console open: page routes read the whole object registry (751 KB), the chrome's dashboard and report getters add a serial wave, and nine calls the page does not read run on open

Filed by the repo:cloud#2 seat (seat post objectstack-ai/objectstack#24145, session session_01Cbdek8SKSHxddGqTPZXdqX) from the objectstack-ai/cloud#2914 round 2 report (comment 6252346977, PR objectstack-ai/cloud#3027), out-of-scope findings 1–3: the console's remainder of cloud#2914's Targets 2–4. The maintainer reports opening an environment as slow today. Triage grades this; the seat suggests p1, the same as cloud#2914. ⛔ Not a claim.

What was measured

The measurement: one full console open at cloud's pins (objectui 43efd078, framework ec749fbb44b5), as round 1 measured it: Postgres behind a 50 ms RTT proxy, Chromium with CDP initiators, and the pinned console built with source maps.

Where the open already improved: with console open asks the same resource several times: environment-entitlements ×3 on the cloud pricing page (one fetch per hook instance), get-session ×2 and runtime/config ×2 on every open #12808, console splash waits on six serial request waves: the organisation chain (list → get-full-organization → get-active-member) and AppContent's object/dashboard/report/page metadata preload run before the first paint #12809 and GET /api/v1/meta/object loads the whole registry (97 objects, 745 KB) on every console open; the cloud control plane's pricing page reads no object #12811 in the pin, no duplicate URL appears on any of 22 opens, and the splash ends at 2.3–2.6 s (round 1: 4.0–4.2 s).
1. Page routes still read the whole object registry (cloud#2914 Target 3).
- Every measured control-plane open waits on GET /api/v1/meta/object: 97 objects, 751 KB, 307 ms of server time. The pricing page and the home are page/:pageName routes.
- packages/app-shell/src/console/appRouteMetadataReads.ts:63: page/:pageName reads ['page','object'].
- appObjectScope.ts:46: APP_OBJECT_SCOPE_ROUTES excludes page routes.
- AppContent.tsx:301: the objects getter.
- The home's package-scoped read (meta/object?package=com.objectstack.cloud) answers 28 B empty, because the control plane's objects belong to another package (ADR-0048). A package scope alone cannot serve it.
- GET /api/v1/meta/object loads the whole registry (97 objects, 745 KB) on every console open; the cloud control plane's pricing page reads no object #12811's triage (6234766460) named this re-measure as the pull for a new card.
2. A sixth serial wave before the splash leaves (cloud#2914 Target 2).
- meta/dashboard (useNavigationSync.ts:587, get dashboards) and meta/report (AppHeader.tsx:462, get reports) start only when the chrome mounts, after the route reads. That is about 0.35 s at 50 ms RTT.
- The splash left 6–71 ms after meta/report landed in 6 of 6 warm opens. This is a correlation; the dependency was not traced.
3. Calls the page does not read (cloud#2914 Target 4's budget of 15).
- A warm pricing open makes 24 calls, and the home 26.
- Nine calls the pricing page does not read: meta/dashboard, meta/report, approvals/requests (sharedUserFeeds.ts:683), data/sys_activity (:1037), data/sys_inbox_message (:1229), ai/agents (plugin-chatbot useAgents.ts:237), data/sys_user/ID (useUserLocale.ts:269) beside me/localization, api/settings (WorkspaceTimezonePrompt.tsx:88), and i18n/locales.

Wanted

A page route reads only the objects its page binds, or a scoped set the host declares. It does not read the whole registry.
The chrome's dashboard and report reads are off the splash's critical path (after first paint, with skeletons).
The bell feeds, the agents list, the timezone prompt's settings read and the duplicate user-locale read load after first paint, or only when opened.

Acceptance

The same measurement on a control-plane page route: no whole-registry meta/object read, the splash ends without waiting on the dashboard and report reads, and a warm open makes ≤ 15 calls.
A test per item.

Dedupe

300 objectui issues (open and closed, by last update) scanned for appObjectScope, "page route" with meta/object, "sixth wave", sharedUserFeeds on open and useAgents on open: #12811 and its PR #12871 (closed, the app-route half) are the only neighbours.

Generated by Claude Code

Triage: p1 · `domain:ui` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-18T16:57Z

Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015). ⛔ Not a claim, ⛔ not a dispatch.

Grade p1, as suggested. The maintainer reports opening an environment as slow today. This is the console's remainder of objectstack-ai/cloud#2914 (p1), measured: the whole object registry (751 KB) on page routes, a sixth serial wave before the splash leaves, and nine calls the page does not read.
Lane domain:ui. The three findings can land as separate PRs, in the card's order. The registry read on page routes is the largest share.
Paired platform half: data reads: every GET /api/v1/data read makes five serial sys_metadata reads of the target object's own active row (about 250 ms at 50 ms RTT) objectstack#24371 (five serial sys_metadata reads per data read, p1, domain:engine). Together they are the two halves of the same slow open.

Generated by Claude Code

Claim: PM loop round 10 (priority:p1; part 1 of 3: the page routes' whole-registry read)
Session: session_01XVp19E1MYUTFM3dnkwz8gB
Account: baozhoutao (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-12980-page-route-objects
Worktree: objectui-issue-12980
Domain: domain:ui
Seat: domain:ui#2
File surface: Wanted item 1 only, as triage ordered: packages/app-shell/src/console/appRouteMetadataReads.ts (page/:pageName's reads), packages/app-shell/src/console/appObjectScope.ts (APP_OBJECT_SCOPE_ROUTES), the objects getter in packages/app-shell/src/console/AppContent.tsx (about :301), the data-source object read they call (packages/data-objectstack/src/ only if a scoped read needs a parameter the client already declares), and tests beside each. Plus one .changeset/12980-*.md. ⛔ Not items 2 and 3 (the chrome's dashboard / report wave; the bell feeds, agents, timezone and locale reads): later claims on this card. The PR says Part of #12980. Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui reads no path-derived mandate)
Clause-②: yes
Clause-② note: amended in place by the seat rulings on this card: readObjectScope reads names without packageId as a names-only scope, which widens the documented accept set of @object-ui/react's AppShellContext contract, so @object-ui/react levels at least minor and one CONTRACT_REVIEW_TIER review is owed before the queue.
File surface widened in place by the same rulings: packages/app-shell/src/views/PageView.tsx, InterfaceListPage.tsx, nav-menu-renderer.tsx, global-search-renderer.tsx, packages/app-shell/src/providers/MetadataProvider.tsx, packages/react/src/context/AppShellContext.tsx (the docblock), and the openPathObjectReaders-12811 census rows.
Responsibility: @object-ui/app-shell's page/:pageName route reads ['page','object'], the whole object registry, and the app-object scope excludes page routes | objectui#12811 (PR #12871) scoped the app routes' object read, and the server already serves meta/object with a scope filter | every console open on a page route; the control plane's pricing page and home measured (751 KB, 307 ms of server time) in objectstack-ai/cloud#2914 round 2.
Thread-read: 6252596234
Serial constraints cleared: no open objectui PR touches appRouteMetadataReads.ts, appObjectScope.ts or AppContent.tsx; the in-flight claims (objectui#12954, #12957) name none of them. This seat's objectui#12977 (console/organizations/) is disjoint.
Claimed at: 2026-10-18T17:12Z

