[PM seat] domain:ui · seat 2 — 🟢 os-sales · session_01JG2jy8 · R1 · batch 3 · 1 landed (objectui#11382)

[os-tesla]

Seat registration post for domain:ui seat 2 @ objectui. Job description: objectstack:.claude/skills/pm-dispatch/references/lanes/ui.md. The body is authoritative; the title and assignee are derived views. Only the sitting seat PM writes it. Body refreshed 2026-10-01T14:59Z on a maintainer stance (round 1). The previous revision (the stand-down write of session session_011p7ikEivgXefNDaE5S5Uec) is the archive for sections 2–4, which carry over unchanged.

1. Current PM

status	🟢 seated: GitHub account os-sales (as GET /user answers it), session session_01JG2jy8a9su7ia4Hx7zxv42, seated 2026-10-01T13:47Z on the maintainer's summons /pm-dispatch ui 2. The stand-down brief 5927807388 was the latest event, so the seat was taken at once.
previous occupant	GitHub account os-support-ai, session session_011p7ikEivgXefNDaE5S5Uec. It stood down on the maintainer's word (brief 5927807388) with nothing in flight.
maintainer's word this session	「seat 1 also has no tokens left; take over 11382 too」 (verbatim: 「seat 1 也没有 token了。11382 你也接手」). Done: objectui#11357 / PR objectui#11382 was taken over (claim 5932833214), accepted (5932883731) and landed (see the ledger row).
maintainer ruling (contract-review faces)	「我已经同意了，不需要严格的规则」, given in this session's chat on 2026-10-01. It answered the seat's proposal 「changeset 散文由席位在 ACCEPT 自核，不起 fable」. So .changeset / CHANGELOG prose is checked by this seat in its ACCEPT, sentence by sentence against the diff, and no isolated CONTRACT_REVIEW_TIER subagent is spawned for it. The isolated at-tier review stays on Clause-②: yes, packages/spec/src/** and governed text. The skill text change is filed as objectstack-ai/objectstack#21187 for the skills seat.
stance (this session only)	「准备发版本，暂时只处理 p0 p1 p2 的卡」, given in this session's chat on 2026-10-01: a release is being prepared, so candidate selection takes only priority:p0 / p1 / p2 cards. p3 and unprioritized cards stay in pm:queue, untouched. The three cards in flight (p1, p2, p2) are inside it. Like every stance, it lapses with this session.
batch	the lane default, 3.
wake-up	the self-bound hourly Routine trig_0146NkrbAV8DqGWP3BLJi7ra (minute 46).
PR subscriptions	none held. subscribe_pr_activity answered "Could not subscribe to this PR" twice on objectui#11382, so the seat watches its PRs with an in-session poll of the PR state and timeline, and the hourly Routine re-reads them.
in flight	Three os-dev subagents, which fill batch 3 of 3: objectui#10872 batch 10 (p1, the flat-props refusal half; claim 5933064114, branch claude/issue-10872-flat-props-refusal), objectui#10993 batch 4 (p2, the closing batch, Clause-②: yes; claim 5933074589, branch claude/issue-10993-batch4-ts-face), and objectui#11358 (p2, a declared gate that cannot be evaluated fails closed; claim 5933280307, branch claude/issue-11358-declared-gate-fails-closed).
charter	check-harness-current.mjs: CURRENT at objectstack origin/main fbcc05f400. Every harness path is in the shared checkout's HEAD.

2. Inheritance ledger

This seat's landings (session session_01JG2jy8a9su7ia4Hx7zxv42):

• objectui#11357 → PR objectui#11382, taken over from seat 1, merged as 3ae919307 at 2026-10-01T14:10Z. All 5 files on origin/main are blob-identical to the landing head 25db06c522, and the co-author trailer is model-free. Fixes closed the card, and pm:dispatched was stripped in the following write. The at-tier contract review 5933224786 (changeset prose) came back PASS. It names one loose changeset sentence ("a key neither editor shows": the metadata editor's JSON source tab does show requires) for whoever compiles the release notes. That is an Acceptance note, not a card.

In flight: none. No open card carries a live Seat: domain:ui#2 claim, and no open card is assigned to os-support-ai. Every dev and review subagent of the session has returned, and no cloud session was created.

Closed this round without code:

• objectui#10107: the probe re-run on 17.5.0 found outcome (d), explain and PATCH agreeing in 108 of 108 cells, with a pre-fix control at 32 disagreeing. Close note 5906835929.
• objectui#10750: its coordination had already landed in PR objectui#11086, and A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition objectstack#15429's engine reads mode at 17.5.0. Close note 5907036956.

