deps: bump objectui's @objectstack/* to 17.7.0 — eight held cards' conditions are met on npm, and objectui's lockfile still resolves 17.6.0

[objectstack-fleet]

Filing gate: ④ a coordination node. One change unblocks many held cards. Filed by the triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015, session_01AavokzJ5DndAwitDXvKy4U) at the unlock scan after @objectstack/spec 17.7.0 reached npm (published 2026-10-06T12:22Z). objectstack-ai/objectstack#21352 merged at 2026-10-06T11:49Z, and the tag commit is 4e4e881427. The shape is objectui#11438's, the 17.6.0 bump. ⛔ Not a claim.

Why one card

These objectui cards are pm:on-hold on "an @objectstack/* release that carries X, and objectui resolves it". Each X was tested against the 17.7.0 tag commit 4e4e881427 with git merge-base --is-ancestor.

card	its condition	17.7.0
objectui#11649	carries f5b8e29e37 (PR objectstack-ai/objectstack#21890)	✅ in tag
objectui#11645	carries 48297ad980 (PR objectstack-ai/objectstack#21833)	✅ in tag
objectui#11545	carries 6dd99b82c3 (PR objectstack-ai/objectstack#21566)	✅ in tag
objectui#11439	carries 39119014dc (PR objectstack-ai/objectstack#21344)	✅ in tag
objectui#11428	carries dcc5ef4c5a (PR objectstack-ai/objectstack#21256)	✅ in tag
objectui#11389	carries 99e1912afc (PR objectstack-ai/objectstack#21342)	✅ in tag
objectui#11334 (p1)	carries 32d5769080	✅ in tag
objectui#11417	carries 32d5769080	✅ in tag

Not released by 17.7.0: objectui#11168 (it waits on a maintainer ruling), objectui#10971 (a producer measurement), objectui#6152, objectui#5144, objectui#3592 and objectui#2763 (each waits on something other than a release).

objectui origin/main still declares "@objectstack/spec": "^17.6.0" in packages/types/package.json, and pnpm-lock.yaml resolves @objectstack/spec@17.6.0 (read at filing). Every row above names "objectui resolves it" as half its trigger. Eight claimants each bumping the lockfile would collide, so this card does it once.

The change

• Update objectui's @objectstack/* dependencies to 17.7.0, and regenerate pnpm-lock.yaml with the repository's own tooling. ⛔ Never edit it by hand.
• Where a package will import a 17.7.0-only export, raise that package's declared floor (^17.7.0). The consuming card may do this instead: say which in the PR.
• The mirror rides this PR. objectui#11334's placement ruling (5956401275) says the PR that bumps @objectstack/spec carries the mirror, re-attaching checkDashboardWidgetChartMeasureArity under its new name by import. objectui#11417 closes with it. ⛔ No second copy of the check.
• Run the full gate set. 17.7.0 also carries changes objectui did not ask for. A gate that goes red on one is fixed here if it is mechanical, and reported here by row if it is not.

One red the bump will meet, by design

This is from the cross-lane memo of domain:spec @ objectui (objectstack-ai/objectstack#6015, comment 6016611286).

• objectui#11624 (landed as 1c2e2c46) pins sortField on the object-master-detail-form details entry as accepted. 17.7.0 carries 6ec54f00, which retires that key, so those rows go red at the bump.
• The bump flips them to refused, with the spec's retired-key message. It also updates the prose that calls the retirement unreleased:
  • the MasterDetailDetailConfig docblock in MasterDetailForm.tsx;
  • the header of masterDetailDetailsMembers-8071.test.tsx and the header of the types pin;
  • the details registration description in plugin-form's index.tsx;
  • the console parity row object-master-detail-form.details.

After it lands

Each row above then has its Restart-when: met. Triage's next unlock scan releases each hold, after re-reading its own condition on the merged ref. ⛔ No hold is released by this card's own act.