os-dev-report

{
  "issue": 12980,
  "status": "blocked",
  "branch": "claude/issue-12980-page-route-objects",
  "pr": null,
  "session": "session_01XVp19E1MYUTFM3dnkwz8gB (the dispatching seat's; this run is its subagent)",
  "premise_still_valid": true,
  "summary": "Item 1's premise holds on objectui origin/main 76f91f80d9: `page/:pageName` reads ['page','object'] in APP_ROUTE_METADATA_READS, APP_OBJECT_SCOPE_ROUTES does not name it, so AppContent's allObjects falls to `metadata.objects` (the whole-registry GET meta/object). Stopped on the file-surface STOP condition before any edit: the ruling cannot hold inside the claimed surface, because PageView.tsx (not in the claim) reads `useMetadata().objects` itself at render (ConsoleActionRuntimeProvider, RelatedRecordActionsBridge and pageRecordActionsValue's routable set), so scoping AppContent alone leaves the whole-registry read on every page open. Three page blocks read the whole registry the same way (InterfaceListPage.tsx, nav-menu-renderer.tsx, global-search-renderer.tsx), and an app item with no `_packageId` has no names-only read in MetadataProvider.readObjectScope (no packageId = whole registry by its documented contract). Needed claim amendment and one decision are below; the empty branch is pushed as the claim marker, nothing else was written.",
  "tests": "None run: no code was written (stopped before the first edit). Readings, all on objectui origin/main 76f91f80d9 and objectstack origin/main 6a49efb9c4: (A1) server bounded reads for meta/object = `?package=` only (meta-collection-routes.ts list handler passes `packageId` from metaItemPackageBinding(req.query.package); the list chain createMetaListAnswer reads only `id` (apps), `object` (views), `include` (docs), `locale`) plus the per-object GET meta/object/NAME; no names list. The pinned client @objectstack/client 18.0.0-next.3 meta.getItems options = { packageId } only. Per-object reads are what a page needs: the two measured control-plane pages bind ZERO objects (cloud packages/service-tenant/src/pages/pricing.page.ts and welcome.page.ts: every dataSource is `provider: 'api'`, HOME_SUMMARY_BINDING / WARM_BINDING; no objectName), so derive-from-page reads 0 objects there. Arithmetic at the card's 50 ms RTT: today the page wave waits on meta/object, at least 50 ms RTT + 307 ms server time + 751 KB transfer, about 357 ms or more before transfer; after: 0 object requests on those two pages. A page binding N objects outside its package pays one added serial wave of about 50 ms + the slowest single per-object server time, the N reads in parallel. (A2) waves today: app list, then page and object in parallel (1 route wave). After: page and the app package list in parallel (the chrome's useNavMemberGates / useServedViewItems already read readObjectScope({packageId})), then the page's bound names by name only when it binds objects the package does not hold: 1 route wave for pages binding none (both measured pages), 2 for the rest. (A3) page-route readers that assume the whole registry: AppContent allObjects (ConsoleLayout, CommandPalette, useTrackRouteAsRecent) [in surface]; PageView.tsx [NOT in surface]; InterfaceListPage.tsx (finds `cfg.source` in objects), nav-menu-renderer.tsx (routable-object set), global-search-renderer.tsx (searchable objects) [NOT in surface]; useNavMemberGates / useServedViewItems / useObjectPresence already read the package scope while the object type is idle (no change). The census openPathObjectReaders-12811.test.ts classifies PageView, InterfaceListPage, nav-menu-renderer and global-search-renderer as `off` with the why 'a page route', which is the row set this card moves to `scope`.",
  "mcp_calls": "1 — claude-code-remote add_repo (objectstack-ai/cloud, read) to read the measured pages; zero GitHub MCP calls, no MCP write tool",
  "api_writes": "2 — git push of the empty branch claude/issue-12980-page-route-objects (not REST); POST repos/objectstack-ai/objectui/issues/12980/comments (this os-dev-report, via scripts/pm/post-stamped.mjs)",
  "open_questions": [
    {
      "question": "File surface: the ruling needs files the claim does not name. Amend the claim to add them?",
      "options": [
        "A: add packages/app-shell/src/views/PageView.tsx (take the route's bounded objects from AppContent, fetch an unlisted block object by name on demand) plus InterfaceListPage.tsx, nav-menu-renderer.tsx and global-search-renderer.tsx, and the census row edits in console/__tests__/openPathObjectReaders-12811.test.ts",
        "B: add PageView.tsx only; leave the three page blocks reading the whole registry when such a block is on the page (a page that places nav:menu, global:search or an interface source still reads 751 KB), recorded as a named gap with Part of #12980",
        "C: keep the surface; deliver AppContent only (does not meet the ruling: PageView still reads the whole registry on every page open)"
      ],
      "recommendation": "A, because the ruling is absolute (a page route never reads the whole registry) and B leaves the same read behind any page that places one of three shipped blocks; C would land a PR whose own pin (a) cannot go green"
    },
    {
      "question": "An app item without `_packageId`: MetadataProvider.readObjectScope with no packageId answers the whole registry (documented in @object-ui/react MetadataContextValue.readObjectScope). How does a page route in such an app read only its page's objects?",
      "options": [
        "A: extend readObjectScope so `names` without `packageId` is a names-only scope (MetadataProvider.tsx and the AppShellContext.tsx docblock join the surface; no member added, Clause-② no, the docblock's 'no packageId = whole registry' becomes 'no packageId and no names')",
        "B: such an app keeps the whole registry on page routes, as 12811 kept it for package-less apps (a narrower ruling than Section 1 states)",
        "C: PageView asks getItem('object', NAME) itself per bound name, bypassing the scope cache"
      ],
      "recommendation": "B if the PM can show every served app carries `_packageId` (then the branch is test-only), otherwise A, because it keeps one scope cache and one read path (C duplicates the by-name bookkeeping readObjectScope already owns); four-axis: A is contract-first and adds no new surface, B is the startup-scope default if no producer serves a package-less app"
    }
  ],
  "out_of_scope_findings": [
    "carrier: the claim amendment above (PageView.tsx, InterfaceListPage.tsx, nav-menu-renderer.tsx, global-search-renderer.tsx) · noted, not filed: they are this card's item 1, not a separate finding"
  ],
  "gates": "NOT MEASURED: every family (no diff exists), reason: stopped on the file-surface condition before the first edit",
  "files_changed": [],
  "deviations": "add_repo attached objectstack-ai/cloud (read) to read the measured pages' definitions over REST; not cloned. Worktree /home/user/objectui-issue-12980 removed after the report (no commits, clean)."
}

Generated by Claude Code

Seat rulings on the part-1 dev's open questions · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T17:20Z

These answer os-dev-report 6252817084 (stopped before the first edit on the file-surface condition; readings on objectui 76f91f8, objectstack 6a49efb).

Surface: A. The claim widens to packages/app-shell/src/views/PageView.tsx and the three page blocks that read the whole registry: InterfaceListPage.tsx, nav-menu-renderer.tsx and global-search-renderer.tsx. It also covers the census rows in console/__tests__/openPathObjectReaders-12811.test.ts, which move from off to scope. The ruling is absolute (a page route never reads the whole registry), and B or C would leave the 751 KB read behind every page open or behind three shipped blocks.
A package-less app: A. readObjectScope treats names without packageId as a names-only scope. MetadataProvider.tsx and @object-ui/react's AppShellContext.tsx docblock join the surface, and the rule becomes "no packageId and no names = the whole registry". B needs proof that every served app carries _packageId, and this seat has none. C would duplicate the by-name bookkeeping readObjectScope already owns.
- Clause-②: yes. This widens the documented accept set of a published @object-ui/react contract (names alone now means something it did not), even with no new member. So @object-ui/react is at least minor, and one CONTRACT_REVIEW_TIER review is owed before the queue. The claim is amended in place.

Readings carried into the order: the server's only bounded list read is ?package=, beside per-object GET meta/object/NAME; there is no names list. The two measured control-plane pages (pricing, welcome) bind zero objects, so they read none after the change. A page binding N objects outside its package pays one added parallel wave (about 50 ms plus the slowest single read at the card's RTT).

Generated by Claude Code

os-dev-report

{
  "issue": 12980,
  "status": "done",
  "branch": "claude/issue-12980-page-route-objects",
  "pr": "https://github.com/objectstack-ai/objectui/pull/12984",
  "session": "session_01XVp19E1MYUTFM3dnkwz8gB (the dispatching seat's id; this run is its subagent)",
  "premise_still_valid": true,
  "summary": "Part 1 is done under the seat's rulings (comment 6252837193; the claim was widened in place). The names-only scope is in: readObjectScope({ names }) with no packageId now reads each name by name. In MetadataProvider the scope key is '', and a refresh asks no package list. The AppShellContext docblock now says 'no packageId and no names = whole registry'. Clause-② yes: @object-ui/react minor, app-shell patch. `page/:pageName` joined APP_OBJECT_SCOPE_ROUTES. A page route now reads the objects its page definition binds; pageObjectNames derives them from the page's object, interfaceConfig.source, and every objectName and dataSource.object at any depth. They are read by name, beside the app's package list or through the names-only scope, and never the whole registry. The chrome gets the same scope. PageView reads the same scope, and reads any other object a block asks its navigator about by name on demand. InterfaceListPage reads its source object by name, nav:menu's capability gate reads its requiresObject names by name, and global:search reads the registry only once the user types. Named gap, outside this surface: in an app without _packageId, the chrome hooks useNavMemberGates, useServedViewItems and useObjectPresence still read `objects`, as #12811 left them.",
  "tests": "At da32c4273c (final head). New pins: AppContent.pageRouteObjects-12980.test.tsx 4/4, covering (a) package plus by-name reads with no 'object' call, (a) a zero-binding page in an app whose package holds no objects reads only the package list plus the one object a block asked about, (b) CONTROL: the #12811 app-route call set is unchanged, (c) an on-demand read whose href appears when it lands. MetadataProvider.namesOnlyScope-12980.test.tsx 5/5. pageObjectReads-12980.test.tsx 4/4 (PageView in a package-less app; the three blocks). Families, run as `pnpm exec vitest run --maxWorkers=1` over 58 files (those naming appRouteMetadataReads, APP_ROUTE_METADATA_READS, appObjectScope, APP_OBJECT_SCOPE_ROUTES, readObjectScope, PageView, InterfaceListPage, nav-menu, global-search and the census, plus AppContent.objectScope-12811 and the scripts suites that read these files): 'Test Files 58 passed (58) / Tests 571 passed (571)'. Lock verdicts: command-exit 0, 547s. Type-check: `pnpm --filter @object-ui/app-shell --filter @object-ui/react run type-check` printed 'type-check: Done' for both, after a closure build (`turbo build --filter=@object-ui/app-shell^...`, 28/28). Ablation, once, with scripts/ablation-replace.mjs: removing 'page/:pageName' from APP_OBJECT_SCOPE_ROUTES took the anchor from 1 to 0 and the blob from e3475fd4 to 567a1e43, and the result was 'Tests 3 failed | 1 passed': both (a) pins and (c) went red, and (b) CONTROL stayed green. Restore was proven: the blob after restore is e3475fd4, equal to HEAD, and `git diff HEAD` is empty. The tests import src relatively, so no dist build was needed for the ablation.",
  "mcp_calls": "1 (earlier leg): claude-code-remote add_repo (objectstack-ai/cloud, read access). Zero GitHub MCP calls and no MCP write tool.",
  "api_writes": "3 REST writes through the fleet-write relay: POST issues/12980/comments (first-leg os-dev-report 6252817084); POST pulls (draft #12984) together with its assignee POST issues/12984/assignees (one pr_create action, read back identical, assignee baozhoutao); POST issues/12980/comments (this report). Not REST writes and not counted: 4 git pushes of the branch. Label writes: 0, because the order named no label and objectui path labels come from labeler.yml.",
  "open_questions": [],
  "out_of_scope_findings": [
    "class: a · reach: a page route open in an app item without _packageId still issues GET meta/object, through the chrome's useNavMemberGates (sidebar), useServedViewItems (header) and useObjectPresence (bell), which fall back to `objects` when the app has no package · evidence: those hooks' 'no package = whole registry' branch, unchanged since #12811; the AppContent.objectScope-12811 CONTROL pins it on app routes · dedupe words: names-only scope, package-less app, useNavMemberGates objects, page route registry · carrier: a follow-up for this card's seat (these files were not in the widened surface); also noted in the PR's Acceptance notes"
  ],
  "gates": "exit 0 each: pnpm check:phantom-deps, check:self-import, check:unreferenced-sources, check:doc-example-readers, check:handler-key-reads, check:side-effects-array, check:element-data-source-declaration, check:spec-symbols, check:action-forward-parity, check:designer-field-key-parity, check:icon-record-names, check:i18n-keys, check:i18n-drift, check:i18n-designer-parity, check:control-bytes, check:new-line-citations, check:test-path-roots; node scripts/check-hand-rolled-comment-mask.mjs; check-changeset-fixed.mjs; check-changeset-presence.mjs; check-changeset-no-major.mjs. eslint on the 13 changed ts/tsx files (inline config honoured, as the package lint script runs it): exit 0, 0 errors; the warnings added on changed lines are all @typescript-eslint/no-explicit-any. Type-check: app-shell and react passed. CI is declared the full package directories of @object-ui/app-shell and @object-ui/react, and the repo-wide lint. check:spec-floors was not run.",
  "files_changed": [
    "packages/react/src/context/AppShellContext.tsx",
    "packages/app-shell/src/providers/MetadataProvider.tsx",
    "packages/app-shell/src/console/appObjectScope.ts",
    "packages/app-shell/src/console/AppContent.tsx",
    "packages/app-shell/src/console/appRouteMetadataReads.ts",
    "packages/app-shell/src/views/PageView.tsx",
    "packages/app-shell/src/views/InterfaceListPage.tsx",
    "packages/app-shell/src/views/nav-menu-renderer.tsx",
    "packages/app-shell/src/views/global-search-renderer.tsx",
    "packages/app-shell/src/console/__tests__/openPathObjectReaders-12811.test.ts",
    "packages/app-shell/src/console/__tests__/AppContent.pageRouteObjects-12980.test.tsx",
    "packages/app-shell/src/providers/__tests__/MetadataProvider.namesOnlyScope-12980.test.tsx",
    "packages/app-shell/src/views/__tests__/pageObjectReads-12980.test.tsx",
    ".changeset/12980-page-route-objects.md"
  ],
  "deviations": "1) The ablation differs from the order's literal one ('restore [page, object]'), because the route table row keeps ['page','object']: the record overlay on a page route still waits on the whole registry. I ablated the equivalent change instead, removing 'page/:pageName' from APP_OBJECT_SCOPE_ROUTES. 2) global-search-renderer's census row moved from off to `action`, not `scope`: a search covers every object, so it reads them after ensureType('object') once the user types. 3) appRouteMetadataReads.ts received only a comment on the page row. 4) The pricing and welcome page definitions were read from objectstack-ai/cloud over REST, after add_repo in the first leg. 5) The PR footer is the session-URL form the role file requires, not the harness's emoji footer."
}

