fix(plugin-form,components): master-detail edit lines and element:record_picker options re-read on the data-invalidation bus (objectui#10853)

[objectstack-fleet]

Fixes #10853
Clause-②: no — two readers subscribe to the existing invalidation bus; no declared key, export or accept set moves.

This is the fifth batch of bus readers under decision B on objectui#10519 ("readers first"). After PR objectui#10777 removes PageView's remount, these two blocks would otherwise keep showing data from before a page action. objectui#10519 remains open.

Dispatched by the domain:ui seat 4 PM (claim comment 5860833997, triage 5860720827). Session: https://claude.ai/code/session_01MCg3z5cGrV51xEddpGcxEY. In mode:subagent that id is also this run's. Branch claude/issue-10853-master-detail-picker-bus-readers was cut from origin/main a7fd877 (BASE). Head is a13048c, after one patch round on the contract review of ac526b1 (see Patch round below).

What changed

Each read now takes a nonce from the data-invalidation bus for the object it queries. This follows the objectui#10623 / objectui#10778 shape: the same bus, the same matcher (dataChangeMatches), no second channel and no remount.

• packages/plugin-form/src/MasterDetailForm.tsx, edit mode. Each detail collection's find of its child object now re-reads when the bus reports a change to that child object, or '*'. The re-read happens in place, and the rows and the save baseline move together.

  • The collections are authored, so their number is not fixed, and one useDataInvalidation call per object cannot be made. A local useObjectsInvalidation(names) subscribes through subscribeDataChanges, the listener set useDataInvalidation itself uses. It tests each change with dataChangeMatches at object level and keys on the set's content.
  • Nothing new is exported. The claim's Clause-②: no holds.
  • The load now runs one collection at a time: a per-collection read number, plus a generation the load effect's teardown bumps. So the load on mount, a record change, a bus re-read and a replay each commit only if they are that collection's latest read.
  • A collection only re-reads for its own child object. The header re-reads through its own ObjectForm (objectui#10572), which is unchanged.
  • The row editor ("Open row") holds its collection's re-read while it is open (patch round).

• packages/components/src/renderers/basic/record-picker.tsx. The option read's effect now names useDataInvalidation(object), subscribed only when the adapter reads and the object has resolved. A read of exactly the query whose options are already on screen (same adapter, object, filter, sort and limit) runs in place:

  • there is no setLoading(true), so the control stays enabled and shows no "Loading…";
  • the answer swaps the options when it lands and clears an earlier failure;
  • the bound page variable is never written.

  Any other read starts from "Loading…", as before. The placeholder input's description now says so ("while it reads a new set of options"). That is a text change only; the input's type is unchanged.

• object-form with subforms routes to the same MasterDetailForm, so the same reader covers it. This is pinned. The spec refuses subforms on object-form, so this does not gate (triage note 5).

• .changeset/10853-master-detail-picker-bus-readers.md: @object-ui/plugin-form patch and @object-ui/components patch.

• .changeset/10778-dashboard-bus-readers.md: one appended dated note in the objectui#10533 form (triage note 4). It says the drillDown.report drawer does re-read since objectui#10814 (PR objectui#10836), and points at drillDownReport.invalidationRefetch-10814.test.tsx. The single-record drawer half of that sentence still holds. The frontmatter is byte-identical: md5 of lines 1-3 is d527064c725ab089f1c5e8f4c419e8cd before and after, and check-changeset-overwrite.mjs reports the declared bump unchanged (@object-ui/plugin-dashboard: patch).

Unsaved lines (B3): held, then replayed after a save or a revert

This is the objectui#10712 R3 / objectui#10572 rule, applied to each collection separately.

• Unsaved means the edit save's row diff (buildMasterDetailEditBatch) finds something to write to that collection other than the parent link. A user's edit therefore reads as unsaved exactly when the save would send it, and a line changed back to its stored value reads as saved again.

  • The parent link is set aside on purpose, and it is the one place the two disagree. The save restates the link on every row it writes, as the parent id in string form. Against a backend that returns the link as a number or an expanded lookup, the save restates it on every row while the lines read as saved. That restatement is not user input, so it does not hold a re-read. sendBatch is unchanged.

• An open row editor counts as unsaved for its collection (patch round). The editor keeps its draft in its own form, not in the rows, so the rows alone cannot show it.

• A held bus re-read is replayed once, when the row editor is closed (Apply or Close) and either:

  • revert: the collection's lines read as saved again; or
  • save: this form's save has landed. sendBatch marks every collection the batch wrote, and the replay runs once the save guard releases.

  After Apply the applied line is unsaved, so the re-read stays held until that line is saved or changed back.

• A line typed while a re-read is in flight is kept, in the grid or in the row editor.

  • The answer only commits over the same rows the read was issued over, and never while the row editor is open on that collection. That check sits both before and inside the setRowState updater.
  • Otherwise the re-read is held behind the edit.
  • Unlike LineItemsPanel, the grid is not disabled during a re-read, and the row editor is never reset, re-keyed or disabled by one. Only a save disables it, as before.

• A failed re-read keeps the lines on screen and warns. A failed load still leaves the collection empty, as before.

  • A bus event during the FIRST load supersedes that load. If that re-read then fails, the warning now says "no lines have been read for it yet" instead of "the lines on screen are kept" (patch round, a one-line change).

The picker's bound value (B4)

• A re-read never writes the page variable.
• If the bound record is no longer among the re-read options, the variable keeps its value and the trigger shows no label. It shows neither the old label nor the placeholder: Radix draws the placeholder only for an empty value, and no offered item carries the value. This is the same thing the picker has always shown for a value outside its options, for example one beyond limit.
• When a later read offers that record again, its label returns.

All of this is pinned.

Patch round (contract review of ac526b1: FAIL)

The review (comment 5861421338) found one defect, and it is verified: with the row editor open on a line, a bus event committed fresh row copies under it. expandedRow then changed, and the editor's inline initialData reset overwrote the draft.

• Measured at ac526b1 with the new pins: the draft read 'first (server)' where the user had typed 'first (draft)'.
• When the server dropped the line, the editor was reset to its neighbour ('second'), so a later Apply would have mapped onto that row.

What changed in the fix:

• MasterDetailForm.tsx: the expanded state moves above the lines reader. A render-time rowEditorEntryRef (the same pattern as rowStateRef) names the collection the editor is open on, and three places read it:
  • the hold decision counts that collection as unsaved;
  • both in-flight commit sites hold instead of committing while it is open;
  • the replay effect skips it while it is open and runs again when expanded changes.
• The failed re-read warning is now accurate for a collection that never loaded.
• linesUnsaved's doc, the changeset and this body now claim only what holds about the parent link.
• The changeset sentence about a line typed in flight now names both the grid and the row editor, and says an open editor is never reset by a re-read.
• Two pending changesets re-read against this diff, per the changeset-claim-re-read bot. Both hold, and neither is edited:
  • .changeset/6726-find-envelope-records-arms.md: its row "element:record_picker options" still describes the picker, whose read is still res?.data ?? (Array.isArray(res) ? res : []). A { records } answer still reads as no rows, on a first read and on a re-read.
  • .changeset/6783-readprops-degenerate-config-bag.md: the picker still reads its bag through readProps, which this diff does not touch.