---

Generated by Claude Code

[objectstack-fleet]

Claim: PM loop round 1 (maintainer direct dispatch)
Session: session_01XgevVHGcN8dnegiiY3v75V
Account: os-support-ai (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-11717-objectstack-17-7-0
Worktree: objectui-issue-11717
Domain: domain:devx
Seat: domain:devx#2 (objectui#10917, round-open marker 6019163160)
File surface: the @objectstack/* ranges in objectui's package.json files, and pnpm-lock.yaml regenerated by the repository's own tooling (⛔ never by hand). The card's three named riders: the objectui#11334 mirror in packages/types/src/zod/complex.zod.ts, with its census row in spec-object-refinements-7715.test.ts and the three OWED TO objectui#11334 bookings it strikes; objectui#11417's single-series arm pin; the objectui#11624 sortField flips in the five places the card lists. Plus the mechanical fixes for gates that 17.7.0 turns red, each named by row in the PR. A non-mechanical red is reported by row, not fixed. Stop on breach and explain in the report.
Container & model: M, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui printed "no path-derived mandate")
Clause-②: yes
Thread-read: none
Serial constraints cleared: PR objectui#11721 (objectui#11686) also edits pnpm-lock.yaml and packages/plugin-tree/package.json. That is ordinary concurrency: the later lander merges main and regenerates the lockfile with tooling. PR objectui#11600 is the release PR, and PR objectui#11069 (objectui#5250, parked) touches none of this surface. All eight held cards in the body are pm:on-hold with no assignee, so none is in flight. pnpm-lock.yaml is not a SINGLE_CLAIM_PATHS entry.

Clause-② amended from no to yes by this seat, after report 6024525892. The claim delegated the answer to the dev's measurement on the built declaration closure of @object-ui/types. Report 6024525892 and PR objectui#11742 measure yes: the emitted zod declarations inline 17.7.0's shapes, and four .d.ts files differ in declarations; the two chained checks also narrow the runtime accept set. An at-tier contract review is owed before landing, and it is running.

Authority. The maintainer's words in this seat's session chat, verbatim: 「11717 最优先」 and 「11717 如果衍生出相关的阻塞卡片，你也负责派发处理」. Cards derived from this one that block it are dispatched by this seat, each with its own claim.

Placement rulings carried in: objectui#11334 5925828806 (triage: the PR that bumps @objectstack/spec carries the mirror) and 5956401275 (chain the renamed checkDashboardWidgetChartMeasureArity). Precedent: objectui#11438 (17.6.0), rulings A′ 5965062249 and A″ 5968177777.

_{Stamp: 2026-10-06T15:08Z · read against objectui origin/main 7300fca; npm @objectstack/spec latest = 17.7.0 (published 2026-10-06T12:22Z), and the lockfile resolves 17.6.0}

---

Generated by Claude Code

[objectstack-fleet]

This amends my filing body above. Its "Not released by 17.7.0" line lists objectui#6152, and that is wrong. objectui#6152's condition is an @objectstack/spec release on npm that refuses resizableColumns on object-grid. 17.7.0 does: measured at this write, with { resizable: true } as the lit control. So objectui#6152 waits on this card, like the eight rows in the table.

Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-06T16:20Z. ⛔ Not a claim. ⛔ No scope change for the claim 6019218018.

• For this PR: nothing new to do. objectui#6152's four strokes stay on objectui#6152, which is claimed afresh after this card lands.

• A pointer for the claimant: objectui#6152's comment 5961014236 lists, as measured at pin 89cad75d55, the objectui places that 17.7.0's object-grid row (PR feat(spec)!: an object-grid page block types the seven members the grid reads, and resizableColumns retires for resizable (#21445) objectstack#21463) turns red:

  • operations, which becomes strict: read and import are refused, the root README.md example included;
  • conditionalFormatting;
  • apps/console's registry-inputs-spec-parity.test.ts row for object-grid.resizableColumns.

  This card's own rule covers them: fix a mechanical red here, and report a non-mechanical one by row.

[objectstack-fleet]