Generated by Claude Code

ACCEPT (part 1 of 3) · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T18:11Z

PR #12984 at head da32c42 (Part of #12980), reviewed against the claim 6252756345 (amended in place: Clause-②: yes, surface widened), the seat rulings 6252837193, the dev reports 6252817084 and 6253248137, and the contract review 6253321662.

What changed (Wanted 1): a page/:pageName route no longer reads the whole object registry (751 KB). page/:pageName joins APP_OBJECT_SCOPE_ROUTES. pageObjectNames derives the objects a page binds from its definition (object, interfaceConfig.source, every objectName, every dataSource.object), and they are read by name beside the app's package list. PageView reads the same scope and reads any other object a block asks for by name on demand. InterfaceListPage reads its source by name, nav:menu's capability gate reads its requiresObject names by name, and global:search reads the registry only once the user types. @object-ui/react's readObjectScope treats names without packageId as a names-only scope (the docblock rule becomes "no packageId and no names = the whole registry"). The two measured control-plane pages (pricing, welcome) bind no objects, so they now read only the package list the chrome reads anyway.
Clause-②: yes (widening): the documented accept set of @object-ui/react's readObjectScope widens; no member is added. Changeset @object-ui/react: minor, @object-ui/app-shell: patch.
Contract review: PASS 6253321662 at CONTRACT_REVIEW_TIER on head da32c42. Existing packageId and no-argument callers hit byte-identical branches. Nothing new is exported from the app-shell barrel. The route-table row keeping ['page','object'] for the record overlay, while the scope list intercepts the open, is GET /api/v1/meta/object loads the whole registry (97 objects, 745 KB) on every console open; the cloud control plane's pricing page reads no object #12811's own design.
Tests: AppContent.pageRouteObjects-12980 (4), MetadataProvider.namesOnlyScope-12980 (5) and pageObjectReads-12980 (4) pin: no whole-registry read on a page open; a zero-binding page reading only the package list plus one on-demand object; the GET /api/v1/meta/object loads the whole registry (97 objects, 745 KB) on every console open; the cloud control plane's pricing page reads no object #12811 app-route call set unchanged (CONTROL); the on-demand href; the names-only scope; and the three blocks. Ablation (page/:pageName removed from the scope list) turns both (a) pins and (c) red with the CONTROL green; restore proven by blob hash. 58 files / 571 tests pass; both type-checks and the pre-tsc battery exit 0.
Carried on this card for the next claim (not filed separately; this card stays open):
1. In an app without _packageId, the chrome hooks useNavMemberGates, useServedViewItems and useObjectPresence still read the whole registry, as GET /api/v1/meta/object loads the whole registry (97 objects, 745 KB) on every console open; the cloud control plane's pricing page reads no object #12811 left them.
2. pageObjectNames does not yet collect data.object (the ViewDataSchema object arm on object-grid / object-map, the canonical spelling), element:metadata_viewer.object, record:line_items.childObject / parentObject or addPicker.object. Such an object is read by name on demand, never through the whole registry, but it is outside the route's scope until then.
3. Inside a package app, InterfaceListPage's names-only read likely fetches its source a second time; passing the app's packageId, as nav:menu does, would reuse the route's scope.
Items 2 and 3 of the card (the chrome's dashboard / report wave; the calls the page does not read) are still open.
Trailers: all three commits carry only the model-free pair.
CI at head: all 36 check-runs on da32c42 are complete: 33 success, 3 skipped (dependabot and the two coverage jobs), none failed; mergeable_state: clean. Path surface (14 files, +1004 / −41): packages/app-shell/src/{console,providers,views}/**, packages/react/src/context/AppShellContext.tsx (docblock) and one changeset; no governed path.

Next: pr_ready + automerge_enable.

Generated by Claude Code

Landed (part 1 of 3) · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T18:26Z

PR #12984 (Part of #12980) merged through the merge queue as 024b7b66594952958c0a3a1d869b94791161e5f0, which is on main. Verified by content: all 14 of the PR's files on main are blob-identical to the accepted head da32c42. The merge commit's message names no model.

What landed: a page/:pageName route reads only the objects its page binds, never the whole object registry. The two measured control-plane pages bind none, so they read only the package list. @object-ui/react's readObjectScope takes a names-only scope (minor). Contract review PASS 6253321662.

This card stays open and stays with this seat. Still to do: Wanted item 2 (the chrome's dashboard / report reads off the splash's critical path), claimed next; Wanted item 3 (the calls the page does not read); and the three part-1 leftovers in the ACCEPT 6253365954. pm:dispatched stays.

For cloud (objectstack-ai/cloud#2914): the page-route half of the slow open applies once cloud's objectui pin covers the merge commit above.

Generated by Claude Code

Claim: PM loop round 11 (priority:p1; part 2 of 3: the chrome's dashboard / report wave off the splash's critical path)
Session: session_01XVp19E1MYUTFM3dnkwz8gB
Account: baozhoutao (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-12980-chrome-wave
Worktree: objectui-issue-12980b
Domain: domain:ui
Seat: domain:ui#2
File surface: Wanted item 2 only: the chrome's meta/dashboard read (packages/app-shell/src/hooks/useNavigationSync.ts, about :587) and meta/report read (packages/app-shell/src/layout/AppHeader.tsx, about :462); whatever holds the splash (the console's boot / splash gate under packages/app-shell/src/console/ or layout/) only if it waits on those reads; and tests beside each. Plus one .changeset/12980-*.md. ⛔ Not item 3 (the bell feeds, agents, timezone and locale reads) and ⛔ not part 1's leftovers: later claims on this card. The PR says Part of #12980. Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui reads no path-derived mandate)
Clause-②: no
Responsibility: @object-ui/app-shell's chrome starts meta/dashboard and meta/report only when it mounts, after the route reads, a sixth serial wave the splash appears to wait on | part 1 (PR #12984, 024b7b6) already moved page routes off the whole-registry read | every console open, about 0.35 s at the card's 50 ms RTT (objectstack-ai/cloud#2914 round 2).
Thread-read: 6253522300
Serial constraints cleared: part 1 (PR #12984) landed as 024b7b6. No open objectui PR touches AppHeader.tsx, useNavigationSync.ts, AppContent.tsx or a splash gate; the in-flight claims (objectui#12954, #12957) name none of them.
Claimed at: 2026-10-18T18:27Z

Landed · objectui#12980 part 5 · PR #13003 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:20Z

Merged as d3fd5e4 on main. Verified by content: all 12 files in d3fd5e4^1..d3fd5e4 are blob-identical to the accepted head 3c48263. The squash message is model-free.
What landed:
- A page route's scope read now also holds objects bound through data.object (the object arm), parentDataSource.object, a state_machine metadata viewer's object, and record:line_items' child and parent objects.
- An interface list page inside a package app reuses the route's scope instead of reading its source again.
The card stays open with pm:dispatched and its assignee. Part 6 (part 1's leftover 1, the package-less chrome hooks) is in flight (claim 6256858128). The dashboard route's chunk-graph wait at boot is still carried.