Mechanism assumptions, measured

• B1 (templates). I read both merged diffs.

  • PR objectui#10836: LineItemsPanel holds edits and keeps the grid drawn with refreshing.
  • PR objectui#10837: DatasetWidget / DashboardRenderer, one reader per widget component.

  I reused the seam. The picker needed the in-place path because its re-read flipped loading. The master-detail lines needed no refreshing state: at BASE their edit-mode read had no loading branch, so a re-read never blanked them.

• B2 (sites re-read 0 times at BASE). Confirmed with an uncommitted scratch probe on BASE a7fd877: the real SchemaRenderer, real registrations, one notifyDataChanged({ objectName: '*' }), and a bare useDataInvalidation reader as the lit control. Results, counted as reads on mount → reads after the event:

  site	before → after
  detail lines (find of po_line)	2 → 2
  header (findOne)	1 → 2
  picker options	1 → 1
  control nonce	1

  Mount reads the lines twice because the load effect runs again once the resolved details are set. That is unchanged, and the pins count from rest.

• B3 / B4. See the two sections above.

Pins and ablation

• The pins. MasterDetailForm.busReread-10853.test.tsx (12 tests: 8, plus 4 for the row editor) and record-picker.busReread-10853.test.tsx (4 tests) render through the real SchemaRenderer and the packages' own registrations. The master-detail pins use the real line-item grid; the picker pins sit in a real PageVariablesProvider with a variable bound to the picker. A bare useDataInvalidation reader is the lit control, and reads can be held open to look at the block mid-flight.
• What they cover.
  • '*' re-reads once in place: the same input node, no blanking. A save afterwards writes no line, so the baseline moved.
  • A child-object event re-reads; an unrelated object does not.
  • Two collections each answer only to their own object.
  • Held then revert; held then save (exactly one replay each).
  • A line typed in flight is kept.
  • Row editor, through the real form ("Open row", type into its fields):
    • a bus event rewrites the line: the draft is intact, Apply writes that line, the save sends update l1 { label, memo }, and the held re-read replays once after the save;
    • the server drops the line: the draft is intact, and Apply and the save target l1, not its neighbour;
    • Close without Apply replays the held re-read once;
    • a re-read already in flight when the editor opens does not commit under it, and replays once on close.
  • The subforms route re-reads.
  • Picker: in place with the value kept; its own object versus an unrelated one; the bound record gone and then back.
  • Controls: create mode and a picker with no object read nothing.

leg	mutation (on-disk proof)	result
head ee5c87d	none	Tests 12 passed (12)
A1, reverse	both source files checked out at BASE under the committed pins (trap restore); blobs 157fa2762a / b5217ead2d == BASE; reader hits linesBusNonces 0, invalidationNonce 0	Tests 10 failed, 2 passed (12). Every red is a read-count miss (expected 2 to be 3, expected "vi.fn()" to be called 2 times, but got 1 times); the 2 green are the two controls. Restored with git checkout HEAD -- ABSOLUTE_PATH: blobs bc3724ddf8 / d11d0e1fcc == HEAD, git diff HEAD 0 bytes
A2, ablation	ablation-replace.mjs (objectstack origin/main): if (unsavedLines(entry)) { → if (false) {, anchor 1 → 0, blob bc3724ddf8 → 9347b55e8e	Tests 3 failed, 5 passed (8): exactly the three hold cases ("a bus re-read ran over unsaved lines" ×2, "re-read again over the unsaved line"). Restored by the tool: blob == HEAD, git diff HEAD empty
A3, ablation	same tool: the picker's const inPlace = …signature; → const inPlace = false;, anchor 1 → 0, blob d11d0e1fcc → 4fb57ece38	Tests 1 failed, 3 passed (4): exactly "the re-read disabled the picker". Restored by the tool: blob == HEAD, git diff HEAD empty
A4, reverse (patch round)	MasterDetailForm.tsx checked out at ac526b1 under the committed pins at ea52c79 (trap restore); blob c4a5fdc81c == ac526b1; rowEditorEntryRef hits 0	Tests 4 failed, 8 passed (12): all four editor cases, each on the draft (expected 'first (server)' to be 'first (draft)' ×3; expected 'second' to be 'first (draft)'). Restored: blob d147657e95 == HEAD, git diff HEAD 0 bytes
A5, ablation (head a13048c)	same tool: rowEditorEntryRef.current === entry.id || → false || (the hold decision), anchor 1 → 0, blob 1a74a580d7 → b2dda3c6ba	Tests 3 failed, 9 passed (12): the three editor cases, on "a bus re-read ran under an open row editor". The draft stays intact there because the commit guard still holds. Restored by the tool
A6, ablation (head a13048c)	both in-flight commit sites at once, by a scripted replace with anchor counts 1/1 → 0/0 and a trap restore; blob 1a74a580d7 → 4fc16f6691	Tests 1 failed, 11 passed (12): exactly "the in-flight re-read reset the editor’s draft". Restored: blob == HEAD, git diff HEAD 0 bytes

Two ablation attempts in the patch round, reported as they happened:

• The first A5 attempt, at ea52c79, crashed the worker (Worker exited unexpectedly) instead of failing. At that head the commit guard held by re-asking, and with the hold decision ablated the re-ask looped. The shipped code cannot loop, because both checks read the same ref, but a13048c makes the guard hold directly so the two checks no longer depend on each other. A5 was then re-run cleanly (above).
• Ablating only the outer commit site stayed green (Tests 12 passed): the updater site alone covers it. That is why A6 mutates both sites.

No dist/ preflight applies: each pin imports its package's registration through a relative source path, and both mutated files are reached through it.

Gates

Pinned to head a13048c unless marked. packages/components is byte-identical since ee5c87d: git diff ee5c87d a13048c -- packages/components is empty, so its runs at ee5c87d stand.

• turbo run build --concurrency=2 --filter=@object-ui/plugin-form^... (Tasks: 11 successful), then pnpm --filter @object-ui/plugin-form type-check: script echoed, VERDICT command-exit 0. --listFiles counts the pin in tsconfig.test.json: 1.
• pnpm exec vitest run --maxWorkers=2 packages/plugin-form/ packages/components/src/renderers/basic/__tests__/record-picker.busReread-10853.test.tsx: Test Files 137 passed (137), Tests 1597 passed | 1 skipped (1598), VERDICT command-exit 0.
• At ee5c87d:
  • pnpm --filter @object-ui/components type-check: exit 0.
  • packages/components/ in 3 shards, all VERDICT command-exit 0: 108 files / 1039 tests passed; 107 files / 1212 passed + 10 skipped; 106 files passed + 1 skipped / 887 passed + 7 skipped.
  • Importers that render these blocks (the 10 files listed in the first round): Tests 1166 passed (1166).
• Lint, narrowed and measured, at a13048c:
  • eslint --format json over MasterDetailForm.tsx and its pin read 2 files: 0 errors / 23 warnings and 0 / 6. MasterDetailForm.tsx is 0 / 23 as at ac526b1 (BASE 0 / 25); with --no-inline-config it is 0 / 29 (exhaustive-deps 6, the file's existing disables).
  • record-picker.tsx is unchanged since ee5c87d (0 / 10 = BASE).
  • eslint.config.js is not type-aware, so this diff moves no verdict in an untouched file.
• Script gates at a13048c, all exit 0:
  • check:new-line-citations: VERDICT … 0 new citation(s).
  • check:control-bytes: OK. A self-scan of the changed files found no control byte.
  • check:test-path-roots, check:vi-mock-specifiers, check:vi-mock-inherit, check:unreferenced-sources, check:element-data-source-declaration: OK.
  • check:phantom-deps, check:changeset-claims, check:pending-changeset-literals: ✅.
  • scripts/check-changeset-presence.mjs ✅, scripts/check-changeset-no-major.mjs ✅, scripts/check-changeset-fixed.mjs ✅, scripts/check-type-check-coverage.mjs ✅.
  • scripts/check-changeset-overwrite.mjs (report-only): 1 added, 1 modified; the 10778 note, case 2.
• NOT MEASURED: check:sdui-registration-pins. Reason: it exits 2 without a console build. This diff changes no registration and no sideEffects array. CI builds the console.
• CI owns the repo-wide lint and the full test run.