os-dev-report

{
  "issue": 11717,
  "status": "done",
  "branch": "claude/issue-11717-objectstack-17-7-0",
  "pr": "https://github.com/objectstack-ai/objectui/pull/11742",
  "session": "session_01XgevVHGcN8dnegiiY3v75V",
  "premise_still_valid": true,
  "summary": "pnpm-lock.yaml resolves @objectstack/* 17.7.0 (pnpm-only regeneration, census: only the seven packages move). The card's three riders are done: the objectui#11334 mirror (checkDashboardWidgetChartMeasureArity chained by import, the census row named, and all three OWED TO objectui#11334 bookings struck by flipping rows to the refusal), objectui#11417's single-series arm pinned, PR objectui#11433's door assertions flipped, DASHBOARD_WIDGET_MULTI_MEASURE_TYPES imported in place of a restated list, and the objectui#11624 sortField rows flipped to refused with the spec's retired-key message, with prose updated in the five places. Every other gate 17.7.0 turned red was fixed here as mechanical, by row; checkPageRequiresKind was attached under ruling B1. One red stays, and it is not mechanical: check:eager-closure (Bundle Analysis), vendor-objectstack +182,909 gzip bytes over main's control. Floors: @object-ui/types ^17.7.0, @object-ui/plugin-dashboard ^17.6.0. Clause-②: measured yes.",
  "tests": "(1) Full root suite under the lock on the lockfile-only commit 60ed1ad: Test Files 24 failed | 4458 passed | 7 skipped (4489); Tests 41 failed | 60842 passed | 359 skipped (61242). All 24 are rows in the PR. (2) Full root suite under the lock on the PR head 50ed869: Test Files 4 failed | 4494 passed | 7 skipped (4505); Tests 5 failed | 61065 passed | 359 skipped (61429). Two are CI shard 7's rows, fixed in c7bbce48 (twins-spec-by-reference-9736 Page control; exclusion-reason-truthfulness vacuity row). Two are not this branch's: record-details.hideFieldsUncast-9965 timed out at 15000ms under full load and passes alone; date-only-zone-pins-10183 fails only with this worktree's vitest results cache present. Moving node_modules/.vite/vitest/.../results.json aside, the same head passes 27/27, and a control worktree at main 02a38967 passes 27/27. (3) After the fixes: packages/types/ + app-shell previews/ + the two files, 456/457 files green, the date-zone file red with the cache; the three changed or timed-out files rerun 24/24 green. (4) PR CI on 50ed869: 39 success, 3 failure (Test shard 7/8 with its Test aggregate, Bundle Analysis), 3 skipped. Spec Main Shape Gate, Lint, Type Check, Build & E2E, Live E2E (informational, backend 17.7.0) and the other 7 shards are green. CI on c7bbce48 at report time: 24 success, 17 in_progress, 3 skipped, 0 failure. The CI state is in_progress. No ablation was owed: the reverse proofs are the census and cap rows themselves, which went red on the bump-only tree and green with the mirror.",
  "gates": [
    {
      "cmd": "pnpm install --frozen-lockfile (engine-strict, merged head 25251a80)",
      "exit": 0
    },
    {
      "cmd": "pnpm install (lockfile byte-identical after)",
      "exit": 0
    },
    {
      "cmd": "turbo run build --filter=!@object-ui/site --concurrency=2 (25251a80, 43/43)",
      "exit": 0
    },
    {
      "cmd": "turbo run type-check --concurrency=2 --continue (81/81, on 47b0ff99's tree)",
      "exit": 0
    },
    {
      "cmd": "turbo run lint --concurrency=2 --continue (47/47, 0 errors; 50ed869)",
      "exit": 0
    },
    {
      "cmd": "pnpm exec vitest run (full, 60ed1ad bump-only)",
      "exit": 1
    },
    {
      "cmd": "pnpm exec vitest run (full, 50ed869)",
      "exit": 1
    },
    {
      "cmd": "vitest targeted reruns after fixes (c7bbce48 tree)",
      "exit": 0
    },
    {
      "cmd": "check:lockfile-integrity",
      "exit": 0
    },
    {
      "cmd": "check:lockfile-dedupe",
      "exit": 0
    },
    {
      "cmd": "check:spec-floors",
      "exit": 0
    },
    {
      "cmd": "check:spec-symbols",
      "exit": 0
    },
    {
      "cmd": "check:installed-pin-claims",
      "exit": 0
    },
    {
      "cmd": "check:component-surface-parity (report-only)",
      "exit": 0
    },
    {
      "cmd": "check:doc-snippets",
      "exit": 0
    },
    {
      "cmd": "check:doc-types",
      "exit": 0
    },
    {
      "cmd": "check:doc-examples",
      "exit": 0
    },
    {
      "cmd": "check:readme-exports",
      "exit": 0
    },
    {
      "cmd": "check:new-line-citations",
      "exit": 0
    },
    {
      "cmd": "check:control-bytes",
      "exit": 0
    },
    {
      "cmd": "check:test-path-roots",
      "exit": 0
    },
    {
      "cmd": "check:changeset-claims",
      "exit": 0
    },
    {
      "cmd": "check:pending-changeset-literals",
      "exit": 0
    },
    {
      "cmd": "node scripts/check-changeset-presence.mjs",
      "exit": 0
    },
    {
      "cmd": "node scripts/check-changeset-no-major.mjs",
      "exit": 0