Generated by Claude Code

os-dev-report

{
  "issue": 12980,
  "status": "done",
  "branch": "claude/issue-12980-packageless-chrome-scope",
  "pr": "https://github.com/objectstack-ai/objectui/pull/13005",
  "session": "session_01XVp19E1MYUTFM3dnkwz8gB (subagent: the parent's id)",
  "premise_still_valid": true,
  "summary": "P0: package-less apps are real. The console's CreateAppPage saves with meta.saveItem('app', name, schema) and names no package, and objectstack serves such a row without _packageId (withRowBinding / boundPackageOf; DERIVED_PROVENANCE_KEYS strips any stored copy). The showcase and CRM apps are package-stamped. P1: useObjectPresence and useServedViewItems are class (a) and now read the names-only scope in an app that has a name and no _packageId while the object type is idle. Presence asks its one name. Served view items go through the app's requiresObject names (asked by AppContent anyway), and read nothing when the app requires none. Outside an app, once the whole registry is loaded, and in package apps, behaviour is unchanged. useNavMemberGates is NOT changed: the fork rule fired (see open_questions). Until it is ruled, a package-less page open still reads the whole registry once, from the sidebar's gate.",
  "tests": "At head 38c40e92bc: pnpm exec vitest run --maxWorkers=1 over 11 files (the new packagelessChromeScope-12980.test.tsx, plus every test naming useNavMemberGates / useServedViewItems / useObjectPresence / objectPresence / pageRouteObjects-12980, plus MetadataProvider.namesOnlyScope-12980) -> 'Test Files 11 passed (11) / Tests 75 passed (75)', lock VERDICT command-exit 0. Pre-fix measurement (rendered test): one bare 'object' call per mount from useServedViewItems / useObjectPresence, and one from useNavMemberGates (throwaway test, deleted). Under the whole registry, checkCapability('object','other_missing') answered false. Ablation via objectstack scripts/ablation-replace.mjs, with a trap restore. (1) useObjectPresence, condition narrowed back to packageId: anchor 1->0, blob 55f29b78ee2a->69aabf05eec9. All 3 (a) tests failed ('expected [ object/note, object ] to not include object'), 3 CONTROLs passed. Restored: blob == HEAD 55f29b78ee2a, git diff HEAD empty. (2) useServedViewItems, package-less branch disabled: blob 964149b01cd7->37c7dc7addcd. All 3 (a) tests failed ('expected [ object ] to not include object'), CONTROLs passed. Restored: blob == HEAD 964149b01cd7, diff empty. No build/dist leg: the mutated modules are imported by relative path from the same package's tests. Type-check: closure build pnpm --workspace-concurrency=2 --filter '@object-ui/app-shell^...' build BUILD_EXIT=0. pnpm --filter @object-ui/app-shell type-check (tsc --noEmit && tsc -p tsconfig.test.json) exit 0 at 38c40e92bc. --listFiles counts the new test and both hooks (3 hits). Lint: eslint --no-inline-config --format json on the 3 changed files -> 3 files, 0 errors, 0 warnings. The population is the 3 changed files. eslint.config.js sets no parserOptions.project/projectService, so linting is not type-aware and the diff cannot move any untouched file's verdict. The repo-wide pnpm lint is declared to CI. Battery: all 15 check:* in the order, check-hand-rolled-comment-mask, check-changeset-fixed, check-changeset-presence and check:new-line-citations exit 0, each printing its own OK line. Declared to CI: the full packages/app-shell/ directory.",
  "mcp_calls": "0",
  "api_writes": "2 relay dispatches (fleet-write, as objectstack-fleet[bot]): (1) pr_create -> POST /repos/objectstack-ai/objectui/pulls (draft) + POST /repos/objectstack-ai/objectui/issues/13005/assignees [baozhoutao]; read-back identical, 7397 bytes. (2) os-dev-report comment -> POST /repos/objectstack-ai/objectui/issues/12980/comments. Also 3 git pushes (not REST). Zero label writes: the order names no label, and objectui's labeler owns path labels.",
  "gates": [
    "vitest targeted families (11 files / 75 tests) @38c40e92bc: exit 0",
    "closure build --filter '@object-ui/app-shell^...': exit 0",
    "@object-ui/app-shell type-check @38c40e92bc: exit 0",
    "eslint --no-inline-config on 3 changed files: exit 0 (0 errors, 0 warnings)",
    "check:phantom-deps / self-import / unreferenced-sources / doc-example-readers / handler-key-reads / side-effects-array / element-data-source-declaration / spec-symbols / action-forward-parity / designer-field-key-parity / icon-record-names / i18n-keys / i18n-drift / i18n-designer-parity / control-bytes: all exit 0",
    "scripts/check-hand-rolled-comment-mask.mjs: exit 0",
    "scripts/check-changeset-fixed.mjs: exit 0",
    "scripts/check-changeset-presence.mjs: exit 0",
    "check:new-line-citations: exit 0 (0 new)",
    "NOT MEASURED locally: the full packages/app-shell/ directory (the order rules out whole-package runs) and the repo-wide pnpm lint. Both declared to CI."
  ],
  "files_changed": [
    "packages/app-shell/src/hooks/useObjectPresence.ts",
    "packages/app-shell/src/hooks/useServedViewItems.ts",
    "packages/app-shell/src/hooks/__tests__/packagelessChromeScope-12980.test.tsx",
    ".changeset/12980-packageless-chrome-scope.md"
  ],
  "line_budget": "n/a: no skills/** or governed path touched",
  "deviations": [
    "The order conflicts with itself. Gates says to run 'each touched package's dir under the lock'; Lock says 'No whole-package-directory runs ... declare the full packages/app-shell/ directory to CI'. I followed the narrower Lock clause and declared the directory to CI.",
    "The harness attribution reminder asks for a model-named Co-Authored-By trailer and a different PR footer. Per the order and AGENTS.md, commits carry only the model-free pair, and the PR body ends with the role file's session-URL footer.",
    "The P1 measurement of useNavMemberGates used a throwaway test file in the worktree. It was deleted before any commit and never pushed."
  ],
  "open_questions": [
    {
      "question": "useNavMemberGates in a package-less app: the order's fork rule fired. A names-only scope over this app's requiresObject names answers the capability gate differently from the whole registry for some nav items. (1) activityObjectRoster (the bell's activity feed) runs this one gate over EVERY app's navigation and areas. Measured: under the whole registry, another app's entry requiring an object the runtime lacks is hidden (checkCapability false). Under the names-only scope it reads as registered (registeredObjectNames.has || !missingObjectNames.has). (2) readAppObjectScope names requiresObject from app.navigation only, not from app.areas[].navigation, which UnifiedSidebar draws. Which name set should the gate read in an app without _packageId?",
      "options": [
        "A: Leave the hook on the whole registry. Business need: exact gates, but the 751 KB read stays on every page open of a console-created app. Long-term: an inconsistency with package apps remains. AI-error: none added. Scope: zero work.",
        "B: This app's requiresObject names, as part 1's pageObjectNames path asks them. The roster divergence is accepted, as package apps already accept it under #12811. Business need: removes the read; the divergence only adds an object_name to the activity feed's $in filter (no visible rows). The app's own area entries also stay shown when the runtime lacks their object. Long-term: inherits the areas gap. AI-error: the gate would be lenient (unknown reads as registered), the consumer-tolerance shape the framework warns against. Scope: small, inside the hook.",
        "C: The union of requiresObject across every app's navigation and areas (from the app list the chrome already holds), through the names-only scope. Business need: removes the read, and every gate answers exactly as the whole registry would, at one by-name GET per distinct required object (usually a handful). Long-term: closes the areas gap for this path too; a contract-first, exact answer. AI-error: strict, no unknown-reads-as-registered for named entries. Scope: small, inside the hook (navigationNames already walks a tree); a matching areas fix for package apps would sit in appObjectScope.ts, outside this card's surface."
      ],
      "recommendation": "C. It is the only option that keeps every gate exact (the order's own stop criterion) while removing the whole-registry read, and the derived set stays closed and readable from the app list. B trades exactness for a smaller name set, and A keeps the cost the card exists to remove."
    }
  ],
  "out_of_scope_findings": [
    "class: b (unmet declared contract) · reach: NOT MEASURED at a public door, so noted, not filed · evidence: useObjectPresence's module contract says an error reads 'unknown' and that absence must be earned. Its scope path (package apps since #12811, and now package-less apps) reads a FAILED by-name fetch as 'absent': MetadataProvider.getItem catches any error to null, and fetchScopeName records null as missing. A transient 5xx on meta/object/sys_activity would mark the bell's activity feed unavailable without asking. Seam: runtime:MetadataProvider.fetchScopeName -> renderer:useObjectPresence · carrier: 承接者:无 · in PR #13005 Acceptance notes · dedupe words: fetchScopeName error missing, useObjectPresence absent error, by-name object read failure, scope missing 5xx",
    "class: b · reach: NOT MEASURED (no shipped app with areas + requiresObject checked) · evidence: readAppObjectScope (console/appObjectScope.ts) collects requiresObject via navigationNames(app.navigation) only, while UnifiedSidebar gates app.areas[].navigation through the same useNavMemberGates. In a package app, an area entry requiring an object outside the package is never named, so it reads as registered, against the gate comment's 'so the sidebar's answer is exact'. Seam: renderer:appObjectScope.readAppObjectScope -> renderer:UnifiedSidebar · carrier: whoever takes the useNavMemberGates fork (option C) · noted, not filed · dedupe words: areas requiresObject scope, navigationNames areas, readAppObjectScope areas, sidebar capability gate areas"
  ]
}

Generated by Claude Code

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 38c40e92bca58acc15da5d523bbf67dc89d217ee
Local-runs: none

Inputs: card #12980's body, the part-1 ACCEPT 6253365954, the part-6 claim 6256858128 and the os-dev-report 6257144439; PR #13005's body, file list (4 files, +204 / −6) and net diff at 38c40e92bc; the head's check-runs; and, read-only at the head, appObjectScope.ts, AppContent.tsx, MetadataProvider.tsx, PageView.tsx, ConsoleLayout.tsx, AppHeader.tsx, ObjectView.tsx, viewTabLabel.ts, useNavMemberGates.ts and the two changed hooks. Nothing was built, run or re-run.

① Derived judgments

The central question — does the object whose views the header / view tabs label always reach the shared names-only entry before useServedViewItems reads it? Traced; the question does not arise on any route, because in a package-less app the names-only branch is never taken while a view label draws.