Serial

• git merge-tree --write-tree HEAD at a13048c against a fresh origin/main (e32dae1, fetched into a private ref, since deleted): clean. No commit on main since BASE touches these three files.
• PageView is not touched (PR objectui#10777, seat 2).

Acceptance notes

Not changed here. Each was read at source unless it says it was measured.

• The lines subscribe to each collection's child object only. A write declared on the PARENT object, for example a record action that rewrites lines on the server and announces the parent, re-reads the header but not the lines. '*', which PageView will send, does both. This is the same boundary LineItemsPanel has.
• Measured at BASE and unchanged: the edit-mode lines are read twice on mount. The load effect runs again when the resolved details are first set, even when nothing needed resolving, and the first read is superseded. Carrier: none.
• Unchanged: a failed first load of a collection's lines still leaves it empty with no message. Only the new re-read path keeps the lines and warns. Carrier: none.
• The picker's in-place re-read that fails shows "Failed to load" and disables the control, the same as a failed first load. The options it had stay in state and come back with the next read that lands.
• With the console's mutation bridge mounted, the save's own batch echoes on the bus while its collections still read unsaved. By the ordering read at source, that echo is held and replayed once after the save, so there is one extra read per written collection, in place. The pins' adapter has no onMutation, so this is not pinned.
• element:record_picker's emptyText input description cites a line address, :213 read at a7fd877, for the "No records" default. At a7fd877 that line is inside a comment block, not the default. This is a stale same-file line address in published registry text, outside this card's class, so it is not repaired here. Carrier: none.

---

Generated by Claude Code

[github-actions]

changeset-claim-re-read

⚠️ 1 address(es) in this pull request's own prose name a tree it replaced

Each was read from a tree this change itself moves, so a reader who follows it lands somewhere else. ⛔ Nothing here blocks and nothing here says the sentence is false — the question asked is arithmetic: does this diff move the line that number points at?

• in this body, record-picker.tsx:213 — this change moves packages/components/src/renderers/basic/record-picker.tsx:213 to :245

  • element:record_picker's emptyText input description cites record-picker.tsx:213 for the "No records" default.

⛔ The repair is not to correct the number. Changing :246 to :274 is true today and born false again on the next insertion — objectui#9509 states that before anything else. Bind the number to the tree it was read from (`:246` at `b8a006883d`, `:274` at this head), which cannot re-stale because each number names its own tree; or state a rule instead of a coordinate, the way objectui#9495 replaced a file count with "every file in git diff --name-only against the merge base".

⚠️ 2 pending changeset(s) describe a file this change touches

Their bodies publish verbatim into the CHANGELOG at the next release, so this is a request to re-read them against your diff — addressed here because you are the one seat that can answer it without re-deriving anything.

⛔ Nothing here blocks, and nothing here is a verdict on your change. This gate exits 0, is not a required context, and judges name resolution, never meaning: it asked whether a pending body names a file you touched. "Is this sentence still true?" is the one question it will not answer, and the one you are being asked to answer.

.changeset/6726-find-envelope-records-arms.md

• names components/src/renderers/basic/record-picker.tsx → packages/components/src/renderers/basic/record-picker.tsx — edited by this change

  | module | what it does | | --- | --- | | components/src/hooks/related-count-store.ts | related-list tab badge count | | components/src/renderers/basic/data-list.tsx | element:repeater rows | | components/src/renderers/basic/elements.tsx | element:number client-side aggregate | | components/src/renderers/basic/record-picker.tsx | element:record_picker options | | plugin-detail/src/renderers/record-activity.tsx | record:activity self-fetch | | plugin-detail/src/renderers/record-history.tsx | record:history self-fetch | | plugin-view/src/ObjectView.tsx | non-grid (kanban / calendar / gallery / timeline) fetch |

.changeset/6783-readprops-degenerate-config-bag.md

• names record-picker.tsx → packages/components/src/renderers/basic/record-picker.tsx — edited by this change

  Five modules under packages/components/src/renderers/basic/ — elements.tsx, data-list.tsx, text-input.tsx, record-picker.tsx, metadata-viewer.tsx — each carried a copy of the same reader, { ...(schema?.props ?? {}), ...(schema?.properties ?? {}) }. ?? only replaces null/undefined, so a non-object bag went into the object spread and came back out as indexed keys: for properties: 'not-a-bag', the config bag a renderer received was { '0': 'n', '1': 'o', … '8': 'g' } — nine keys nobody authored. The five copies are now one readProps (renderers/basic/readProps.ts) that asks isConfigBag, and a degenerate bag on either side contributes no keys.

Read the paragraph, not the line: both false halves of the objectui#8617 claim sat in one paragraph, and correcting either alone would have left it asserting the same wrong thing.

If a claim did go false, correct the body. That is precedented and prose-only, frontmatter untouched; check-changeset-overwrite.mjs will report the correction as its own case 2 ("correcting a declaration on purpose … legitimate"), which is the intended shape — one gate asks for the read, the other records the write.

Not covered, stated so nobody reads this as more: a born-false claim that spells no line address at all (objectui#9495 coordinated one by ORDINAL — "a grep finds that member first" — and deciding that means reading what the sentence means), a claim spelled as a symbol or a package rather than a backticked file name, and a file named ambiguously.

Compared the checked-out tree with fab627ff9 (merge-base with origin/main): 4 file(s) changed outside .changeset/, read against 1643 pending declaration(s) that publish a body (2233 pending in total). · run

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3083.2 KB	3104.5 KB
Main entry chunk (gzip)	148.5 KB	350 KB
Entry file	index-COyRhQrg.js	—
Status	PASS	—

The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it.