    },
    {
      "cmd": "check:eager-closure",
      "exit": 1
    },
    {
      "cmd": "check:eager-locale-catalogues",
      "exit": 0
    },
    {
      "cmd": "check:skill-examples / prompt-keys / sdui-registration-pins / i18n-keys / i18n-drift / i18n-dead-keys / action-forward-parity / designer-field-key-parity / phantom-deps / unused-deps / registry-bare-names / handler-key-reads / metadata-write-doors / doc-fences / doc-example-ids",
      "exit": 0
    }
  ],
  "line_budget": {
    "additions": 942,
    "deletions": 574,
    "files": 85,
    "against": "merge-base 02a38967 (GitHub reads the same on the PR at c7bbce48); under the 5,000 human-merge line"
  },
  "files_changed": 85,
  "eager_closure_measurement": {
    "instrument": "the console build's apps/console/dist/eager-closure.json (emitEagerClosureReport, gzipBytes per eager chunk), read by pnpm check:eager-closure; control = main 02a38967 built with turbo run build --filter=@object-ui/console... in its own worktree (spec 17.6.0); head = 25251a80 (shipped source identical to c7bbce48), same command family; 331 eager chunks on both",
    "chunk_rows_gzip_bytes": [
      {
        "chunk": "vendor-objectstack",
        "control": 1421565,
        "head": 1604474,
        "delta": 182909
      },
      {
        "chunk": "ui-components",
        "control": 282560,
        "head": 282557,
        "delta": -3
      },
      {
        "chunk": "framework",
        "control": 74652,
        "head": 74670,
        "delta": 18
      },
      {
        "chunk": "src",
        "control": 176055,
        "head": 176052,
        "delta": -3
      },
      {
        "chunk": "i18n-locale-en",
        "control": 44497,
        "head": 44525,
        "delta": 28
      },
      {
        "chunk": "everything else (326 names)",
        "control": 1406387,
        "head": 1406376,
        "delta": -11
      },
      {
        "chunk": "aggregate",
        "control": 3405716,
        "head": 3588654,
        "delta": 182938
      }
    ],
    "ceilings_that_fired": "MAX_EAGER_CLOSURE_GZIP_BYTES 3,410,284: head over by 178,370. PER_CHUNK_GZIP_CEILINGS vendor-objectstack 1,439,157: head over by 165,317 (PER_CHUNK_BASELINE 1,420,958). ui-components, framework and i18n-locale-en pass. The control passes all of them, with 4.5 KB aggregate headroom.",
    "attribution_below_chunk": "Instrument: raw and gzip-9 byte sizes of each ESM entry file in the installed @objectstack/spec 17.7.0 against the npm 17.6.0 tarball. Gzip deltas for the entries objectui's shipped code imports: shared +152,445 (raw +522,843; same 79 exports, while the bundle inlines 71 more strictObject( and 58 more lazySchema( call sites); browser (root) +66,383; browser/system +61,940; ui +42,688; browser/kernel +23,229; api +17,065; browser/data +11,196; ai +3,004; identity +956; security +355; contracts +115; automation -862. migrations grew +203,315, but no shipped objectui import names it (only tests do). Other @objectstack packages are small by comparison: lint +9,125, types +6,089, core +1,497, formula +929, client 0, sdui-parser -33. Prose proxy: 211 long string literals (80 chars or more) new in 17.7.0 are found in the head vendor chunk, 57,905 chars, about 19,830 gzip bytes on their own, roughly 11% of the growth.",
    "not_measured": "Which entries and modules the vendor chunk actually keeps after tree-shaking: the build emits no module-level sizes or sourcemaps for the chunk. So the split of the remaining growth, about 163 KB gzip, is NOT MEASURED. Whether any of it is code the first screen does not need, a lazy-load candidate on the objectui side versus a payback on the objectstack side, is also NOT MEASURED. The shared entry's growth behind an unchanged export list is the strongest objectstack-side payback lead.",
    "handling": "Left red. MAX_EAGER_CLOSURE_GZIP_BYTES, BASELINE, PER_CHUNK_BASELINE and the headroom constants are untouched, per the PM's guidance and the 11088 precedent (raised only on an explicit maintainer ruling)."
  },
  "clause_2": "yes, by measurement. The @object-ui/types dist .d.ts files were compared, with comments dropped and line order ignored, between main 7300fca (spec 17.6.0, its own worktree) and this branch (17.7.0). 13 files differ, and 4 differ in declarations: zod/index.zod.d.ts (2,498 lines), zod/objectql.zod.d.ts (3,324), zod/public-blocks.zod.d.ts (374) and zod/layout.zod.d.ts (54). The emitted zod declarations inline spec shapes in both directions: typed action:group / action:menu members, new enum members (new-tab, separator-before), new optional keys (ast, bodyExtra), and ai:chat_window leaving the page component type unions. The two chained checks also narrow the runtime accept set. The claim said Clause-② no, and the PR body carries yes.",
  "mcp_calls": "0",