Running ledger: 103 PRs landed on this seat this shift. Each was verified by content on origin/main: every file blob-identical to the PR head (or its own hunks identical where main moved the file), the model-free co-author trailer, and closed cards stripped of pm:dispatched. Every Fixes card is closed by the seat's close-if-open step when Fixes leaves it open; since objectui#11204, Fixes has closed some on its own. This round:

• objectui#11189 → PR objectui#11202 (0389650f3, 3 blob-identical). It was dequeued once by the external Spec Main Shape Gate red, then re-armed on the same head after objectui#11206's fix.
• objectui#11199 → PR objectui#11215 (f8334f877, 3 blob-identical; the body was corrected before the merge).
• objectui#11173 → PR objectui#11214 (52bf34824, 4 blob + 2 own-hunk identical). It was re-queued once after an API 502 in the Test rollup gate (5912387790).
• objectui#11190 → PR objectui#11219 (ed498ac91, 4 blob + 1 own-hunk identical; filed objectui#11220).
• objectui#11195 → PR objectui#11217 (1263e405d, 23 blob-identical; after review round 1 FAILED on an undeclared widening).
• objectui#11204 (p1) → PR objectui#11230 (395f4fa51, 3 blob-identical; Fixes closed it this time; filed objectui#11232).
• objectui#11196 slice 1 → PR objectui#11224 (8741cb71a, 19 blob + 1 own-hunk identical; after review round 1 FAILED on a conflict with main, so no CI had run).
• objectui#11201 stage 1 → PR objectui#11225 (cb2f6fb5b, 6 blob + 2 own-hunk identical). Released to pm:queue + pm:retriage (5915594509); triage then handed stage 2 to the maintainer (5915782920).
• objectui#10783 → PR objectui#11238 (b32e7debc, 8 blob-identical; Fixes closed it; filed finding(spec): an inline grid column is judged on only one of its carriers — FormViewSchema.subforms[].columns is z.array(z.any()), and a currency column's scale reaches no refusal when its type comes from the child field objectstack#20901).
• objectui#10872 (p1) batch 7 → PR objectui#11249 (32dcf2965, 2 blob-identical): the properties.children runtime pin plus a measured pull reading. It was released to pm:queue (5917083739), and the next batch declares responsiveStyles. Pointers: objectui#5250 5916790889 and objectui#11183 5916799642.
• objectui#10948 → PR objectui#11247 (68c9ca721, 13 blob + 1 own-hunk identical; Fixes closed it). Review round 1 passed with a README note; the dev then added the docs, and round 2 passed on the landing head.
• objectui#9843 → PR objectui#11246 (741864f7b, 12 blob + 1 own-hunk identical; Fixes closed it). Round 1 FAILED on shard 1 (the objectui#9784 pin's spelling), fixed inside the claimed surface. The PR's pnpm-lock.yaml plugin-grid hunk was already on main via PR objectui#11243, and the end state carries all three ^17.5.0 floors. Filed objectui#11254 and objectui#11255.
• objectui#8069 → PR objectui#11233 (af9e9572c, 21 blob + 13 own-hunk identical; Fixes closed it). Round 1 FAILED; option B with a fork-stop against objectui#7530 (it did not fire); round 2 PASSED. Filed objectui#11256 (the governed skill sentence) and objectui#11262 (the ADR-0137 D4 successor).
• objectui#8894 step 1 → PR objectui#11257 (dae76a456, 3 blob-identical; Part of). The door pins landed. Step 2's premise was measured false, so the card was released to pm:queue + pm:retriage with the fork for triage (5918585815), and objectui#11261 was filed.
• objectui#11196 slice 2 → PR objectui#11259 (02a22957c, 14 blob + 1 own-hunk identical; Fixes closed it). This freed StudioDesignSurface.tsx; seat 1 claimed objectui#11232 on it.
• objectui#11223 → PR objectui#11267 (7d074baae, 8 blob-identical; Fixes closed it). It landed after objectui#10872 batch 8 as noted. The page:card footer gap in the spec's exported walk is filed as spec(ui): walkAddressedPageComponents skips page:card's properties.footer as a back-compat spelling, but PageCardProps.footer is a declared, rendered slot, so nodes there go unjudged; the three platform page walks disagree on positions objectstack#20940; objectui#11170 waits for it.
• objectui#11242 → PR objectui#11268 (18d1a0abd, 7 blob + 1 own-hunk identical; Fixes closed it). The global-runner residual is ruled A, with B named as its fix; pointer on objectui#4421 5920153819. This freed ActionRunner.ts for objectui#10929.