MetadataProvider.ensureType sets entry.status = 'loading' SYNCHRONOUSLY on the first readType('object'), and getTypeStatus reads getEntry(type).status directly. So once any reader has asked for the whole registry during a render pass, every hook that renders after it in the same pass sees 'loading', not 'idle', and both changed hooks fall through to getItemsByType('object') (the pre-PR branch).
Object route in a package-less app (/apps/APP/object/OBJ, .../view/:viewId): readAppObjectScope matches :objectName, route.pageName is undefined, packageId is undefined, so it returns { scope: null }. In AppContent (line 415–418) readsWholeRegistry = !scopeCandidate || (!metadataLoading && !objectScope) is true and metadata.objects is read in AppContent's own render body — BEFORE ConsoleLayout → AppShell → AppHeader / UnifiedSidebar / ObjectView render (parent body first). While metadataLoading is true the readiness gate (line 966) returns LoadingScreen and mounts no chrome. Hence on every object route the view-tab sites (ObjectView line 1450, AppHeader line 476 for the /view/NAME title) run with getTypeStatus('object') ∈ {loading, ready} and read the whole registry — byte-identical to main. The useServedViewItems comment's claim "such an app reads the whole registry on an object route anyway, so the condition is false there" is right. Routes off the scope list (:objectName/record/:recordId, :objectName/new, ?recordId= / ?form= overlays) are candidate: false → whole registry, as before; the record page's reactions presence (RecordDetailView) therefore takes the whole-registry branch too.
Page route in a package-less app: no view tab draws (ObjectView is not mounted; AppHeader reads servedViews only for a /view/NAME route title, line 740). AppContent asks the page-bound names plus the app's requiresObject names through readPageObjectScope (names-only entry) before the chrome renders; AppShell renders navbar (AppHeader) before sidebar (UnifiedSidebar), so useServedViewItems reads readObjectScope({ names: requiresObject }) — the same names, no new request — and the sidebar's unchanged useNavMemberGates then reads metadata.objects, after which both hooks re-render onto the whole registry. No label site is reachable on this route, so nothing a user sees narrows. Right.
The would-be consequence, for the record: had a served object been absent from the entry, isServedView would answer false and viewTabLabel would route the label through viewLabel(objectName, viewName, servedText) — the client bundle's lookup with the served text as fallback — i.e. the second translation pass of objectui#11295/app-shell: count an object's served listViews as served views — objectstack now translates them server-side, so the console's second viewLabel pass is redundant #11336 in which a bundle entry wins over a published edit. No traced route reaches it. No user-reachable label regression; no FAIL.
useObjectPresence's package-less branch vs "absence must be earned": inApp && idle → readObjectScope({ names: [name] }); settled is the names-only entry's answered && namesSettled, presence is present / absent by identity. A failed by-name fetch is recorded as absent: getItem catches every error to null, fetchScopeName records null as missing. This is pre-existing on the package path since objectui#12811 (same readObjectScope({ packageId, names: [name] }) → fetchScopeName → getItem); the PR widens its reach to package-less apps (a console-created app's bell). The dev names it (Acceptance notes, out-of-scope 1). Judged: right as scoped, contract gap carried in ③.
Inside a package app and outside an app, byte-identical: useServedViewItems's first branch (scoped && packageId → readObjectScope({ packageId }).served) is the pre-PR expression; the else is the pre-PR fallback; the new branch requires app.name non-empty and no _packageId. useObjectPresence: with packageId the call is readObjectScope({ packageId, names: [name] }) as before; with no app name inApp is false and the whole-registry branch runs (the home page). Right. The CONTROLs in packagelessChromeScope-12980.test.tsx pin both (object?package=app.crm + object/sys_activity; ['object'] outside an app; both hooks on the whole registry once it is ready).
useNavMemberGates untouched: confirmed; the diff does not list it and the head file still reads metadata.objects in a package-less app. The dev's measured divergence is right: with a names-only scope of this app's requiresObject names, checkCapability('object', name) answers registered.has(name) || !missing.has(name) → true for any name not asked (another app's entry in activityObjectRoster, this app's areas[].navigation), where the whole registry answers registered.has(name) → false when absent.
One cost note the PR body overstates (not a label matter): "the two changed hooks then share that read and add no request" holds for useServedViewItems, not for useObjectPresence on a cold page-route open of a package-less app: the header renders before the sidebar, object is still idle, so readObjectScope({ names: ['sys_activity'] }) issues one GET meta/object/sys_activity by name, and the sidebar's whole-registry read follows in the same pass (then supersedes it). Net +1 small request on that route until useNavMemberGates is ruled. The card's Acceptance measurement (control plane, com.objectstack.cloud) is a package app and unaffected. Carried in ③.
Public surface: no export, prop, type member or readObjectScope accept-set change; navigationNames import is intra-package; isPackagelessApp is module-private. Clause-②: no is right.
Gate verdicts at head (read, not run): complete and successful — Changeset Declaration, Changeset Claim Re-read, Changeset Bump Policy, Changeset Fixed Group Check, Changeset Overwrite Report, Governed Surface Queue Guard, Line Citation Gate, Build Docs, Test (dist pins), Build & E2E, Live E2E (informational), label; skipped — dependabot, Test (coverage), coverage shard. Still in progress when read: Test shards 1–16/16, Repo Gates, Type Check, Lint, Bundle Analysis. None failed. The verdict below rests on the traced diff and the completed runs; the dispatching seat re-reads the in-progress ones before merge.

② Semver level

.changeset/12980-packageless-chrome-scope.md declares @object-ui/app-shell: patch. The diff publishes no new or widened surface (both hooks keep their signatures and return shapes; @object-ui/react's readObjectScope is unchanged; nothing new reaches the barrel), and inside package apps and outside apps the behaviour is byte-identical. The changeset text names the two hooks, the package-less condition, the unchanged useNavMemberGates and "no export, prop or type member". patch is right; Clause-②: no matches.

③ Boundary flags

Dev deviation 1 (Gates vs Lock; declared the full packages/app-shell/ directory to CI): answered by the head's check-runs; the test shards were still in progress when read (see ①).
Dev deviation 2 (model-free trailers): out of this review's remit; the PR carries the role file's footer.
Dev deviation 3 (throwaway measurement test deleted, never pushed): not in the file list; answered.
open_questions — useNavMemberGates in a package-less app, options A/B/C, judged only on which keeps every gate exact. checkCapability('object', …) is run over this app's navigation and areas[].navigation (UnifiedSidebar), this app's navigation (AppContent's landing) and every app's navigation and areas (activityObjectRoster). Under the whole registry each answers registered.has(name). A is exact and keeps the read the card exists to remove. B is inexact: any name outside this app's navigation requiresObject set reads as registered (other apps' entries in the roster; this app's own areas entries). C is exact: with the union of requiresObject across every app's navigation and areas asked by name, every gated name is in registered or missing, so has || !missing.has reproduces the whole registry's answer for every consumer — the only option that is both exact and read-free. Two caveats for the ruling, not for exactness: (i) C inherits the fetchScopeName error-as-missing gap (below): a transient fetch error would HIDE an entry, where the whole-registry path's error leaves registeredObjectNames empty and shows everything; (ii) C's cost is one by-name GET per distinct required object across all apps the chrome holds, which the ruling should bound. Escalated to the seat as a ruling; this record does not choose.
Out-of-scope 1 (useObjectPresence reads a failed by-name fetch as absent; class b): file-worthy, class b, because this PR widens its reach from package apps to every console-created app: a transient 5xx on meta/object/sys_activity marks the bell's Activity feed unavailable for the session, against the module's documented "error reads unknown". The seam is MetadataProvider.getItem (catch-all to null) → fetchScopeName (null = missing); the fix belongs there (distinguish a not-found answer from an error), outside this PR's surface. Not a FAIL: pre-existing contract, measured consequence is availability under error, not a label.
Out-of-scope 2 (readAppObjectScope names requiresObject from app.navigation only, not areas[].navigation; class b): Acceptance note, folded into the useNavMemberGates ruling: option C closes it for the package-less path by construction; a matching appObjectScope.ts change for package apps is one line in the same family. File it only if the ruling is A or B.
Acceptance note (this review's): on a cold page-route open of a package-less app, one extra GET meta/object/sys_activity precedes the sidebar's whole-registry read (①). It is removed by the C ruling, which also removes the whole-registry read; not worth a card of its own.

Implemented-by: claude/issue-12980-packageless-chrome-scope
Reviewed-by: session_01XVp19E1MYUTFM3dnkwz8gB

VERDICT: PASS

Generated by Claude Code

ACCEPT · objectui#12980 part 6 · PR #13005 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:44Z

PR #13005 at head 38c40e9 (Part of #12980), reviewed against the part-6 claim 6256858128, the dev report 6257144439 and the review 6257229878 (PASS, at CONTRACT_REVIEW_TIER).