---

📦 Bundle Size Report

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.57KB	6.15KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	27.95KB	10.04KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.17KB	10.58KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.39KB	1.21KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.52KB	3.45KB
auth (index.js)	3.63KB	1.64KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	27.13KB	7.95KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	557.59KB	133.61KB
core (index.js)	9.93KB	3.94KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	225.01KB	62.50KB
fields (index.js)	261.26KB	66.37KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	2.59KB	1.22KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.24KB	2.27KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	39.40KB	12.91KB
i18n (translateFn.js)	0.20KB	0.18KB
i18n (useDisplayLocale.js)	3.52KB	1.76KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	39.28KB	11.09KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	6.62KB	2.45KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.33KB	3.07KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	16.01KB	3.93KB
plugin-calendar (index.js)	51.73KB	14.72KB
plugin-charts (index.js)	83.99KB	22.86KB
plugin-chatbot (index.js)	197.67KB	46.90KB
plugin-dashboard (index.js)	136.82KB	36.44KB
plugin-designer (index.js)	216.42KB	44.47KB
plugin-detail (index.js)	233.48KB	61.79KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	160.95KB	41.33KB
plugin-gantt (index.js)	169.83KB	41.99KB
plugin-grid (index.js)	218.66KB	59.90KB
plugin-kanban (index.js)	48.43KB	15.11KB
plugin-list (index.js)	114.78KB	28.44KB
plugin-map (index.js)	22.90KB	7.62KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	44.17KB	12.20KB
plugin-timeline (index.js)	31.00KB	9.09KB
plugin-tree (index.js)	11.21KB	3.89KB
plugin-view (index.js)	87.86KB	22.02KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.81KB	3.58KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	119.16KB	39.05KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.03KB	1.86KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (body-dialect.js)	4.78KB	2.09KB
sdui-parser (codegen.js)	7.50KB	3.05KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	6.16KB	2.71KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.84KB	1.90KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	18.27KB	6.20KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	3.52KB	1.36KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.26KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	2.52KB	1.31KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	17.15KB	6.32KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.27KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

• ✅ Core packages should be < 50KB gzipped
• ✅ Component packages should be < 100KB gzipped
• ⚠️ Plugin packages should be < 150KB gzipped

[objectstack-fleet]

Contract review

Served-tier: 229/229 CONTRACT_REVIEW_TIER
Head-sha: ac526b1a299586395c3e4504b6f19328566f5cc5

Read: card objectui#10853 body and its three comments (triage 5860720827, claim 5860833997; the os-dev-report 5861273028 treated as claims only); the objectui#10519 decision 5830213285; merged PR objectui#10836 and objectui#10837 diffs (via the API); PR #10873 body, file list (6 files) and its diff against the merge-base a7fd877c (origin/main is fab627ff; the PR is 2 commits ee5c87d, ac526b1); at the head: packages/react/src/data-invalidation.ts, packages/plugin-form/src/MasterDetailForm.tsx (head and base), masterDetailTx.ts, sanitize.ts (changedFields, isSameStoredValue), ObjectForm.tsx (routesToMasterDetail, the record-fetch effect, useRecordInvalidation), initialRecord.ts, packages/components/src/renderers/form/form.tsx (the defaultValues layout-effect reset), packages/fields/src/widgets/GridField.tsx (ghost row, per-keystroke onChange), SelectField.tsx, renderers/form/select.tsx, record-picker.tsx (head and base line 213), plugin-dashboard DrillDownDrawer.tsx, ObjectDataTable.tsx, RecordDetailDrawer.tsx; both pin files; AGENTS.md §5 and §9; .changeset/config.json; .changeset/10778-dashboard-bus-readers.md at base and head; .changeset/10853-master-detail-picker-bus-readers.md; every pending changeset naming the touched files (grep for sentences about loading, in-flight input, baseline, slots); packages/components/package.json and packages/plugin-form/package.json (exports, files); the head's check-runs twice. Ran the two pin files once at the head in my own throwaway worktree under os-verify-lock.sh (slot review-10873): Test Files 2 passed (2), Tests 12 passed (12); the worktree was clean and is removed.

① Derived judgments

1. One bus, not a second channel — RIGHT. useObjectsInvalidation(names) subscribes through subscribeDataChanges, which adds to the same module-level listeners Set that useDataInvalidation subscribes through; it tests each change with dataChangeMatches(change, name) with no recordId, which is exactly what useDataInvalidation(name) does (its first line returns true for '*'; both compare change.objectName with strict equality, neither normalises names). It subscribes only when the set is non-empty, mirroring if (!objectName) return; the effect cleanup unsubscribes on unmount and on every namesKey change; it is one unconditional hook call with fixed inner hooks (useState, useEffect), keyed on the set's content (JSON.stringify of the sorted unique names), so a fresh array of the same names neither resubscribes nor resets a nonce, and a nonce survives a set change the way useDataInvalidation keeps its nonce across an objectName change. Nothing is exported. No divergence in substance; the only difference is that nonces for names no longer authored stay in the state object, which nothing reads. Reuse of useDataInvalidation was possible (one child component per detail collection, reporting its nonce up), but the triage's "no second channel" reads as no other event mechanism (a window event, a refreshKey, a relay), and this is a composition of the bus's own exported primitives.

2. Held dirty lines — WRONG on one path; the traced paths otherwise hold.

• Traced and holding: a 'bus' read over a collection that linesUnsaved reads as dirty is added to heldLinesRereadRef and not issued; the replay effect (keyed on rowState, saving) replays a held entry once its lines read clean again (held.delete before readLines(e, 'replay'), so once), and after a save through linesSavedRef (marked in sendBatch after childRowsAfterSave; the replay effect returns before clearing that set while saving is true, so the mark survives until the guard releases). A line typed in the grid while a read is in flight changes rowState.rows identity (GridField's cells call onChange per keystroke), the commit is refused against rowsAtIssue both before and inside the setRowState updater, and the read is re-asked as 'bus', which holds it. One commit per collection: a per-collection seq plus a gen the load effect's teardown bumps supersede every older read. Every commit writes rows and original together, and the save's baseline advance (childRowsAfterSave) is unchanged, so I found no sequence that leaves a stale baseline or applies one answer twice. The pins cover hold-then-revert, hold-then-save, and typed-in-flight.
• The defect: the row editor ("expand to full form", offered in grid mode whenever formFields outnumber columns, and always in inlineMode: 'form') holds the user's draft inside its nested ObjectForm's react-hook-form state, not in rowState, so linesUnsaved cannot see it and the collection reads clean. A bus event then commits fresh row copies (rows.map((r) → ({ ...r }))). expandedRow is rowState[entryId].rows[rowIdx], handed to the editor as an inline schema.initialData; ObjectForm's record-fetch effect lists schema.initialData in its deps and, with no recordId, calls setInitialData(resolveInitialRecord(schema)); the form renderer's defaultValues layout effect then runs form.reset(defaultValues) whenever the JSON of the defaults differs from the last one, carrying only fields absent from both the outgoing and incoming defaults. So: open the editor on a line, type, and let a page action that rewrites that line (or '*') land: the re-read commits under the open editor and the reset overwrites every drafted field the row names. If the server no longer returns the row, expandedRow becomes another row or undefined, the editor is reset to that, and a later "Apply" calls applyRowEdit(entryId, rowIdx, values), which maps by index onto whatever row now sits there, or onto nothing. The head itself treats the editor as an input surface that must be held: it wraps it in ‹fieldset disabled=saving› and the pending 10631 changeset says "An open row editor … is disabled too, 'Apply' included". The LineItemsPanel precedent has no row editor, so its refreshing grid-disable never had to cover one; this form does, and the PR's deviation (grid left enabled, identity guard on rows only) leaves that one surface unheld. Not pinned. The changeset's sentence "A line typed while a re-read is in flight is kept" is therefore true for grid cells and false for the editor's draft.
• Initial load: the change from one Promise.all commit to one commit per collection alters the FIRST load only in that collections now render progressively (the first to answer draws before the others; base drew all at once), which nothing reads wrongly: sendBatch, sumRows and MasterDetailLines all read by the current entries' ids. Ordering is unchanged (the derive effect's setResolvedEntries still re-runs the load effect, the first run's reads now superseded by the gen bump where base cancelled them). Not pruning slots of entries no longer authored is unread by anything (all readers iterate entries). The unresolved early-return now commits an empty slot synchronously on the first run instead of inside the cancelled Promise.all; the grid shows no rows either way.