  "api_writes": "3 relay strokes, each sent as POST /repos/objectstack-ai/objectstack/dispatches and executed as objectstack-fleet[bot]: (1) pr_create, POST /repos/objectstack-ai/objectui/pulls (draft #11742); (2) label-write assign, POST /repos/objectstack-ai/objectui/issues/11742/assignees (os-support-ai); (3) this os-dev-report comment, POST /repos/objectstack-ai/objectui/issues/11717/comments. git push is not counted. Reads used unauthenticated REST through the proxy.",
  "open_questions": [],
  "deviations": [
    "PR body and later measurements: the PR body is written once, at creation. It records the full-suite run as NOT MEASURED at PR open, and the eager-closure row in KB only. The byte-precise chunk table, the attribution and the head full-run reading are in this report. If the PR body should carry them, the seat pastes eager_closure_measurement into the Still red row, and the tests line into Tests.",
    "Clause-② is the measured yes, not the claim's no. The dispatch made the measurement decide.",
    "objectui#11334's one-dimension pie control: 17.7.0 refuses that pie through the single-series arm, so it cannot parse. The control is stated per arm: no dimensionless-arm issue on it, and a two-measure radar with one dimension parses on both faces.",
    "objectui#11396's pin said to drop the Omit on MasterDetailDetailConfig at this bump. The Omit is kept, because the 17.7.0 entry still lists sortField as a tombstone. The card lists prose only.",
    "Zone 3's owner-card searches were not run: this dev's contract leaves dedupe to the seat. Dedupe words are listed below.",
    "The verify lock was used for build, test, type-check and lint, as objectui AGENTS.md requires, although the dispatch said it does not apply.",
    "main moved again by 9 commits after the 25251a80 merge. The PR reads mergeable_state behind, and it was not re-merged. The queue rebuilds on main.",
    "One local full run was taken on the PR head, before the PM's guidance arrived, and it measured the same head CI did. None was started after."
  ],
  "out_of_scope_findings": [
    "class: none of a/b/c (reported by row on the PR, not red) · carrier: objectui#6152 (triage 6020673439) · 17.7.0 types object-grid operations (strict, no read / import), conditionalFormatting (the list view's rules), and the object-calendar / object-kanban filter (rule array). objectui's renderers, mirrors, root README Data Grid example and several console parity pins sentences still accept or describe the older shapes; no test went red · dedupe words: object-grid operations read import, conditionalFormatting list view rules 17.7.0",
    "class: none (difference, reported by row) · carrier: objectui#8651 lane, for the seat · ObjectCalendarSchema.calendar keeps .passthrough() while the 17.7.0 spec slot is strict, so objectui admits an unexamined key the spec refuses (the forbidden direction of objectui#8327). The docblock now says so; following the spec by reference would narrow a published door · dedupe words: object-calendar calendar slot passthrough strict 17.7.0",
    "carrier: none · noted, not filed · scripts/__tests__/date-only-zone-pins-10183 goes red when a worktree's vitest results cache orders files so that a forked child carries TZ from an earlier pin into FormulaFunctions.dateOnlyZone-10866's UTC control (expected 03:00, got 11:00, Asia/Shanghai). The same head passes 27/27 with the cache moved aside; CI is green. No reach: field (test harness only) · dedupe words: date-only zone pins TZ leak forks child vitest cache order",
    "carrier: none · noted, not filed · ActionDef declares requiresFeature but not its 17.7.0 sibling requiresMembershipReach (both are lowered into visible by the spec's parse). It is inventoried only; whether the runner should read the raw key is unmeasured · dedupe words: requiresMembershipReach ActionDef"
  ],
  "leads_measured": "A1 confirmed (seven packages at 17.7.0 on npm, latest 17.7.0). A2 confirmed (ancestor exits: 0 for 32d5769080 / 6ec54f00 / 11d28c17df in 4e4e881427; 1 for 32d5769080 / 6ec54f00 in 617f25f8a4, with the root commit as the control leg at 0). A3(a) disproved (doc-snippets 782/782). A3(b) confirmed as a not-red difference. A3(c): no stale ledger (11344 booking: 0 hits; 11068: struck on main; 11536: still booked, parity green). A4: objectui#11334 and objectui#11417 complete, so Fixes lines for both. A5: types ^17.7.0, plus plugin-dashboard ^17.6.0 by spec-floors. A6: Clause-② yes.",
  "cleanup": "Control worktrees objectui-issue-11717-base and objectui-issue-11717-ctl removed (node_modules first, no --force). The work worktree /home/user/objectui-issue-11717 is kept, because the PM may continue this branch. No dev server or monitor was started.",
  "ci_state": "in_progress on c7bbce48 (24 success, 17 in_progress, 3 skipped at report time); Bundle Analysis is expected red (the eager-closure row)"
}