• objectui#11234 → PR objectui#11269 (52aad5cef, 20 blob + 2 own-hunk identical). ObjectKanbanSchema.onQuickAdd is retired and an internal board takes the quick-add pair; the body was corrected through the relay before the merge (ACCEPT 5920016891).
• objectui#9217 → PR objectui#11271 (1cb3732ce, 5 blob-identical). Ruling B (5919298624): the flow policy renders the platform's reason verbatim on its presence, muted, never parsed (ACCEPT 5920018476). Its FlowStatusDot follow-up is filed as objectui#11281.
• objectui#11261 → PR objectui#11278 (b4333ab8a, 4 blob-identical; Fixes closed it). A dimensionless table / pivot or bar / line / area / combo widget renders every measure. The report's two open questions (nine unnamed chart types still drop measures; the warning's metric-only narrowing rests on a false premise) went to objectui#8894 with pm:retriage (5920992338). The PR body was corrected before the merge.
• objectui#11252 → PR objectui#11279 (7fed09d07, 9 blob-identical; Fixes closed it). The last three Select… placeholders read their locale keys, and an enumeration pin closes the family.
• objectui#11251 → PR objectui#11277 (1b30c0fe0, 3 blob-identical; Fixes closed it and objectui#10239). The generic metadata editor renders a view's served top-level union through the one existing resolver, which now scores a bare nested union member by member and ranks same-kind members by contradictions. objectui#10239's scalar-arm guard rides it.
• objectui#11262 → PR objectui#11283 (58da8aeca, 12 blob-identical; Fixes closed it). ADR-0137 D4's three remaining blank-gate paths now report through the one diagnosed guard, with no verdict moved, and the two mirror keys refuse a blank (minor on @object-ui/types). The family's next member (the pred === '' pre-screens) is recorded, not filed, with the measurement that would file it named in ACCEPT 5921492233.
• objectui#10929 → PR objectui#11282 (e2271568f, 5 blob-identical; Fixes closed it). The served 17.5.0 action schema strips the aria tombstone, so Studio offers no aria without an editor change. @object-ui/core's ActionDef drops the dead mirror, minor with the break spelled out, after review 5921429109 ② and delta 5921635685.
• objectui#11201 stage 2 → PR objectui#11297 (1ccb5ba7d, 5 blob-identical; Fixes closed it). Under ruling B (5921167691), a present nav label renders verbatim with no translate-if-equal exception, and an inline locale-map label renders at the spec resolver's floor. The follow-up family is objectui#11299.
• objectui#10859 batch 5 → PR objectui#11309 (997ce38cb, 26 blob + 3 own-hunk identical; Part of). An authored object-map takes the spec's properties bag, the flat spelling is refused by name, and seven producers were rewritten. The maintainer's record-source rule is kept and read in the bag. The card was released to pm:queue for batch 6 (object-gantt, carrying the measured dataSource-only refusal).
• objectui#11276 object-chart batch → PR objectui#11328 (5262f7dd3, 13 blob + 4 own-hunk identical; Part of). With no spec row, the bag is judged by objectui's own mirror members by reference. Its PR body's stale base sentences were corrected before the merge. The card was released to pm:queue for object-grid (5924879598).
• objectui#10859 batch 6 → PR objectui#11337 (db0beb2aa, 20 blob + 3 own-hunk identical; Part of). An authored object-gantt takes the spec's properties bag; this was the last flat-arm block under 5917261346. The landed head b75d89daa merged main (objectui#11330's gate fix) into the ACCEPTed 0c673ccb8, with all 23 PR files blob-identical. Released to pm:queue (5926748567): the governed page-builder gantt example, the designer six, and the (ii)/(iii) keys stay on the card.
• objectui#11322 → PR objectui#11350 (5638529e6, 14 blob + 1 own-hunk identical; Fixes closed it; stripped). The grid's selection bar asks the action family's one declared-gate definition, so a blank-gated bulk action shows and runs as on the rows. The built-in Delete keeps main's answer: triage's "Not A" (5926391467) on the seat's retriage question 5926251271 withdrew the core half of the first direction. Filed objectui#11358 (hasDeclaredPredicate's silent ast-only / non-predicate answer, measured through the unmirrored object-grid bulkActionDefs door).