3. The picker — RIGHT. inPlace is true only when committedReadRef holds the same adapter and the same [object, filterKey, sortKey, limit] signature, and then setLoading(true) is skipped, so disabled={loading || !!error || !object} stays false and the SelectValue placeholder stays placeholder, not 'Loading…'; the ref is written only when a read commits, so a re-read after a failed first read still starts from "Loading…". binding.setValue is called only from handleChange; the effect never touches it (pinned: bound-value stays a1). Bound record gone: value is kept and Radix draws neither the old label nor the placeholder (pinned textContent ''), and the label returns when the record is offered again (pinned). Acceptable face: it is what the head's form:select renders for a value outside its authored options (Radix, no fallback), and what this picker already showed at base for a defaultValue beyond limit. The counter-example is SelectField in @object-ui/fields, which prunes a value no longer offered (isValueStillOffered → onChange(null)), but that is a cascade-clear of a record field, and triage note 2 requires the bound page variable to be kept. Registration metadata is published: packages/components/src/index.ts imports ./renderers at module scope (→ ./basic → ./record-picker), and the package's exports . points at ./dist/index.js with files: ['dist', …], so the inputs[].description string ships in the bundle. The new text is true at the head: the placeholder is replaced by 'Loading…' only when loading is set, which happens only for a read that is not in place (first read, or a changed adapter or query), and by 'Failed to load' on error, which a failed in-place re-read also sets.

4. Pins — RIGHT. MasterDetailForm.busReread-10853.test.tsx renders through the real SchemaRenderer inside SchemaRendererProvider, registers through import './index' (the package's own entry) and registerAllFields() (the real grid), fires the real notifyDataChanged, lights BusControl with useDataInvalidation('po_line'), and uses unrelated_object as the negative (plus po_note versus po_line across two collections). record-picker.busReread-10853.test.tsx renders through the real SchemaRenderer, registers through import '../../../renderers', supplies the adapter through AdapterCtx.Provider (the context useAdapter reads) inside a real PageVariablesProvider, lights a control on 'account', and uses unrelated_object as the negative. No timeout is raised anywhere (no testTimeout, no per-test timeout argument); nothing is skipped (no .skip, .only, .todo). Each re-read case asserts a read count one above rest (atRest + 1, toHaveBeenCalledTimes(2), toHaveBeenCalledTimes(3)), and the base effects carry no nonce (base picker deps [adapter, object, filterKey, sortKey, limit]; base form effect [isEdit, dataSource, schema.recordId, resolvedEntries]), so those cannot hold on base; the two controls pass on base by construction, as controls should. My single run at the head: 12 passed.

5. Prose and shape — RIGHT except the one sentence noted under 2. 10853-master-detail-picker-bus-readers.md: frontmatter patch for @object-ui/plugin-form and @object-ui/components, no major, both in the fixed group; every sentence checked at the head holds except "A line typed while a re-read is in flight is kept, and the re-read is held behind it" for the row editor's draft (item 2). One gloss is loose: "(lines the next save would write)" — linesUnsaved normalises the parent link on original, but sendBatch does not, so with a backend that returns the FK as a number or an expanded lookup the real save restates the FK on every row (base behaviour since objectui#10108) while linesUnsaved reads clean; no user input is affected by that difference. 10778-dashboard-bus-readers.md: the first 18 lines are byte-identical to base and the frontmatter md5 is d527064c725ab089f1c5e8f4c419e8cd both sides; the note's claims hold — DrillDownDrawer.tsx renders type: 'spec-report' through SchemaRenderer, PR objectui#10836's own changeset says "The spec-report a drill-down drawer opens for drillDown.report re-reads the same way", apps/console/src/__tests__/drillDownReport.invalidationRefetch-10814.test.tsx exists at the head, and the table row's single-record drawer (RecordDetailDrawer) performs no read and names no bus hook, so "still does not re-read" stands. PR body first line Fixes #10853; Clause-②: no holds (no export, index, spec key, registry input name or type moves; the picker change to registration is description text only). No pending changeset naming the touched files is made false (10564's baseline rule, 10631's in-flight guard, 10664's sort key and 6372's slot identity all still hold).

6. Out-of-scope notes — all three RIGHT. (a) Read twice on mount: at base the load effect keys on resolvedEntries, which the derive effect sets even when nothing needs deriving (if (!needsDerive) { setResolvedEntries(baseEntries); return; }), so the effect runs at mount and again on that set, the first run cancelled; unchanged at the head (superseded by the gen bump). (b) A failed first load: the head's 'load' catch sets rows = [] and warns nothing, as base's catch { return [e.id, { rows: [], original: [] }] } did; only 'bus'/'replay' warn and keep the lines. (c) emptyText's description cites record-picker.tsx:213: at base that line is inside the toText comment block and the default ?? 'No records' sits later (head line 274); a same-file address, so under the AGENTS.md #11 carve-out and not a gate finding, but stale.

② Semver level

patch for @object-ui/plugin-form and @object-ui/components is right: two renderers gain a bus subscription and an in-place re-read; no export, type, spec key or registry input shape changes; the registry description is text. No major (AGENTS.md §9 version policy).

③ Boundary flags

• Row-editor draft (item 2): the fix shape is the seat's call — count an open editor for an entry as unsaved in unsavedLines, or hold that entry's re-read while expanded.entryId names it and disable the editor's fieldset during an in-flight re-read the way saving does — and a pin should open the editor, type, land a changed row on the bus, and assert the draft and the target row.
• Narrow race, not a defect on its own: a bus event during the FIRST load supersedes the 'load' read (per-collection seq); if that 'bus' read then fails, its warning says the lines on screen are kept while the collection has none yet.
• The FK-restatement gloss in the 10853 changeset and the PR body's "cannot disagree" (item 5); user input is unaffected.
• The picker's "no label" face is base behaviour for any value outside the options; SelectField in @object-ui/fields takes the opposite policy (prune) for record fields.
• With the console's mutation bridge, a save's own echo is held and replayed once per written collection (one extra in-place read per save), as the PR says; not pinned (the pins' adapter has no onMutation).
• CI as read: poll 1 at 2026-09-28T00:32:40Z — 42 check-runs, 29 success, 3 skipped, 10 in progress (Lint, Skill Example Check, Spec Main Shape Gate, Test shards 1/8 to 8/8, Type Check). Poll 2 at 2026-09-28T00:45:36Z — 43 check-runs, 40 success, 3 skipped (Test (coverage shard …/4), Test (coverage), dependabot), 0 failures, 0 in progress; among the successes: Lint, Type Check, Test shards 1/8 to 8/8, Test (dist pins), Build & E2E, Changeset Declaration, Changeset Bump Policy, Changeset Fixed Group Check, Changeset Claim Re-read, Changeset Overwrite Report, Line Citation Gate, Control Byte Scan, Governed Surface Queue Guard, Spec Main Shape Gate, Bundle Analysis, Build Docs.
• Nothing in the PR touches PageView (PR objectui#10777's file); the branch merges cleanly by the API's mergeable: true (mergeable_state: behind).

