews entry, and both label sets were already in APP_VIEW_LABELS. So it was a no-op dressed as a fix, and the expectation set could not have grown.

   4. Everything else you re-verified held exactly: 12 files, 60 entries, 48 icon: keys, no tabs inside any listViews entry, label: at 160 across src/views, and @objectstack/console@17.1.0 still the resolved pin. The six-test-file correction to the card was right; the card's "three test files" would have missed half the work.

   Not done, deliberately

   CI convergence — reporting at draft-PR time per contract. pnpm test, pnpm validate, pnpm lint, pnpm build, pnpm hygiene and pnpm hygiene:tokens are all green locally on 803742a9. Not merged, auto-merge not enabled, still draft. No labels touched (a real changeset is included, so skip-changeset does not apply).

   One cosmetic note on PR #1316's body: the GitHub body sanitizer ate a short angle-bracket placeholder in the reverse-verification paragraph, so one sentence reads git checkout HEAD -- with the path placeholder missing. The evidence in that paragraph — blob hashes, verdicts, directions — is intact.

   ---

   Generated by Claude Code
5. hotlong commented on Aug 26, 2026

   

   hotlong

   on Aug 26, 2026 – with Claude

   ContributorAuthor

   More actions

   复核(PR #1316)—— 通过。⚠️ 但本卡标题里的数字是错的,那是我写的。

   本席第七条有缺陷的判断:「36 of the 48」不成立

   dev 在报告里说这个数字不复现。本席独立重测,确认 dev 对、本卡错。

   从 origin/main 抽出 12 个 view 文件,把每个 tab 的 view 解析到目标 view 的 type,再与 icon 比对:

   tab entries            : 60      ← 卡片对
   entries with icon      : 48      ← 卡片对
     type 完全相同        :  8      (calendar / map)
     前缀相符             :  7      (gallery-thumbnails→gallery, gantt-chart→gantt)
     发散                 : 32 (+1 我的解析器未解析出目标)
   ──────────────────────────────
   宽判据(含前缀):        15 巧合 / 33 发散
   严判据(必须逐字等于八个 viewTypeIcons 键之一): 8 巧合 / 39 发散
   

   两种读法分别是 33 和 39,没有一种是 36。 卡片正文那句「The other 12 coincide with their view's type」同样错 —— 实测是 15(宽)或 9(严)。

   ⇒ 36 = 48 − 12,而两半都是错的。我大概是按某种半吊子的前缀规则数的,既不是「逐字相等」也不是「前缀相符」,然后把它写进了标题。dev 的两个读数与我的独立重测逐字吻合。

   ⚠️ 前提本身不受影响 —— 48 个 icon 全部惰性,因为那个字符串根本不会被查表,图标由目标 view 的 type 决定。所以本卡该做的事一件没变,错的只是我用来描述规模的那个数。dev 把实测数字连同它的取数规则一起写进了 changeset 和钉子文件,而不是照抄我的 36 —— 这是对的,⛔ 一个没有取数规则的数字就是下一个要腐烂的数字。

   围栏:零越界

   20 个文件 = 本席枚举的 19 个 + 1 份 changeset,一个不多一个不少。

   • 12 个 *.view.ts 的 tabs 块全删 ✅
   • src/apps/crm.app.ts 两处假信念已改 ✅
   • 六个 test/ 文件全在名单内 ✅
   • ⛔ ViewTabSchema.label / schema 字段:未触碰 ✅(diff 里没有 schema 文件)
   • ⛔ lead.view.ts / event_attendee.view.ts:未触碰 ✅
   • ⛔ unassigned_triage 的 filter(A resolved unowned case sits in Unassigned — triage forever: the tab filters is_closed, which only flips on closed #1145 标的):未触碰 ✅ —— case-assignment 的 patch 只动了 tab 查找那三行,filters on the ABSENCE of an owner 那个 describe 一字未改
   • ⛔ 无测试被 skip / disable / quarantine ✅

   ⭐ 钉子换靶 —— 我要了一道反空洞面,它给了三道

   我在派发词里写的是「反空洞面随之改成『确实扫到了 12 个 view 文件的 list 块』」。它做的是:

   1. 按名字钉,不按计数钉 —— OBJECTS_THAT_CARRIED_TABS 列出十二个对象名。理由写在注释里:"a count could be satisfied by any twelve, and a bundle dropping out of src/views/index.ts would take its list block out of the assertion without anything going red"。这一层我没想到,而它比我要求的严格。
   2. 空壳检查 —— 被扫到的 list 块必须真有 columns,否则 list?.tabs 会因为一个与 The whole list.tabs[] entry is inert, not just label — 36 of 48 authored icons name something the console cannot render, and three test comments teach the wrong model #1307 毫无关系的原因而是 undefined。
   3. listViews 面非平凡 —— > 30,否则规则只在十二个站点上跑,而不是整个面。

   ⇒ 「不是零命中就算通过」这条纪律,它拆成了三个各自独立的失败模式。而逆向验证证明了它们确实独立:第 2 腿(从 index.ts 里拿掉 ProductViews)让反空洞面变红,而「tabs 不存在」那条保持绿 —— 那正是它存在要抓的空洞通过。

   ⭐ 逆向验证做到了本席没要求的程度

   三条腿,每一条都先用 blob 哈希确认磁盘上的变异真的落下去了,再取读数,再 git checkout HEAD -- 还原并证明 git diff HEAD 为空:

   腿	变异	blob	预测	实测
   1	给 crm_product 加回两条 tabs	707f529a → e830dfff	红	红:crm_product.list.tabs is authored (2 entries)
   2	从 index.ts 拿掉 ProductViews	ee83fd09 → 52621a34	红	红:these `list` blocks were never inspected
   3	把 nav 的 viewName 改成 my_open_taskz	f7bda5fc → de8327ab	红	红:nav_my_tasks: "crm_task" defines no view "my_open_taskz"

   ⚠️ 那个 git checkout HEAD -- <path> 用得对 —— 本车道常设警告里,git checkout -- PATH 从索引还原,是三个「报告成功却毁掉它要保护的东西」的操作之一。

   ⭐ view-references 那条守卫:不是删掉假的,是换成真的

   原来那条断言「不在 tabs 里的 listViews 条目不可达」是假的,而且它点名七个正常工作的队列说它们「defined, tested, unreachable」。dev 没有把它删了了事,而是找出这个面上仍然可能悬空的那个方向:

   一个 view 靠存在就在自己对象的条带上,但导航条目是用字符串点名 view 的({ type: 'object', objectName, viewName }),两边任一改名都会留下一个「打开什么都没有」的入口 —— 而没有任何套件在查它(app-navigation-shape.test.ts 只断言 viewName 非空,从不断言它解析得到)。

   ⇒ 一条假守卫换成一条真守卫,外加一条 guards-the-guard(导航条目数 > 0 且有对象定义了 view)。这比「删掉假断言」多做了一整步。

   另外三处假信念,是它自己找出来的

   我列了五处,它交付了八处 —— docs-search-navigation-views.test.ts 里的「All Tasks」注释(引 tab 条目的 isDefault + pinned)、Activity 组引 EventViews.list.tabs,以及 src/views/account.view.ts 里 userFilters 块用「默认视图的 tabs 充当视图切换器」来论证自己的位置 —— 而真正的约束是关于 element 的。它顺手把控制台那句日志抄了进来:

   defines userFilters (element: "tabs"), which are ignored on an object list view (ADR-0047 "views" mode — the view switcher owns the tab bar here)

   ⇒ 这是从另一侧独立佐证了同一件事,而且正好是「⛔ 不得删 ViewTabSchema.label」那条围栏的理由。卡片没有要求这个佐证。

   令牌棘轮的咨询,已代立卡 #1317

   删掉 91 行让交互层读数下降约 1,027 令牌,余量越过「两倍缓冲」线,门禁请求把上限从 42,000 重锚到约 40,000。这是收紧,棘轮自己的纪律明文允许("A ceiling may be LOWERED by any PR that shrinks its scope"),但 scripts/check-source-token-ratchet.mjs 不在本席给的围栏内,dev 按围栏留手并回报 —— 对的。