Lane queue at stand-down (read at 2026-10-01T08:33Z): 43 open domain:ui cards; 13 dispatchable (pm:queue, unassigned, without pm:retriage), none of them p0 or p1.

• The p1 objectui#11326 (a red row in the P0 release sweep), which this seat left for the next seat, has since been claimed by another seat (os-litant).
• p2, this seat's families:
  • objectui#10859, released after batch 6 (5926748567). Left on it: the governed page-builder.md gantt example (a governed draft PR, after maintainer approval), the designer six (5861220023), and the (ii)/(iii) keys (5861157913).
  • objectui#11276: object-grid is its next batch. It touches objectql.zod.ts's union and the census pins, so cut it from main (PR objectui#11337 has landed).
• objectui#11211 is claimed by seat 1 (os-warren).

Decision box (needs-user-decision, objectui, domain:ui): empty at 08:33Z. Held: 6 domain:ui + pm:blocked and 16 pm:on-hold, read at 08:33Z.

Untriaged findings: objectui#11358, filed by this seat at 08:03Z from PR objectui#11350's acceptance note (a) on triage's instruction (5926391467 point 3), with the review's addendum 5927514000. It awaits triage's first grade.

Filed this shift: objectui#11035 (an activity row's Add reaction writes sys_comment by a sys_activity id) · objectstack-ai/objectstack#20558 (sys_comment.reactions described as an array, stored as a map) · objectui#11037 (a week/day drag of an overnight event shortens it) · objectui#11040 (the comment-mask sweep races a tsup temp file) · objectstack-ai/objectstack#20571 (the fleet relay's issue_patch replaced multi-byte characters in a ~41 KB PR body with U+FFFD, deterministically) · objectstack-ai/objectstack#20578 (the save-gate port of objectui#11008's bind / hidden, Blocked-by: objectstack-ai/objectui#11008) · objectui#11051 (element:* write aria.ariaLabel as aria-arialabel) · objectui#11056 ("Change undone" hard-coded English) · objectui#11058 (DefaultLoginPage maps no sign-in refusal code; now claimed) · objectui#11061 (modalCloseButton declared, never read) · objectui#11062 (the console never arms dashboard auto-refresh) · objectui#11067 (Studio's 「Platform event」 start option writes a trigger kind no engine routes) · objectui#11068 (eight declared ObjectGridSchema keys ObjectGrid never reads) · objectui#11075 (the generated JSX types fail tsc over the real public manifest: TS2430 on RecordAlertProps) · objectui#11079 (a gantt-variant timeline draws its bars one column narrower than its headers) · objectui#11083 (the nested aria family: the page root drops it; ListView and useRecordAriaProps onto the shared reader) · objectui#11085 (the flow designer's scope warns on references the engine writes: the Problems panel's connector outputs, an edge guard's source outputs) · objectui#11100 (Studio's app preview mis-renders a locale-map app, nav-item or group label) · objectui#11106 (record:path with a statusField and no stages renders "no stages configured"; the promised derivation from field metadata runs only in the synthesized default page) · objectui#11112 (a gantt-variant timeline reads a date-only end as the start of that day: an inclusive end draws a day short, a same-day task zero-width) · objectui#11122 (Undo of an undoable update that writes a lookup sends the expanded related record back as the value; measured, and the published skill example ReassignLeadAction is a named producer) · objectui#11128 (the app designer's canvas names a locale-map nav item "Item N", and an inline rename writes a string over the map) · objectui#11131 (the record page's line-items panel reads Line Items / Save / Saving… and the grid's Add line / No items yet… in English under zh-CN; measured) · objectui#11132 (the master-detail form's Subtotal / Tax / Total prints a hard-coded ¥; named producer app-showcase's invoice) · objectstack-ai/objectstack#20694 (ComponentPropsMap['object-grid'] lacks description / emptyState, which ObjectGrid reads since objectui#11130) · objectui#11136 (Studio's Interfaces pillar hands its page inspectors readOnly={false} on a read-only package while its autosave discards the edit) · objectui#11140 (a multi-value select drops an authored placeholder; the catalog example teaches it) · objectui#11141 (plugin-gantt reads a date-only end as that day's start, so its catalog plans draw a day short; option A: one end rule in @object-ui/core) · objectui#11144 (a master-detail child with inlineColumns derives no sort or amount field: the showcase invoice's drag order and total are lost) · objectui#11145 (the line-items family's remaining chrome, English under zh) · objectui#11148 (Studio's nav-item inspector shows a map label as [object Object] and overwrites the map) · objectui#11153 (the Interfaces pillar deletes its own ?sel=nav:ID deep link on mount; its apply effect also enters editing without reading readOnly)) · objectui#11158 (the Interfaces pillar's nav rail crashes on a locale-map nav-item label: NavTree renders it raw, resolveSurface stringifies it; measured through the real InterfacesPillar)) · objectui#11160 (the line-items family's last English under zh: GridField's invalid-cell LABEL is required at three sites, and MasterDetailForm's three config hints)) · objectui#11165 (slotted-pages' header example writes eyebrow and icon on a page:header, which the contract refuses) · objectui#11166 (seam: page:header's breadcrumb, default true, draws only an empty slot; enforce or remove)) · objectui#11167 (the Interfaces rail keeps nav editing open after a late read-only answer, so edits are never saved)) · objectui#11173 (the Studio page designer's "Show breadcrumb" toggle still writes page:header.breadcrumb, a producer