Implemented-by: claude/issue-10853-master-detail-picker-bus-readers
Reviewed-by: session_01MCg3z5cGrV51xEddpGcxEY

VERDICT: FAIL

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3083.2 KB	3104.5 KB
Main entry chunk (gzip)	148.4 KB	350 KB
Entry file	index-YI2Fkfxl.js	—
Status	PASS	—

The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it.

---

📦 Bundle Size Report

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.57KB	6.15KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	27.95KB	10.04KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.17KB	10.58KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.39KB	1.21KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.52KB	3.45KB
auth (index.js)	3.63KB	1.64KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	27.13KB	7.95KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	557.59KB	133.61KB
core (index.js)	9.93KB	3.94KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	225.01KB	62.50KB
fields (index.js)	261.26KB	66.37KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	2.59KB	1.22KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.24KB	2.27KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	39.40KB	12.91KB
i18n (translateFn.js)	0.20KB	0.18KB
i18n (useDisplayLocale.js)	3.52KB	1.76KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	39.28KB	11.09KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	6.62KB	2.45KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.33KB	3.07KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	16.01KB	3.93KB
plugin-calendar (index.js)	51.73KB	14.72KB
plugin-charts (index.js)	83.99KB	22.86KB
plugin-chatbot (index.js)	197.67KB	46.90KB
plugin-dashboard (index.js)	136.82KB	36.44KB
plugin-designer (index.js)	216.42KB	44.47KB
plugin-detail (index.js)	233.48KB	61.79KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	161.16KB	41.40KB
plugin-gantt (index.js)	169.83KB	41.99KB
plugin-grid (index.js)	218.66KB	59.90KB
plugin-kanban (index.js)	48.43KB	15.11KB
plugin-list (index.js)	114.78KB	28.44KB
plugin-map (index.js)	22.90KB	7.62KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	44.17KB	12.20KB
plugin-timeline (index.js)	31.00KB	9.09KB
plugin-tree (index.js)	11.21KB	3.89KB
plugin-view (index.js)	87.86KB	22.02KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.81KB	3.58KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	119.16KB	39.05KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.03KB	1.86KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (body-dialect.js)	4.78KB	2.09KB
sdui-parser (codegen.js)	7.50KB	3.05KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	6.16KB	2.71KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.84KB	1.90KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	18.27KB	6.20KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	3.52KB	1.36KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.26KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	2.52KB	1.31KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	17.15KB	6.32KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.27KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

• ✅ Core packages should be < 50KB gzipped
• ✅ Component packages should be < 100KB gzipped
• ⚠️ Plugin packages should be < 150KB gzipped

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3083.2 KB	3104.5 KB
Main entry chunk (gzip)	148.4 KB	350 KB
Entry file	index-DI7c78sK.js	—
Status	PASS	—

The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it.

---

📦 Bundle Size Report

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.57KB	6.15KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	27.95KB	10.04KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.17KB	10.58KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.39KB	1.21KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.52KB	3.45KB
auth (index.js)	3.63KB	1.64KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	27.13KB	7.95KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	557.59KB	133.61KB
core (index.js)	9.93KB	3.94KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	225.01KB	62.50KB
fields (index.js)	261.26KB	66.37KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	2.59KB	1.22KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.24KB	2.27KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	39.40KB	12.91KB
i18n (translateFn.js)	0.20KB	0.18KB
i18n (useDisplayLocale.js)	3.52KB	1.76KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	39.28KB	11.09KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	6.62KB	2.45KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.33KB	3.07KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	16.01KB	3.93KB
plugin-calendar (index.js)	51.73KB	14.72KB
plugin-charts (index.js)	83.99KB	22.86KB
plugin-chatbot (index.js)	197.67KB	46.90KB
plugin-dashboard (index.js)	136.82KB	36.44KB
plugin-designer (index.js)	216.42KB	44.47KB
plugin-detail (index.js)	233.48KB	61.79KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	161.21KB	41.41KB
plugin-gantt (index.js)	169.83KB	41.99KB
plugin-grid (index.js)	218.66KB	59.90KB
plugin-kanban (index.js)	48.43KB	15.11KB
plugin-list (index.js)	114.78KB	28.44KB
plugin-map (index.js)	22.90KB	7.62KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	44.17KB	12.20KB
plugin-timeline (index.js)	31.00KB	9.09KB
plugin-tree (index.js)	11.21KB	3.89KB
plugin-view (index.js)	87.86KB	22.02KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.81KB	3.58KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	119.16KB	39.05KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.03KB	1.86KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (body-dialect.js)	4.78KB	2.09KB
sdui-parser (codegen.js)	7.50KB	3.05KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	6.16KB	2.71KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.84KB	1.90KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	18.27KB	6.20KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	3.52KB	1.36KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.26KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	2.52KB	1.31KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	17.15KB	6.32KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.27KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

• ✅ Core packages should be < 50KB gzipped
• ✅ Component packages should be < 100KB gzipped
• ⚠️ Plugin packages should be < 150KB gzipped

[objectstack-fleet]

Contract review

Served-tier: 82/82 CONTRACT_REVIEW_TIER
Head-sha: a13048ce95eacfac6f7ef7d33210e7520bf075cc

Read: the previous record (review/record.md, VERDICT FAIL on item 2); the card thread (triage 5860720827, claim 5860833997, the os-dev-report 5861273028 treated as claims only); the PR body at a13048c and its file list (6 files; 4 commits; mergeable: true, mergeable_state: clean; merge-base with origin/main e32dae16 is a7fd877c); the fetched ref refs/review/pr-10873-re2 and the delta ac526b1..a13048c (two commits ea52c79, a13048c; 3 files: MasterDetailForm.tsx, its pin, the 10853 changeset); the blob ids of all 6 PR files at both heads; at the head: packages/plugin-form/src/MasterDetailForm.tsx in full (useObjectsInvalidation, linesUnsaved, MasterDetailLines, the reader block, the replay effect, applyRowEdit, addRowViaForm, cancelRowEdit, sendBatch, the row-editor Card), ObjectForm.tsx (the inline-fields effect, readsRecord, the record-fetch effect), initialRecord.ts, recordInvalidation.ts (the readsRecord gate), packages/components/src/renderers/form/form.tsx (the defaultValues layout-effect reset), packages/fields/src/widgets/GridField.tsx (emit, applyPatch, the "Open row" button, every emit call site), masterDetailTx.ts (buildMasterDetailEditBatch, diffRows, childRowsAfterSave), sanitize.ts (isSameStoredValue, changedFields), record-picker.tsx (the 6726 arm and readProps); both pin files at the head and the pin's delta; AGENTS.md §5, §9 测试纪律 and 版本号策略; .changeset/10853-master-detail-picker-bus-readers.md at both heads, .changeset/10778-dashboard-bus-readers.md (blob and diff against a7fd877), 6726-find-envelope-records-arms.md, 6783-readprops-degenerate-config-bag.md, 10631-master-detail-save-in-flight.md; the head's check-runs twice. Ran the two pin files once at the head in a throwaway worktree under os-verify-lock.sh (slot review-10873-re2, acquired after 0 s, held 14 s): Test Files 2 passed (2), Tests 16 passed (16); the worktree had no change outside node_modules and is removed; /home/user/objectui is untouched (clean, on its own branch).