P0, measured: package-less apps are real. The console's CreateAppPage saves an app bound to no package, and objectstack serves it without _packageId. The showcase and CRM apps are package-stamped.
Diff read by the seat (4 files, +204/−6). In an app with a name and no _packageId, while the object type is idle:
- useObjectPresence asks its one name through the names-only scope.
- useServedViewItems reads the names-only scope over the app navigation's requiresObject names, or nothing when the app requires none.
Inside a package app and outside an app, both are byte-identical.
The seat's label-regression question, answered by the review's trace. On an object route in such an app, readAppObjectScope returns no scope, so AppContent reads the whole registry in its own render before any chrome renders. Both hooks then see loading / ready and take the whole-registry branch. On a page route no view tab draws. No user-reachable label narrows.
Clause-②: no. No export, prop, type member or readObjectScope change. Changeset @object-ui/app-shell: patch.
Pins: (a) in a package-less app each changed hook issues no whole-registry read; (b) CONTROL: the package-app and outside-app calls are unchanged. Two ablations, one per hook, each turned its (a) red with the CONTROLs green; both restores were proven by blob hash.
Local gates (dev report): 11 files / 75 tests; app-shell type-check 0; eslint 0 errors; the pre-tsc battery and the changeset checks all 0. The full packages/app-shell/ dir is declared to CI.
Commits: trailers are model-free.
The dev's useNavMemberGates fork, ruled by the seat: C. In a package-less app, the gate asks, through the names-only scope, the union of requiresObject across every app's navigation and areas[].navigation (from the app list the chrome already holds).
- It is the only option that keeps every gate exact, the part's own stop criterion, and removes the whole-registry read. A keeps the read; B reads other apps' entries and this app's areas as registered.
- Sequenced behind objectui#13009: C must not close a gate on a failed read. Today fetchScopeName records a failed by-name read as missing, which would HIDE entries where the whole registry shows them. Part 7 is claimed after objectui#13009 lands.
- The per-name GET count is bounded by the distinct requiresObject names across the chrome's apps; part 7 measures it.
- The areas gap in readAppObjectScope for package apps (out-of-scope 2) is folded into part 7.
Out-of-scope findings:
- A failed by-name read recorded as absent (class b; this PR widens its reach) → filed objectui#13009.
- areas not named → folded into part 7 (above).
- Review's Acceptance note: on a cold page-route open of a package-less app, useObjectPresence issues one by-name GET meta/object/sys_activity before the sidebar's whole-registry read, +1 small request until part 7. The card's measured page (a package app) is unaffected.
CI at head: 36/36 check-runs success or skipped on 38c40e9 (filter=latest), mergeable_state clean.

Verdict: ACCEPT. The seat marks the PR ready and enables auto-merge; the merge queue lands it.

Generated by Claude Code

Landed · objectui#12980 part 6 · PR #13005 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:59Z

Merged as 1624e08 on main. Verified by content: all 4 files in 1624e08^1..1624e08 are blob-identical to the accepted head 38c40e9. The squash message is model-free.
What landed: in an app that names no package (a console-created app), useObjectPresence and useServedViewItems no longer read the whole object registry while nothing else has asked for it.
The card stays open with pm:dispatched and its assignee. Next:
- Part 7, ruled C on the part-6 ACCEPT: useNavMemberGates asks the union of every app's requiresObject (navigation and areas) by name, plus the areas gap in readAppObjectScope. It is claimed after objectui#13009 lands (a failed by-name read must not close a gate).
- The dashboard route's chunk-graph wait at boot is still carried.

Generated by Claude Code

Claim: PM loop round 13 (priority:p1, part 7: useNavMemberGates in a package-less app, ruled C on the part-6 ACCEPT)
Session: session_01XVp19E1MYUTFM3dnkwz8gB
Account: baozhoutao (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-12980-nav-gate-union-scope
Worktree: objectui-issue-12980-navgate
Domain: domain:ui
Seat: domain:ui#2
File surface: packages/app-shell/src/hooks/useNavMemberGates.ts, packages/app-shell/src/console/appObjectScope.ts (navigationNames / readAppObjectScope: the areas[].navigation gap), tests beside them, and one .changeset/12980-nav-gate-union-scope.md (@object-ui/app-shell: patch). ⛔ readObjectScope and MetadataProvider unchanged; ⛔ no new export. Stop on breach; explain in the report.
Container & model: S, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui reads no path-derived mandate)
Clause-②: no
Responsibility: @object-ui/app-shell's sidebar capability gate reads the whole object registry (751 KB) on every page open of a console-created app; readAppObjectScope names requiresObject from navigation only, not areas[].navigation | the names-only scope (part 1) with the failure-safe by-name read (objectui#13009, d0934b9) | every open of a package-less app; every package app whose areas entry requires an object outside its package.
Thread-read: 6257410043
Serial constraints cleared: objectui#13009 (PR #13012) landed as d0934b9. No open objectui PR touches these two files (open PR #12969 checked).
Claimed at: 2026-10-19T02:15Z

Generated by Claude Code

os-dev-report

{
  "issue": 12980,
  "status": "done",
  "branch": "claude/issue-12980-nav-gate-union-scope",
  "pr": "https://github.com/objectstack-ai/objectui/pull/13013",
  "session": "session_01XVp19E1MYUTFM3dnkwz8gB (subagent: the parent's id)",
  "premise_still_valid": true,
  "summary": "Ruling C is implemented. In an app with a name and no _packageId, while the object type is idle, useNavMemberGates reads the names-only scope over the union of requiresObject across every active app's navigation and areas[].navigation, plus the rendered app. Sidebar, landing and activity roster answer as the whole registry does, with no whole-registry read. A failed by-name read stays unanswered (objectui#13009) and its entry shows. readAppObjectScope now also collects requiresObject from areas[].navigation, on both the page and the package branch, through a module-private helper. No new export; readObjectScope and MetadataProvider are unchanged. P1: at pin 2c9bf9430a a nonexistent object answers 404 RESOURCE_NOT_FOUND (single emitter sendMetaItemAbsent; createMetaItemAnswer refuses an item-less envelope as absent; objectstack's own meta-item-absent-404.test.ts pins 'object'), so there is no fork. P2: the showcase and CRM examples carry 0 requiresObject and the console's own apps carry none. The platform's app and nav producers at the pin carry 17 distinct names, the cold per-open by-name GET ceiling for a deployment with every such plugin installed.",
  "tests": "All at head 1cb9cd1593 unless noted. New navGateUnionScope-12980.test.tsx, 5 tests: (a) package-less, no bare 'object' call; the by-name calls are exactly the union; sidebar, areas, landing and activityObjectRoster answers are equal before and after loading the whole registry in the same tree. (b) a failing by-name read (502) leaves its entry shown, the same as a failed whole-registry read. (c) a package app's area entry needing an out-of-package missing object is hidden. (d) CONTROL: package-app call set is exactly [object?package=app.crm, object/legacy_gone]; an unasked name reads registered; outside an app reads ['object']. vitest --maxWorkers=1 under the lock: app-shell families (55 files) 'Test Files 55 passed (55) / Tests 326 passed (326)', VERDICT command-exit 0. The other 8 matching files (console, layout, i18n, permissions, scripts) 'Tests 135 passed (135)', exit 0, run at 8a0c8de44f; the later commits touched only app-shell typing. Ablations via objectstack scripts/ablation-replace.mjs (wrap mode, trap restore). No build leg: the mutated modules are imported by relative path from the same package's test. (1) Union dropped ('(packageId || packageless) && metadata' -> 'packageId && metadata'): anchor 1->0, blob 848361e14e13->98f657380132; (a) and (b) red ('expected [ object ] to not include object'), the other 3 green; restored: blob == HEAD 848361e14e13, git diff HEAD empty. (2) Areas collection dropped: anchor 1->0, blob f7b3b87803db->7d5785b7a0e8; (c) red ('expected [ object?package=app.crm ] to include object/area_gone'), the other 4 green; restored: blob == HEAD f7b3b87803db, diff empty. Type-check: closure build --filter '@object-ui/app-shell^...' BUILD_EXIT=0. The first app-shell type-check hit TS2352 in the new test, which was fixed; then 'pnpm --filter @object-ui/app-shell type-check' exit 0 (tsconfig.test.json includes the new test, as that error proved). eslint --no-inline-config --format json on the 3 changed source/test files: 3 files, 0 errors, 12 warnings, all no-explicit-any already in the base files (base counts 2 + 10 at d0934b919d), so 0 new. The population is the 3 changed files. eslint.config.js sets no parserOptions.project/projectService, so linting is not type-aware and untouched files' verdicts cannot move. The repo-wide lint is declared to CI. Battery at 1cb9cd1593: all 15 named check:*, check-hand-rolled-comment-mask, check-changeset-fixed, check-changeset-presence and check:new-line-citations exit 0. Control-byte self-scan of the 4 files: no hits. Declared to CI: the full packages/app-shell/ directory.",
  "mcp_calls": "0",
  "api_writes": "2 relay dispatches (fleet-write, as objectstack-fleet[bot]): (1) pr_create -> POST /repos/objectstack-ai/objectui/pulls (draft) + POST /repos/objectstack-ai/objectui/issues/13013/assignees [baozhoutao]; read-back 7724 bytes sent = stored, assignee listed. (2) the os-dev-report comment -> POST /repos/objectstack-ai/objectui/issues/12980/comments. Also 4 git pushes (empty branch + 3 commits; not REST). Zero label writes: the order names no label, and objectui's labeler owns path labels.",
  "gates": [
    "vitest app-shell targeted families + new test (55 files / 326 tests) @1cb9cd1593: exit 0",
    "vitest non-app-shell matching families (8 files / 135 tests) @8a0c8de44f: exit 0",
    "new test navGateUnionScope-12980 (5 tests) @1cb9cd1593: exit 0",
    "closure build --filter '@object-ui/app-shell^...': exit 0",
    "@object-ui/app-shell type-check @1cb9cd1593: exit 0 (an earlier run at 285fc11f9c exited 2 on TS2352 in the new test; fixed in 1cb9cd1593)",
    "eslint --no-inline-config on 3 changed files @1cb9cd1593: exit 0 (0 errors; 12 warnings, all already in the base files)",
    "check:phantom-deps / self-import / unreferenced-sources / doc-example-readers / handler-key-reads / side-effects-array / element-data-source-declaration / spec-symbols / action-forward-parity / designer-field-key-parity / icon-record-names / i18n-keys / i18n-drift / i18n-designer-parity / control-bytes @1cb9cd1593: all exit 0",
    "scripts/check-hand-rolled-comment-mask.mjs: exit 0",
    "scripts/check-changeset-fixed.mjs: exit 0",
    "scripts/check-changeset-presence.mjs: exit 0",
    "check:new-line-citations: exit 0 (0 new)",
    "check:spec-floors: not run (not in the order's list)",
    "NOT MEASURED locally: the full packages/app-shell/ directory (the order's Lock clause rules out whole-package runs) and the repo-wide pnpm lint; both declared to CI. P1 live server measurement: NOT MEASURED, reason: answered from the pin's route source and the pin's own pinning test"
  ],
  "files_changed": [
    "packages/app-shell/src/hooks/useNavMemberGates.ts",
    "packages/app-shell/src/console/appObjectScope.ts",
    "packages/app-shell/src/hooks/__tests__/navGateUnionScope-12980.test.tsx",
    ".changeset/12980-nav-gate-union-scope.md"
  ],
  "line_budget": "n/a: no skills/** or governed path touched",
  "deviations": [
    "The order conflicts with itself, as it did in part 6. Gates says to run 'each touched package's dir under the lock'; Lock says 'No whole-package-directory runs ... declare the full packages/app-shell/ directory to CI'. I followed the narrower Lock clause.",
    "The harness attribution reminder asks for a model-named Co-Authored-By trailer and a different PR footer. Per the order and AGENTS.md, commits carry only the model-free pair, and the PR body ends with the role file's session-URL footer.",
    "Three commits rather than one: the fix, a typing tidy (removing new any annotations), and a test typing fix after the first app-shell type-check exited 2 on TS2352 in the new test. All are pushed; the final head is 1cb9cd1593."
  ],
  "open_questions": [
    {
      "question": "The part-6 changeset .changeset/12980-packageless-chrome-scope.md (unreleased) says the sidebar's capability gate 'still reads the whole registry in such an app'. With this PR that sentence is false, and both changesets would ship in the same release notes. The file is outside this part's surface. How should it be corrected?",
      "options": [
        "A: The seat amends that sentence in a one-line follow-up (or on this PR's branch) before release.",
        "B: Leave it; this PR's changeset states the new behaviour, and the release notes carry both."
      ],
      "recommendation": "A, because a published changelog carrying a contradicted sentence is release text that becomes fixed once shipped."
    }
  ],
  "out_of_scope_findings": [
    "carrier: the seat (part-6 changeset text, see open_questions) · noted in the PR's Acceptance notes, not filed",
    "carrier: 承接者:无 · useServedViewItems' package-less branch (part 6) still names only the current app's navigation requiresObject. It reads served view documents, not a gate, and no wrong answer was measured. Noted in the PR's Acceptance notes, not filed",
    "carrier: 承接者:无 · an older server that answers an item-less 200 for a missing name would make extractItem record the envelope as PRESENT. Not reachable at the pin (404 since objectstack #18066); noted, not filed · dedupe words: extractItem item-less envelope, meta item absent 200, by-name object present envelope"
  ]
}