left standing by objectui#11166's retirement) · objectui#11189 (Studio's nav autosave drops a shown edit: Done inside the debounce, or an Add while a save is in flight; measured; family closure) · objectui#11190 (the Studio flow screen preview drops options / placeholder / defaultValue; named producer app-todo task.flow.ts) · objectui#11201 (question for triage: does 「present ⇒ verbatim」 retire the nav renderer's machine-name i18n convention?) · objectui#11196 (the designer nav surfaces show a label-less entry blank or unnamed; 24 reads) · objectui#11197 (NavigationItemTypeSchema omits the spec's doc) · objectui#11198 (domain:skills: the governed app-composition.md still says label is required) · objectui#11182 (action:menu's trigger stays pressable while its action is in flight: {...rest} spreads disabled: undefined over disabled={loading}; measured; serial behind objectui#11168 slice 1) · objectui#11203 and objectui#11204 (the autosave family; graded p2 / p1) · objectui#11220 (see above)· objectui#11232 (the autosave family closure: a leaf switch inside the debounce writes the previous item's document into the newly opened one) · objectstack-ai/objectstack#20901 (subforms' columns typed z.any, plus the hydrated scale) · objectui#11254 (the metric-tile half of ruling A′, objectstack-ai/objectstack#19628, with the rest of its "no ?? N" list) · objectui#11255 (the percent docs and catalog teach precision as the decimal width) · objectui#11256 (domain:skills: schema-expressions.md says a broken predicate "never blocks submit") · objectui#11261 (a dimensionless table / chart widget with 2+ measures renders as a one-number tile; the seam behind objectui#8894's fork) · objectui#11262 (the ADR-0137 D4 successor: three silent blank-gate paths) · objectstack-ai/objectstack#20940 (the spec's exported walkAddressedPageComponents skips page:card's footer, so a nested child there is never judged; from PR objectui#11267) · objectui#11299 (the nav entry I18nLabel family after objectui#11201: objectui validate refuses a map label, the renderer shows its en floor, three resolver props are inert) · objectui#11281 (objectui#9217's follow-up: Studio's Automations rail titles a policy-unbound schedule flow "no trigger (run manually)"; serial after objectui#11272) · objectui#11181 (the label-map family's closure card: Studio still reads an app's own I18nLabel through String() at four sites, and the rail heading prints [object Object] · Navigation; measured) · objectui#11358 (hasDeclaredPredicate answers an ast-only or non-predicate gate "not declared" with no report; measured through the unmirrored object-grid bulkActionDefs door; ADR-0137 D4). Recorded, not filed (no measured reach): slice 3's external/ExternalDatasourcePanel.tsx keeps its translated "unavailable" sentence in error state across a language switch; the fix shape is PR objectui#11134's marker-in-state (ImportObjectDialog). ACCEPT 5897159066 has it. Also recorded, not filed (read at source, reach unmeasured), from PR objectui#11137's review: the runner's verbatim Undo capture where no field map exists (retargeted related-list rows, PageView's runtime, a bare ActionProvider); FILE_REFERENCE_TYPES resolved in place on read; other _rowRecord readers reading a relation as the record (ACCEPT 5898225050). From PR objectui#11133 (unmeasured): whether the engine binds bare trigger fields for a START node's out-edge guards. From PR objectui#11142 (no producer): a fixed-currency amount on a fully configured master-detail entry reads the tenant sign in both the line cells and the stack. From PR objectui#11151 (no producer): a fully configured master-detail entry (every inline column typed) derives no sort or amount field; file it once a typed-inlineColumns producer is named. From PR objectui#11154 (unmeasured): a form view's per-field placeholder override never reaches a widget, because widgets read the object field carrier. From PR objectui#11155 (unmeasured): plugin-timeline's DAY_AXIS columns after a viewer-local skipped-midnight day start at 01:00 (localDayStart, then setDate keeps the hour); the public-door measurement is a timeline view spanning 2026-09-06 rendered with the viewer zone America/Santiago. From PR objectui#11156 (unmeasured): ActionRunner.handlePostExecution pushes an undoable result onto the global undo stack only inside its toast branch, so a runner with no toast handler, a silent result, or an authored toast.showOnSuccess: false on an undoable action offers no Ctrl+Z; the public-door measurement is such a metadata action run through the console. From PR objectui#11247 (objectui#10948, read at source, reach unmeasured): the flow inspector's reference-kind fields and list editors have no programmatic label, and no aria-required. The measurement that would file it: render the production FlowNodeInspector on a create_record node, find the Object control by its label text, and read its aria-required. Also from PR objectui#11247 (read at source at objectstack 0f6dcac5e; no 17.5.0 server could run here): saveMetaItem runs the FlowSchema gate on a ?mode=draft save, so a flow holding a fresh http node is predicted to be refused at nodes.N.config.url, and the designer holds that 422 with the draft kept in the editor. Measure that save against a live 17.5.0 server before any card on whether draft saves should skip the executor-contract refusals. From PR objectui#11149: registry-inputs-spec-parity.test.ts:2558 still says fields is "the submitted set" (test prose, no gate; the file is PR objectui#11086's), to be fixed by its next editor. Carrier note on objectui#10859 (5885358145: object-form's properties bag refused).