① Derived judgments

1. The row-editor defect is fixed — RIGHT.

• The hold decision: unsavedLines is now rowEditorEntryRef.current === entry.id || linesUnsaved(…), and readLines(entry, 'bus') consults it before issuing anything, adding the id to heldLinesRereadRef and returning. So editor open, draft typed, a bus event that rewrites the line (or drops it, or '*'): no read is issued at all, rowState does not move, expandedRow keeps its identity, the editor's inline initialData serialises to the same key, and the form renderer's lastDefaultsKey layout effect does not reset. Pins 1 to 3 assert both halves (reads equals atRest; label/memo still the draft).
• Both in-flight commit sites: after isCurrent(), if (inPlace && rowEditorEntryRef.current === entry.id) adds to the held set and returns, placed before the rowsAtIssue re-ask; and inside the setRowState updater the guard is prev[entry.id]?.rows !== rowsAtIssue || rowEditorEntryRef.current === entry.id, returning prev. The outer site holds directly instead of re-asking, so it no longer depends on the hold decision; that is what closes the loop the dev reports at ea52c79 with the hold decision ablated. Pin 4 (read issued, then the editor opened, then the answer): draft intact, grid still ['first', 'second'], reads stays at atRest + 1.
• Never reset, re-keyed or disabled by a re-read: the editor's key is row-${entryId}-${rowIdx} from expanded alone, its fieldset is disabled by saving alone, and neither is touched by the delta; a held or uncommitted read leaves rowState as it was. The editor's nested ObjectForm has no recordId, so readsRecord is false and it holds no bus subscription of its own.
• After Close: cancelRowEdit sets expanded null; the render writes the ref null; the replay effect, whose deps now include expanded, runs: the id is held, the ref is null, linesSavedRef is empty, unsavedLines(e) is false (no grid edit) → held.delete(e.id) then readLines(e, 'replay'), once. Pin 3 asserts atRest + 1 and that it stays there after a pause; pin 4 asserts atRest + 2.
• After Apply: onSuccess runs applyRowEdit (the row at rowIdx becomes { ...r, ...values }) and setExpanded(null); the replay effect finds the ref null, linesSavedRef empty, and linesUnsaved true because buildMasterDetailEditBatch → changedFields(writable, before) finds label/memo moved → it stays held (pin 1: atRest after Apply). Save → sendBatch adds the entry to linesSavedRef and advances original through childRowsAfterSave; the replay effect returns before clearing while saving; releaseSave → the effect replays once → atRest + 1 (pins 1 and 2, and pin 1 asserts it does not replay again). An Apply whose values equal the stored row yields an empty changedFields, no op, so it reads as saved and replays at once, which is the right reading.
• Apply targets the intended row in every one of those sequences: while the editor is open the collection's rows array is never replaced by the bus path (the read is either never issued or never committed), so expanded.rowIdx still addresses the row it was opened on. Pin 1's op is update l1 { label, memo }; pin 2's update ids are ['l1'] after the server dropped l1 and the grid shows ['first (draft)', 'second'].
• Sequences the pins do not cover, traced at the head: (i) the editor switches from a row of collection A to a row of collection B while A is held: the ref becomes B, the replay effect runs on the expanded change and replays A once if its lines read saved; B's own events are then held while open; nothing is stranded. The outgoing editor's draft is discarded by the key change, which is what "Open row" on another row did at base. (ii) Switching rows within A: the ref stays A, the hold stays, Close replays. (iii) The collection removed from the authored details while its editor is open: the derive effect sets resolvedEntries, the load effect clears heldLinesRereadRef and linesSavedRef, bumps gen and issues 'load' reads for the surviving entries; expandedDetail is undefined so the editor is not rendered; the replay effect iterates entries, so a dead id is neither replayed nor waited on. expanded keeps the dead id until the next open; nothing reads it. (iv) Unmount with a held re-read: the load effect's cleanup bumps gen, useObjectsInvalidation unsubscribes, a late answer fails isCurrent(), the held set is a ref and dies with the component; no state write after unmount, no timer. (v) Save clicked with the editor open (the action bar is not gated on expanded): the replay effect after the save skips the editor's entry and then clears linesSavedRef; on Close the replay still fires because childRowsAfterSave advanced original to what was written and linesUnsaved reads false. One replay, none lost. (vi) A bus event during a save with the editor open: held at issue (the hold decision does not consult saving), replayed on Close. (vii) inlineMode: 'form' Add: addRowViaForm appends {} and opens with isNew; Cancel removes the blank row and closes → the rows are the original set again → replay; Apply on the new row is a create → held until the save → the created row takes its id, the baseline advances → replay. (viii) Loop: the only re-ask left is the rowsAtIssue mismatch (a grid edit during the flight), one read per identity change; GridField's emit is reached only from user actions (applyPatch, addRow, remove, duplicate, reorder; no effect emits on a prop change), so identity cannot churn on its own. I found no sequence that loses a draft, double-applies an answer, loops, or leaves a held re-read never replayed.

2. No regression on the paths the previous record judged RIGHT — RIGHT.

• The grid-cell hold: the rowsAtIssue guard stands at both sites (outer: re-ask as 'bus'; updater: hold), and pin 6 (typed in flight kept, atRest + 1) still passes.
• seq/gen: the seq increment, isCurrent, and the load effect's teardown bump are byte-identical to ac526b1.
• linesSavedRef after a save: sendBatch's marking is unchanged, and the replay effect still returns before linesSavedRef.current.clear() while saving is true.
• The first load: both new checks are gated on inPlace, so a 'load' read commits exactly as at ac526b1.
• expanded moved above the reader: a useState declared earlier in the same component; the hook order is fixed across renders; expandedRow, expandedDetail, applyRowEdit, addRowViaForm and cancelRowEdit are unchanged.
• The render-time ref: rowEditorEntryRef.current = expanded?.entryId ?? null is assigned in render, the pattern rowStateRef.current = rowState already uses in this component. Every reader is either an effect of this component, which runs after the commit that wrote the ref and so never sees an older expanded, or the async continuation in readLines, which reads the value of the latest committed render. The two gaps are both safe: between setExpanded(open) and its render an answer sees "closed" and commits, but no editor is mounted yet, so no draft exists, and the editor then opens on the committed row; between setExpanded(null) and its render (including the async onSuccess path) an answer sees "open" and holds, and the replay effect of that very render, keyed on expanded, replays it once. StrictMode's double render writes the same value. No stale read.

3. The narrow-race wording — RIGHT. The inPlace catch now interpolates rowsAtIssue === undefined ? 'no lines have been read for it yet' : 'the lines on screen are kept'. rowsAtIssue is rowStateRef.current[entry.id]?.rows read at issue, undefined only when no commit has ever written the slot, which is exactly the race (a 'bus' read supersedes the first 'load', then fails). A failed 'load' commits rows = [], so a later failed re-read says "kept", which is also true. The else branch yields the byte-identical message of ac526b1 (…after a data change; the lines on screen are kept.), and the 'load' branch (rows = [], no warning) is unchanged.