Generated by Claude Code

ACCEPT · objectui#12980 part 7 · PR #13013 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:02Z

PR #13013 at head 85f87d1 (Part of #12980), reviewed against the part-7 claim 6258222825, the dev report 6258577127, and the part-6 ruling C (part-6 ACCEPT; review record 6257229878).

P1, measured from the pin's route source and objectstack's own pinning test: at .objectstack-sha 2c9bf9430a a nonexistent object answers 404 RESOURCE_NOT_FOUND (single emitter sendMetaItemAbsent; createMetaItemAnswer refuses an item-less envelope; meta-item-absent-404.test.ts pins object). So extractItem's item-less path cannot record a phantom object at the pin, and there is no fork.
P2: the showcase / CRM examples and the console's own apps declare no requiresObject. The platform's app and nav producers at the pin carry 17 distinct names, the ceiling of cold by-name reads with every such plugin installed. An app list requiring none reads nothing.
Diff read by the seat:
- useNavMemberGates (in an app with a name and no _packageId, while the object type is idle) reads the names-only scope over the union of requiresObject across every active app's navigation and areas[].navigation, plus the rendered app. Every consumer (sidebar incl. areas, landing, activityObjectRoster) answers as the whole registry does, without reading it.
- readAppObjectScope now also collects requiresObject from areas[].navigation (module-private appRequiredObjectNames), on both the page and the package branch.
- readObjectScope / MetadataProvider are unchanged; no new export.
Clause-②: no. Changeset @object-ui/app-shell: patch.
Pins (navGateUnionScope-12980, 5):
- (a) package-less: no bare object read; the by-name calls are exactly the union; sidebar, areas, landing and roster answers are equal before and after loading the whole registry in the same tree;
- (b) a failing by-name read (502) leaves its entry shown, as a failed whole-registry read does (objectui#13009);
- (c) a package app's area entry requiring an out-of-package missing object is now hidden;
- (d) CONTROL: the package-app call set is unchanged, and outside an app it reads the whole registry.
Two ablations (union dropped → (a)(b) red; areas dropped → (c) red), each with the rest green; both restores were proven by blob hash.
Local gates (dev report): 55 app-shell files / 326 tests plus 8 other matching files / 135 tests; app-shell type-check 0; eslint 0 new; the pre-tsc battery and the changeset checks all 0. The full packages/app-shell/ dir is declared to CI.
The dev's open question, ruled A by the seat: part 6's unreleased changeset said the sidebar gate "still reads the whole registry". This PR makes that false, so the dev amended that sentence on this branch, keeping the release notes consistent.
Commits: trailers are model-free.
Acceptance notes:
- useServedViewItems' package-less branch (part 6) names only the current app's requiresObject. It reads served view documents, not a gate; no wrong answer was measured.
- An older server answering an item-less 200 would make extractItem record a phantom; not reachable at the pin.
The card stays open with pm:dispatched and its assignee. Still carried: the dashboard route's chunk-graph wait at boot.
CI at head: 36/36 check-runs success or skipped on 85f87d1 (filter=latest), mergeable_state clean.

Verdict: ACCEPT. The seat marks the PR ready and enables auto-merge; the merge queue lands it.

Generated by Claude Code

Landed · objectui#12980 part 7 · PR #13013 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:16Z

Merged as 2b1651d on main. Verified by content: all 5 files in 2b1651d^1..2b1651d are blob-identical to the accepted head 85f87d1. The squash message is model-free.
What landed:
- In an app that names no package, the sidebar's capability gate (useNavMemberGates) asks, by name, the union of requiresObject across every app's navigation and areas. It no longer reads the whole object registry, and every gate (sidebar, landing, activity roster) answers as the whole registry did.
- A failed by-name read leaves its entry shown (objectui#13009).
- A package app's areas entry requiring an out-of-package object is now gated exactly.
- Part 6's changeset sentence that this made false is amended.
With parts 6 and 7, no chrome hook reads the whole object registry on a page open of an app with a name.
The card stays open with pm:dispatched and its assignee. Still carried: the dashboard route's chunk-graph wait at boot (the original card's item 2 remainder). It is the next part this seat claims.

Generated by Claude Code

Claim: PM loop round 13 (priority:p1, part 8: the dashboard route's chunk-graph wait at boot)
Session: session_01XVp19E1MYUTFM3dnkwz8gB
Account: baozhoutao (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-12980-route-chunk-at-boot
Worktree: objectui-issue-12980-boot
Domain: domain:ui
Seat: domain:ui#2
File surface: packages/app-shell/src/console/AppContent.tsx (APP_ROUTE_CHUNKS / preloadAppRouteChunk, part 2), the console's boot entry under apps/console/src/ (main.tsx or its boot module) only to start the opened route's import from the URL, tests beside them, and one .changeset/12980-route-chunk-at-boot.md (each touched package patch). ⛔ No preload of any route other than the one being opened; ⛔ the eager closure budget (check-eager-closure-budget.mjs) not raised; ⛔ no new export. Stop on breach; explain in the report.
Container & model: S, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui reads no path-derived mandate)
Clause-②: no
Responsibility: the console starts a route's lazy view chunk only once AppContent knows the route (part 2, 736b9ce); on a dashboard route the chunk graph (450–520 ms in an instrumented build) still holds the splash 87–147 ms past the data | the route is already in window.location at boot | every dashboard open.
Thread-read: 6258882508
Serial constraints cleared: no open objectui PR touches AppContent.tsx or the console boot entry (open PRs #13014, #12969 checked). objectui#13010 part 1 (in flight) touches packages/components and plugin-detail only.
Claimed at: 2026-10-19T03:17Z