Seat errors, recorded:

• Clause-② declared wrong again (objectui#10929, claim 5920190491). The claim said no, but the built diff removes a member from @object-ui/core's public ActionDef and an element from ACTION_DEF_KEYS, a narrowing. Review 5921429109 ② caught it. The line becomes no (narrowing), and the changeset becomes minor with the break spelled out, before landing. This is round 10's lesson again: a retirement that deletes a typed member is a narrowing even when the member could only be undefined.

• Relay-opened PRs were not subscribed (this shift, up to 22:5xZ). execution-duties.md 〈收集〉 says a report that names a PR is subscribed at once (subscribe_pr_activity) and recorded here, because a relay-opened PR is never attached to the session automatically. landing-operations.md says the same for the landing window. The seat followed its landings by polling scripts instead and did not subscribe. The maintainer pointed it out, and PRs objectui#11277, fix(plugin-dashboard): a dimensionless table renders a row of every measure and a dimensionless chart one mark per measure, as the spec states (objectui#11261) #11278 and fix(app-shell,components,console): the three remaining select placeholders read their locale key, not a hard-coded English "Select…" (objectui#11252) #11279 were subscribed at 22:5xZ. Each is unsubscribed when it merges, and none is left at the stand-down.

• A usage wall (about 19:50Z to 20:30Z). The account's session limit stopped all four dev subagents mid-run; the reset was 20:30Z. The 20:32Z Routine fire coincided with the reset, so no extra timer was set. Each dev was revived from its transcript at 20:33Z, and none redid pushed work. objectui#11232, which this seat had prepared, was claimed by seat 1 during the wall; this seat dispatched objectui#11251 instead.

• objectui#11223's claim raced a serial note (18:49 to 18:50Z). Seat 1 posted 5917593110 ("waits for objectui#10872 batch 8") at 18:49:46Z. This seat's guard read the thread just before that, and the claim went out at 18:50:43Z. The dev was told to keep off the arms files; its surface measured disjoint. The card lands after batch 8 (5918271824). Lesson: re-read the thread once more right before the claim POST, not only before the label write.

• A container restart (about 18:36Z) stopped all five dev subagents. Each was revived from its transcript with a resume note, and none redid pushed work. One locked test run (objectui#8069) survived and was waited on, not duplicated.

• Clause-② declared wrong on two claims (round 10). objectui#11195's claim said no, but the diff widened two published packages (a new root export and optional context members). Review 5912757969 failed on it, and minor was owed. objectui#8069's claim said yes (narrowing), but by clause2-line.mjs that means a diff that widens one surface and narrows another; the ruled change only narrows. Both were amended on the card (5912793723, 5912804042). Lesson: read the four arms in objectstack scripts/pm/clause2-line.mjs when writing a claim, and re-judge the line against the BUILT diff before the review.

• A ruling read too wide (objectui#8069, 16:1xZ). The seat meant to answer the dev's open question (refuse a stored blank field visibleWhen, or keep it as "no rule") from Q2's "no verdict changes". Review 5915380177 failed the head on it. Q2 is ADR-0137 D4 and covers GATE slots only. For the field-rule triad, D1 and D2 refuse a blank ("A blank predicate takes this path too, wherever one is already stored"), and "blank means no rule" is an alternative the ADR explicitly refuses for that triad. Lesson: answer a dev's open question from the governing text's own sentence for the exact slot, not from a neighbouring ruling's summary line.

• A slow run misread as hung (12:33Z). The seat told the objectui#11196 dev that its 43-minute vitest run looked hung, because the log had printed one file in 31 minutes. The dev's bisect showed it slow, not stuck: vitest's default reporter prints nothing for passing files when stdout is not a TTY. The run held the shared verify lock throughout, and the other devs queued behind it. Lesson: judge progress by CPU and a verbose reporter, not by log silence. Orders now ask for heavy directories in chunks under timeout 300, with the verbose reporter.

• objectui#11005's ACCEPT and landing (03:2xZ). The landing ops went out in the same script as the ACCEPT, ungated on its exit code; the ACCEPT was refused on a stamp-contract check (a test-case instant written as a timestamp) and posted a minute later, before the merge, naming the slip (5883018877). Lesson: a landing is sent only after the ACCEPT's post exits 0.

• objectui#10964's claim (03:2xZ to 03:35Z). The label write was the 40th write of the hour and the claim the 41st, refused by write-pace; the card sat labelled and assigned with no claim for about ten minutes, until the window reopened. Lesson: the 40-an-hour budget is shared by every agent in this container, so read write-pace.mjs --status before a claim pair and start none that cannot finish.

• objectui#10982's claim (04:35Z). Its serial line read objectui#10993's in-flight surface from that card's claim, not from its pushed branch, which already edited packages/plugin-view/src/ObjectView.tsx (the form.title / form.description read, outside that claim's surface). The hunks are disjoint (:69, :985, :2480 against :1206 and the non-grid loading state); both devs were told, and the one that lands second merges main. Lesson: read an in-flight card's surface from its branch's three-dot diff, not from its claim.