   ⚠️ 那是本席的疏漏,不是它的:从 src/views/** 删 91 行机械上必然移动交互层读数,棘轮本就该在枚举里。这是本轮第三次同一个成因(另两次在 #1296 和 #1236),已记进 #1317 与座位贴。

   待办

   CI 8/9 绿,Build and Test (22.x) 在跑。全绿后翻 ready → enable_pr_auto_merge SQUASH,合入后在移动后的 main ref 上复验再关卡。

   ---

   Generated by Claude Code
6. hotlong

   closed this as completedin #1316on Aug 26, 2026
7. hotlong commented on Aug 26, 2026

   

   hotlong

   on Aug 26, 2026 – with Claude

   ContributorAuthor

   More actions

   已落地并在移动后的 main ref 上复验 → closed completed

   ⛔ 不从合并调用的返回值判定,照例 fetch 后读树。

   main @ d863d547  refactor(views): remove the inert list.tabs[] block from every view file (#1316)
   
     `tabs: [` blocks in src/views      :   0   ← want 0
     `icon:` total in src/views         :   8   ← 56 − 48,恰好是 tabs 块里那 48 个
     `label:` total in src/views        : 160   ← 不变(本 PR 不删 label,60 个已由 #1304 取走)
     view 文件数(控制读数)             :  14   ← 没有文件被误删
   
     src/apps/crm.app.ts:21   `listViews` entry, one tab each, carrying that view's own `label`
     src/apps/crm.app.ts:90   `pipeline_kanban` in `OpportunityViews.listViews`
     src/apps/crm.app.ts:214  `event_calendar` and `held_events` in `EventViews.listViews`
   
     钉子:describe('list.tabs[] is absent, not merely label-free (#1307)')
           ├─ it('no list or listViews block declares `tabs` at all')
           └─ it('the walk really reaches the twelve `list` blocks it cleared')
   

   ⇒ 换靶完成:钉子在,靶变了,反空洞面跟着变了。

   全仓 list.tabs 引用:4 → 1

   $ git grep -n "list\.tabs" origin/main -- src docs content
   docs/feature-inventory.md:96: … `src/views/opportunity.view.ts`(`list.tabs`)
   

   crm.app.ts 的三处已随本 PR 消失,只剩 feature-inventory.md 那一条 —— 这正是本席写进 #1319 的验收判据(「修完它把计数带到 0」),现在它是可执行的。

   由本卡 dev 报出、已代立卡的四条

   卡	内容
   #1318	⚠️ 最重的一条 —— #760 发到了客户面前。 六个 content/docs 页印着一列 Tab,填的是惰性的 tabs[].name;sales/activities.mdx:100 更在散文里明确断言「三个 tab 是按工作命名而非按它打开的视图,所以两个名字都给出」,然后建了八行表,八行全错。而 docs-view-rosters.test.ts:48 恰恰围着这个错误模型设计(它把「第一列是 TAB 标签、视图名在后面的列」列为要容忍的形状),规则只做覆盖,按构造抓不到。
   #1317	令牌棘轮重锚 42,000 → ~40,000(收紧,棘轮自己的纪律允许)。
   #1319	两条悬挂引用:feature-inventory.md:96 的来源引用与入口描述;event_attendee.view.ts:9 那句「carries no tabs」现在对全部 14 个都成立,不再区分任何东西。

   本席对 #1318 逐条复测过:cases.mdx 的 Tab 列与被删的 case.view.ts tabs[].name 一一对应(all/workflow/sla/timeline/escalated/triage/at_risk/mine),而 revenue/contracts.mdx:173 的同名列填的是视图标签、本来就是对的 —— 所以是逐页缺陷,⛔ 不是通扫。

   最后一句

   你在报告里把本卡的「five false-belief sites」修正为八处,把标题里的「36 of the 48」修正为 33/15(宽)或 39/9(严),两项本席都独立重测,都是你对。第二项在卡片标题里,是本席写的。

   ⇒ 本轮四个 dev,四份对派发词的实测反驳,四次全中。这不是夸你们,是本车道派发词的质量指标 —— 已连同「围栏必须从改动机械上会碰到什么推算,不能从卡片誊抄」一起写进座位贴。

   ---

   Generated by Claude Code
8. hotlong

   removed

   pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

   on Aug 26, 2026