4. Prose — RIGHT.

• The 10853 changeset: frontmatter (lines 1 to 4) md5 329d544f777da815044a51592f11206d at both heads. Each changed sentence holds at the head: "lines the user has changed since they were last read or saved (compared the way the save compares rows, with the link to the parent set aside)" is linesUnsaved running buildMasterDetailEditBatch over rows and original with the FK normalised on original, and original is written at every commit and by childRowsAfterSave; "or while the row editor ("Open row") is open on it, its re-read is held" is the hold decision, and "Open row" is the aria-label of the one expand control GridField renders; "It runs once, when the row editor is closed and either the lines have been changed back or this form's save has landed" is the replay effect's three conditions; "A line typed while a re-read is in flight, in the grid or in the row editor, is kept, and the re-read is held behind it; an open row editor is never reset by a re-read" is the two in-flight sites, pinned by tests 4 and 6. The rest of the file is byte-identical. The linesUnsaved doc now says the save restates the parent link while the lines read as saved and that the restatement does not hold a re-read, which is true (sendBatch unchanged; withFk restates on every row).
• The 10778 note: blob 6da9d731… identical at ac526b1 and a13048c; against a7fd877 the file differs by the 9 appended lines only.
• PR body: first line Fixes #10853. Clause-②: no holds: the delta adds or moves no export or import, no spec key, no registry input. The line address is bound to its tree: ":213 read at a7fd877… At a7fd877 that line is inside a comment block". The 6726 claim is true: record-picker.tsx at the head still reads const data: any[] = res?.data ?? (Array.isArray(res) ? res : []), with records named only in the comment above it, so a { records } answer reads as no rows on a first read and on a re-read alike. The 6783 claim is true: import { readProps } from './readProps' and const props = readProps… are in place and the file is untouched by the delta. The Patch-round section's measurements match the source (see 5).

5. Pins — RIGHT. The four new tests sit in the same file, rendered through SchemaRendererProvider and SchemaRenderer, registered through import './index' (the package entry) and registerAllFields() (the real grid). EDITOR_DETAIL carries two formFields over one column, so the real grid offers onRowExpand; openRow clicks getAllByLabelText('Open row'), GridField's own button; the editor is found by data-testid="md-row-form" (the Card in MasterDetailForm) and its fields by input[name=…] inside it, that is the nested real ObjectForm (key row-…, submitText: 'Apply'); editorButton('Apply') submits through that form's own submit into submitHandler then onSuccess then applyRowEdit; editorButton('Close') is the CardHeader button wired to cancelRowEdit. The inputsOf helper now excludes inputs inside md-row-form, which changes nothing for tests 1 to 8 (no editor is open there). No timeout is raised (the only setTimeouts are 20 ms and 30 ms settles), nothing is skipped (no .skip, .only, .todo), no vi.mock, no fake timers. Each can fail on ac526b1's MasterDetailForm.tsx: there unsavedLines had no editor term and neither commit site had an editor check, so in tests 1 to 3 the bus read is issued (reads becomes atRest + 1, failing the first count assertion) and, had it not, the commit under the editor would reset the draft to first (server) (tests 1 and 3) or to its neighbour second (test 2, l1 dropped); in test 4 the in-flight answer commits and resets to first (server). That is the dev's A4 result ('first (server)' ×3, 'second' ×1). A5 (the hold term ablated to false ||): the read is issued, so tests 1 to 3 fail at "a bus re-read ran under an open row editor", the commit sites still hold so the draft survives, and test 4 (its read was issued before the editor opened anyway) passes: 3 failed, 9 passed, consistent. A6 (both commit sites ablated): tests 1 to 3 are still held at issue and pass; test 4's answer commits and resets: 1 failed, consistent; "the outer site alone stays green" is consistent because the updater's check holds by itself. My run at the head: 16 passed (12 in the master-detail file, 4 in the picker's).

6. Carry-forward — RIGHT. packages/components/src/renderers/basic/record-picker.tsx (blob d11d0e1fcc…), its pin record-picker.busReread-10853.test.tsx (4efdf9671c…) and .changeset/10778-dashboard-bus-readers.md (6da9d731a0…) are byte-identical between ac526b1 and a13048c; the previous record's judgments 3 (the picker), 4 (the picker pin), 5 (the 10778 half) and 6 carry forward. The useObjectsInvalidation hook and the 'load' path (judgment 1 and the first-load half of 2) are also untouched by the delta.

② Semver level

patch for @object-ui/plugin-form and @object-ui/components is right and the frontmatter is unchanged: the delta adds a render-time ref and three checks inside one component, exports nothing, and moves no type, spec key or registry input. No major (AGENTS.md §9 版本号策略).

③ Boundary flags

• Pre-existing, not this card's class: a grid action on the editor's own collection while the editor is open (Remove row, Duplicate row, a cell edit under a sort_field, which maps every row to a new object) changes expandedRow and shifts rowIdx, so the editor resets and Apply maps by the new index. That is setRows from the grid, identical at base; the bus path no longer does it.
• A 'load' (a record change, an adapter change, or a details re-author through resolvedEntries) still commits under an open editor and resets it, as at base; the changeset's "never reset by a re-read" is about the bus re-read, which is what the card is about.
• After a save made with the editor open, the linesSavedRef mark is cleared without a replay; the replay on Close rests on the baseline advance making linesUnsaved false. It holds today; were childRowsAfterSave ever to skip the advance (its original !== diffedAgainst identity check), the held re-read would wait for the next save or revert rather than be lost.
• expanded keeps a removed collection's id after a details re-author; nothing reads it.
• The parent-link restatement is now stated truthfully in the doc, the changeset and the body; user input is unaffected.
• Carried from the previous record: the picker's "no label" face is base behaviour for a value outside its options; with the console's mutation bridge a save's own echo is held and replayed once per written collection, not pinned.
• CI as read: poll 1 at 2026-09-28T01:23:09Z — 43 check-runs, 40 success, 3 skipped (Test (coverage), Test (coverage shard …/4), dependabot), 0 failures, 0 in progress. Poll 2 at 2026-09-28T01:30:02Z — identical. Successes: Test, Test (shard 1/8) to (shard 8/8), Test (dist pins), Type Check, Lint, Build & E2E, Live E2E (informational), Build Docs, Bundle Analysis, Changeset Declaration, Changeset Bump Policy, Changeset Fixed Group Check, Changeset Claim Re-read, Changeset Overwrite Report, Line Citation Gate, Control Byte Scan, Governed Surface Queue Guard, Spec Main Shape Gate, Skill Example Check, Skill Guide Path Check, Skill Eval Token Check, README Export Check, Doc Snippet Type Check, Doc Fence Language Check, Doc Example Id Check, Doc Component Type Check, Internal Docs Link Check, Docs Route Eager Closure Check, Shell Escape Residue Scan, Action Ref Convention, Inert vi.mock Specifier Check, Pre-Install Import Graph Check, label. No annotations to read: no failure.
• The pin run's seconds are shared-box figures (the lock excludes other locked runs only).

Implemented-by: claude/issue-10853-master-detail-picker-bus-readers
Reviewed-by: session_01MCg3z5cGrV51xEddpGcxEY

VERDICT: PASS