• objectui#11002's dispatch (05:58Z). The order assumed ConsoleShell gates on useDiscovery(); the dev measured that false and found no advertised signal at all, and stopped at the order's valve with nothing changed. The seat escalated the card instead of dispatching a second time.

• The write budget. Four ACCEPT pairs, five cards and two claims in one hour held the shared 40-an-hour token at the ceiling twice (06:29Z, 06:41Z). Landing now runs through budget-gated scripts that post each ACCEPT and its landing ops as a pair only when two slots are free, gate the landing on the ACCEPT's exit code, and hold every card number until the board shows it.

• PR objectui#11187's assignee (08:3xZ). The objectui#11167 dev's label-write --assign on its PR was refused by that subagent's permission layer (an auto-mode classifier). The seat did ⛔ not re-route the write, because re-routing a refused write through another identity is permission laundering. It surfaced the refusal to the maintainer, and the PR landed unassigned. No landing gate reads the PR assignee.

• objectui#11181's claim (09:02Z) missed triage's ordering pointer 5907732627. The claim's Thread-read: was the drafting-time read, 07:35Z. The seat ran objectui#11189 (p1) in parallel by named region rather than recalling a pushed dev, and recorded it on the card (5908498158). Lesson: re-read the thread at claim time, not at drafting time.

• objectui#8069 (09:34Z). The candidate was read from a 04:xxZ snapshot, and seat 1 had claimed it at 09:06Z. claim-batch.sh's pre-check refused the claim, so nothing was written.

• PR objectui#11180's squash message carries a PR-body paragraph that was false at its landing head. The correction lives only in comment 5906747518, because objectui squash-merges with the body. Since then, the seat corrects a false PR body before landing through the relay's issue_patch (read-back verified after finding(fleet-write): an issue_patch body of about 41 KB comes back with two multi-byte characters replaced by U+FFFD, about 16 KB apart, identically on a re-send — and dispatch.mjs reports success without comparing the read-back objectstack#20571's fix). This was done on PR objectui#11192.

• Container restart (about 00:01Z, 2026-09-30). The restart killed two dev subagents (objectui#11148, objectui#11145) mid-verification, and the two merge watchers. No write was lost: both devs had pushed, objectui#11080's ACCEPT had posted (5901343062), and both PRs were already in the merge queue. The devs were resumed from their transcripts and the watchers re-armed.

• An unexplained merge-queue hang (02:23Z). PR objectui#11164's first merge group hung in Test (shard 8/8) until its 20-minute job timeout, and the job log is unrecoverable (blob 404). The dev's patch round (5903675431) did not reproduce it: the exact merge tree's shard 8 passed twice, and the PR's pins hash into other shards. The seat re-queued it once, unchanged (5903691501), and it passed. It is recorded as unexplained, not as a flake. The CI keeps no per-file trace of a stalled shard. A progress reporter is a ci.yml change for the maintainer or the gate lane, and it is not filed.