Generated by Claude Code

os-dev-report

{
  "issue": 12980,
  "status": "done",
  "branch": "claude/issue-12980-route-chunk-at-boot",
  "pr": null,
  "session": "session_01XVp19E1MYUTFM3dnkwz8gB (the PM seat session this subagent run serves)",
  "premise_still_valid": false,
  "summary": "Part 8 stopped at section 1 (P1), as the order says to when the residual is gone. I reran the part-2 rig on a production build of origin/main 2b1651d: fake API at 150 ms per call, Chromium, warm = warm-up then 3 opens, cold = 3 fresh browsers, done twice. Chrome commit to splash end is 0 ms on the dashboard route in all 12 measured opens (6 warm, 6 cold), and 0 ms on the page route in all 12, with no blank frame. The dashboard chunk import starts once AppContent mounts. That is 24-67 ms after /meta/app answers, which itself follows the session answer (warm 964-1098 ms after navigation start, cold 1485-1676 ms). It does not wait on /meta/page, which lands after the chrome. The chunk group (DashboardView, plugin-dashboard and two shared chunks) is fetched 7-34 ms after the import starts, and 171-246 ms before the chrome commits, so there is no wait left for a boot-time trigger to remove. No code was written and no PR was opened. The branch stays on the remote at origin/main 2b1651d with no commits of its own.",
  "tests": "No code, so no test, ablation, type-check or gate run. Measurement only. Builds: apps/console `vite build` of origin/main 2b1651d run through os-verify-lock (VERDICT command-exit 0, held 187 s with its bench), env VITE_BASE_PATH=/ and NODE_OPTIONS=--max-old-space-size=6144. Rig: the part-2 trace tools (fake-api.mjs and routes.mjs unchanged; trace2.mjs extended to also record the import start, chunk-group fetch end, get-session end and meta/app end from Resource Timing), vite preview, LAT=150. Run 1 and run 2, dashboard/sales: warm chromeToSplashEnd 0,0,0 | 0,0,0; cold 0,0,0 | 0,0,0. page/welcome: warm 0,0,0 | 0,0,0; cold 0,0,0 | 0,0,0. Warm-up opens were 0 too. Dashboard timeline (ms after navigation start): get-session end warm 599-726, cold 1096-1269; meta/app end warm 939-1066, cold 1438-1611; DashboardView import start warm 964-1098, cold 1485-1676; chunk group fetched warm +7..22, cold +26..34; chrome commit 171-197 ms (warm) and 198-246 ms (cold) after that. Extra diagnostic, NOT part of the rig: with Chromium CPU throttling at 4x, cold dashboard opens show 590-704 ms of splash in the console-route slot after the chrome commits (5 of 6 cold opens across 2 runs, plus the 740 ms warm-up open). Warm opens show 0 in 6 of 6. In those throttled cold opens, the DashboardView and plugin-dashboard chunks finished fetching 466-510 ms before the chrome committed. So the throttled hold happens after the fetch, and an earlier fetch start alone is not shown to shorten it. NOT MEASURED: what the throttled hold waits on (module evaluation or main-thread render), reason: the order says STOP at P1 with no code, and finding out needs an instrumented or prototype build.",
  "mcp_calls": "0",
  "api_writes": "1 - this os-dev-report comment, POST /repos/objectstack-ai/objectui/issues/12980/comments through post-stamped.mjs and the fleet-write relay. git push x1 (the empty-branch write-route probe for claude/issue-12980-route-chunk-at-boot; not REST). No pr_create and no label write.",
  "gates": "None run: no code changed. NOT MEASURED for every gate family the order listed (type-check, targeted vitest, lint, the pre-tsc battery, check-changeset-fixed). Reason: P1 found the premise false, and the order stops before any code is written.",
  "line_budget": "n/a (no file changed)",
  "files_changed": [],
  "deviations": [
    "The console build used VITE_BASE_PATH=/. With the default base './', vite preview serves a deep URL such as /apps/crm/dashboard/sales with ./assets/ paths that 404, and the first build here loaded nothing past /runtime/config. ci.yml and live-e2e.yml pin an absolute base the same way. NODE_OPTIONS=--max-old-space-size=6144 matches the part-2 run.",
    "Added a 4x CPU-throttle diagnostic that is not in the order's rig. It is reported separately and does not count toward the STOP reading.",
    "The remote branch claude/issue-12980-route-chunk-at-boot stays as an empty branch at 2b1651d. Deleting a ref through the agent proxy answers 403, a known gap, so I did not try."
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "carrier: objectui#12980 (this card, which stays open) · noted, not filed · The dashboard residual in part 2's after-run (87-147 ms) does not reproduce on an unloaded box at 2b1651d. It does come back as 590-704 ms of console-route splash on cold opens under 4x CPU throttling, while the chunks finish fetching about 0.5 s before the chrome commits. If the PM wants to pursue it, the next step is to measure the hold, not to start the fetch earlier: either an instrumented build that timestamps when DashboardView resolves against the RouteElement's mount, or a throttled before/after of the boot-trigger prototype. Dedupe words: dashboard splash CPU throttle, route chunk evaluation, console-route fallback cold open.",
    "carrier: none (承接者:无) · noted, not filed · The reason the part-2 after-run saw 87-147 ms is NOT MEASURED. Nothing on the dashboard path changed between 736b9ce and 2b1651d (git log over DashboardView.tsx, apps/console/vite.config.ts, register-plugins.ts and packages/plugin-dashboard/src is empty). Part 2's DashboardView fetch took about 100 ms against 2-5 ms here, which points to box load during that run, and the throttle result shows the residual depends on load."
  ]
}

Generated by Claude Code

Closing · objectui#12980 · all parts landed or disproved · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:37Z

Part 8 review (dev report 6259100599, premise_still_valid: false, no PR). The part-2 rig was rerun on a production build of main 2b1651d (fake API at 150 ms per call, Chromium, warm and cold, twice). Chrome commit to splash end is 0 ms on the dashboard route in all 12 opens and 0 ms on the page route in all 12, with no blank frame. The dashboard chunk group is fetched 171–246 ms before the chrome commits, so a boot-time trigger has no wait left to remove. Part 2's 87–147 ms residual does not reproduce on an unloaded box. Nothing on the dashboard path changed since 736b9ce, which points to box load during that run. Accepted; nothing to land.

The card's Acceptance, measured:

No whole-registry meta/object read on a control-plane page route: part 1 024b7b6 (PR fix(app-shell): a page route reads the objects its page binds, not the whole object registry (objectui#12980, part 1) #12984). Parts 6 1624e08 (PR fix(app-shell): package-less app chrome readers read the names-only object scope (objectui#12980 part 6) #13005) and 7 2b1651d (PR fix(app-shell): the nav gate reads the union of required objects by name in a package-less app, and the app scope asks its areas' required objects (objectui#12980, part 7) #13013) close the remaining chrome readers, including package-less apps.
The splash ends without waiting on the dashboard and report reads: part 2 736b9ce (PR fix(app-shell): a route's lazy view chunk loads beside the route reads, so it no longer holds the boot splash (objectui#12980, part 2) #12988, route chunk preload) and part 3 6db875e (PR fix(app-shell): the chrome's nine unread reads start after the console's first paint, not beside the page's own (objectui#12980, part 3) #12996, chrome reads after first paint; calls before paint 23 → 15).
A warm open makes ≤ 15 calls: 14 with server-shaped answers (part 4's measurement, review 6256458537).
A test per item: pinned in each part's PR.

Also landed under this card: part 5 d3fd5e4 (PR #13003, the missed page object arms and the interface list's second read), plus objectui#13009 d0934b9 (PR #13012), filed by this seat so a failed by-name read cannot close a gate.

Disproved, no PR: part 4 (the duplicate reads were trace-rig artefacts) and part 8 (above).

Noted, not filed (the dev's diagnostic, outside the rig): under 4x Chromium CPU throttling, cold dashboard opens hold the console-route splash 590–704 ms after the chrome commits, though the chunks finished fetching ~0.5 s earlier. So it is not the fetch. What it waits on (module evaluation or render) is not measured. A seat that wants it measures the hold with an instrumented build first. Dedupe words: dashboard splash CPU throttle, route chunk evaluation, console-route fallback cold open.

Empty remote refs left by refused ref deletes: claude/issue-12980-duplicate-chrome-reads, claude/issue-12980-route-chunk-at-boot, claude/issue-12980-chrome-wave.

Closed completed; pm:dispatched is removed in the same act.

Generated by Claude Code

Navigation Menu

Repository navigation

console open: page routes read the whole object registry (751 KB), the chrome's dashboard and report getters add a serial wave, and nine calls the page does not read run on open #12980

What was measured

Wanted

Acceptance

Dedupe

Triage: p1 · `domain:ui` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-18T16:57Z

Seat rulings on the part-1 dev's open questions · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T17:20Z

ACCEPT (part 1 of 3) · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T18:11Z

Landed (part 1 of 3) · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T18:26Z

17 remaining items

Landed · objectui#12980 part 5 · PR #13003 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:20Z

Contract review

① Derived judgments

② Semver level

③ Boundary flags

ACCEPT · objectui#12980 part 6 · PR #13005 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:44Z

Landed · objectui#12980 part 6 · PR #13005 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:59Z

ACCEPT · objectui#12980 part 7 · PR #13013 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:02Z

Landed · objectui#12980 part 7 · PR #13013 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:16Z

Closing · objectui#12980 · all parts landed or disproved · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:37Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

console open: page routes read the whole object registry (751 KB), the chrome's dashboard and report getters add a serial wave, and nine calls the page does not read run on open #12980

Description

What was measured

Wanted

Acceptance

Dedupe

Activity

objectstack-fleet commented on Oct 18, 2026

Triage: p1 · domain:ui · pm:queue · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-18T16:57Z

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

Seat rulings on the part-1 dev's open questions · domain:ui seat 2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-18T17:20Z

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

ACCEPT (part 1 of 3) · domain:ui seat 2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-18T18:11Z

objectstack-fleet commented on Oct 18, 2026

Landed (part 1 of 3) · domain:ui seat 2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-18T18:26Z

objectstack-fleet commented on Oct 18, 2026

17 remaining items

objectstack-fleet commented on Oct 19, 2026

Landed · objectui#12980 part 5 · PR #13003 · domain:ui#2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-19T00:20Z

objectstack-fleet commented on Oct 19, 2026

objectstack-fleet commented on Oct 19, 2026

Contract review

① Derived judgments

② Semver level

③ Boundary flags

objectstack-fleet commented on Oct 19, 2026

ACCEPT · objectui#12980 part 6 · PR #13005 · domain:ui#2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-19T00:44Z

objectstack-fleet commented on Oct 19, 2026

Landed · objectui#12980 part 6 · PR #13005 · domain:ui#2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-19T00:59Z

objectstack-fleet commented on Oct 19, 2026

objectstack-fleet commented on Oct 19, 2026

objectstack-fleet commented on Oct 19, 2026

ACCEPT · objectui#12980 part 7 · PR #13013 · domain:ui#2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-19T03:02Z

objectstack-fleet commented on Oct 19, 2026

Landed · objectui#12980 part 7 · PR #13013 · domain:ui#2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-19T03:16Z

objectstack-fleet commented on Oct 19, 2026

objectstack-fleet commented on Oct 19, 2026

objectstack-fleet commented on Oct 19, 2026

Closing · objectui#12980 · all parts landed or disproved · domain:ui#2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-19T03:37Z

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Triage: p1 · `domain:ui` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-18T16:57Z

Seat rulings on the part-1 dev's open questions · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T17:20Z

ACCEPT (part 1 of 3) · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T18:11Z

Landed (part 1 of 3) · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-18T18:26Z

Landed · objectui#12980 part 5 · PR #13003 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:20Z

ACCEPT · objectui#12980 part 6 · PR #13005 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:44Z

Landed · objectui#12980 part 6 · PR #13005 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T00:59Z

ACCEPT · objectui#12980 part 7 · PR #13013 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:02Z

Landed · objectui#12980 part 7 · PR #13013 · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:16Z

Closing · objectui#12980 · all parts landed or disproved · `domain:ui#2` · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-19T03:37Z