3. Hot-file serial queue

• packages/types/src/zod/public-blocks.zod.ts and the finding(types,cli): objectui validate refuses 79 of the 229 registered component types at type — AnyComponentSchema has no arm for them, ai-form-assist (taught by the plugin-ai README) among them #10859 ratchet test: free since PR objectui#11180 landed. Draft PR objectui#11069 edits other hunks of the ratchet test.
• packages/types/src/objectql.ts, zod/objectql.zod.ts, __tests__/zod-mirror-parity.test.ts: open draft PR objectui#11125 (another session; PR objectui#11115 landed as b0a05dda1f). objectui#11068, objectui#10859's next batch, and objectui#11013 (its allowExport declaration at objectql.ts) wait on them.
• studio-design/StudioDesignSurface.tsx: seat 1's objectui#11272 (p1, the load window and package switch) holds it. objectui#11281 (the FlowStatusDot title) rides after it.
• packages/core/src/actions/ActionRunner.ts: free since PR objectui#11282 (objectui#10929) landed.
• packages/plugin-dashboard/src/DatasetWidget.tsx: free since PR objectui#11278 (objectui#11261) landed. objectui#8894's next claim (the warning's narrowing and the authoring pickers) waits on triage's answer to retriage note 5920992338.
• packages/i18n/src/locales/*.ts: free.
• RecordDetailView.tsx's feed region, plugin-detail's feed renderers and react's DiscussionContext: free since PR objectui#11217 landed.
• The form submit and blank-gate paths (form.tsx, console FormPage.tsx, WizardForm.tsx, sectionFields.ts, core's evaluator, types' form.zod.ts): free since PR objectui#11233 (objectui#8069) and PR objectui#11283 (objectui#11262) landed. The family's next member (the pred === '' pre-screens) is recorded in ACCEPT 5921492233, with the measurement that would file it.
• app-shell/src/hooks/useNavigationSync.ts and the nav-entry producers: free since PR objectui#11225 (objectui#11201 stage 1) landed; objectui#11201 is back in pm:queue for stage 2.
• The flow inspector (FlowNodeConfigField.tsx, FlowNodeInspector, FlowObjectListField.tsx) and clientValidation.ts: free since PR objectui#11247 landed.
• objectui#11021's ListView half (ListView.tsx, ObjectGrid.tsx) is serial behind objectui#9853 / PR objectui#10278 (seat 1). Note 5906922716.
• objectui#10872's next batch: the envelope-key batch, which declares node-level responsiveStyles additively (its pull is lit in PR objectui#11249's reading; release 5917083739). record:line_items is held with no spec row. The flat-props refusal half waits for objectui#11183's docs migration (triage A).
• SchemaForm.tsx (resolveUnionBranch): free since PR objectui#11277 (objectui#11251) landed.
• packages/layout/src/NavigationRenderer.tsx (resolveNavItemLabel): free since PR objectui#11297 (objectui#11201 stage 2) landed. objectui#11299's Q1/Q2 rewrite it and its console call sites.
• packages/types/src/zod/objectql.zod.ts (the authored union, ObjectMapBlockSchema, ObjectGanttBlockSchema, requireRecordSource): free since PR objectui#11337 (objectui#10859 batch 6) landed as db0beb2aa. objectui#11276's object-grid batch is cut from main after it.
• packages/plugin-grid/src/bulkEligibility.ts, ObjectGrid.tsx's built-in bulk Delete call site, and core's listConditional.ts / declaredPredicate.ts docblocks: free since PR objectui#11350 (objectui#11322) landed as 5638529e6.
• packages/layout/src/NavigationRenderer.tsx and packages/types/src/zod/app.zod.ts: objectui#11197 (seat 1). objectui#11196 calls resolveNavItemLabel and does not edit it.

4. Notes (durable)

• check-governed-merges.mjs --pr N and dispatch-gates.mjs gate families answer only for objectstack. For objectui, derive the path surface from the PR's file list; the tier half answers with --tier --repo objectstack-ai/objectui.
• Several domain:ui seats are live at once. Re-read open-PR file lists and live claims before every claim; the board moves in minutes.
• git fetch A B sets FETCH_HEAD to the first ref: fetch into a named ref for any merge-tree probe. Judge a PR by three-dot diff or its file list.
• objectstack-ai/cloud is unreachable from these sessions; cloud-contract cards carry that as a declared reading gap.
• objectui squash-merges with the PR title and body, and GitHub appends every squashed commit's Co-authored-by trailer. A pushed commit that carries a model identifier means moving the change to a fresh branch, because force-push is barred.
• Every write goes through scripts/pm/post-stamped.mjs, label-write.mjs or the fleet relay; the budget is 40 writes an hour per token, and every agent in this container spends the same token (devs' PR, assignee and report writes included). Read node scripts/pm/write-pace.mjs --status before a batch.
• Fixes #N does not reliably close the card on objectui's merge-queue merges. objectui#9830 and objectui#11158 were both still open minutes after the merge, with no closed event on the timeline; objectui#11204 and objectui#10783 were closed by it. The seat reads the state after verifying by content, and closes a card left open itself, with a close note, state_reason: completed.
• A job log is read through the MCP get_job_logs tool; the REST log URL redirects to blob storage the proxy refuses. A failed job is re-run with MCP rerun_failed_jobs (REST answers 403).
• Director reminder 5903848327 (2026-09-30): a contract-review record covers the landing head. Re-record after any base merge, regeneration or changeset-prose edit; the one exception is a Regen-provenance: line. A review face owes a record whatever Clause-② says. Apply it at every enqueue.
• Spec Main Shape Gate is REQUIRED in main's ruleset (read 2026-09-30T11:24Z through GET /repos/objectstack-ai/objectui/rules/branches/main). The workflow file's comment still says it is not. A spec change on objectstack main can therefore dequeue every objectui PR. The standing ruling (objectui#10916, 5866275396) makes such a red lane p0 for any domain:ui seat.
• Cross-lane memos.
  • spec(ui): retire PageHeaderProps.breadcrumb ("Show breadcrumb", default true) — no renderer draws a trail and the console shell already draws one; the spec half of objectui#11166 objectstack#20758 (the breadcrumb spec retirement): the objectui pin-bump PR that installs it must drop the breadcrumb input from page:header's registration, and the "accepts it until objectstack#20758 retires it" sentence in content/docs/layout/page-header.mdx and content/docs/guide/layout.md. nav:breadcrumb is registered only as placeholders.tsx's scaffold (PR objectui#11172's review 5904304282).
  • objectui#11165 (the slotted-pages header example) is domain:devx.