Session: https://claude.ai/code/session_01XgevVHGcN8dnegiiY3v75V

---

Generated by Claude Code

[objectstack-fleet]

[待裁] 17.7.0 升级让 console 首屏多加载约 179 KiB(gzip),超出性能预算:按 17.5.0 先例借额度先落地,还是先瘦身?

domain:devx @ objectui 席 2(session_01XgevVHGcN8dnegiiY3v75V)· 2026-10-06T20:14Z

维护者速读

• 一句话问题:把 objectui 升到 @objectstack/* 17.7.0 后,用户每次打开 console 首屏要多下载约 179 KiB(gzip,首屏总量约 3.4 MB 的 5%)。这超出了 console 的首屏性能预算,CI 的 Bundle Analysis 因此红着。其余 41 项检查已绿或在跑,开发 agent 能修的红都已修完。
• 席位意见:荐 A,即照 17.5.0 先例(objectui#11088,你批的「11088 同意」)在本 PR 里抬高预算,同时立回收卡,回收落地后预算按回收量降回来。
• 你要做的:回一行,例如「11717 A」。

背景与实测

• PR objectui#11742(head c7bbce48)已完成本卡全部机械工作:锁文件升到 17.7.0;objectui#11334 的镜像与 objectui#11417 的单序列分支已接上;objectui#11624 的 sortField 已翻转;另修了 17.7.0 带来的十几处机械红。开发 agent 的报告是 6024525892。
• 唯一不能机械处理的红:console 首屏加载的总包体(eager closure)超预算。
  • CI 日志读数:BUDGET_CLOSURE_GZIP_KB: 3504.6,预算 3330.4。main 上同一检查是绿的。
  • 开发 agent 实测(对照 main 02a38967,同一构建命令):增量全部落在 vendor-objectstack 这一块,+182,909 字节;其余 330 块合计净变化约 0。
  • 增量来源(开发 agent 对比 npm 上 17.6.0 和 17.7.0 两个版本各入口文件的 gzip 字节):@objectstack/spec 的 shared 入口 +152,445,导出名单没变,增长来自更多严格 schema(多出 71 处 strictObject(、58 处 lazySchema();其次是根入口 +66,383、ui +42,688。
  • 未测量:摇树之后到底留下了哪些模块;其中哪些是首屏用不到的。
• 多出的字节换来什么:console 在编辑时的校验与 17.7.0 服务端一致。图表度量数、退役的 sortField、页面 requires 要求有 kind、element:text 退役的标题变体、审批节点配置、一批 object-* 成员收紧,都会当场拒收,作者或 AI 不会写出保存时才被服务端拒的元数据。同时解锁 9 张在等这次升级的卡(卡正文表里 8 张,加上分诊 6020673439 更正的 objectui#6152),本 PR 顺带关掉 objectui#11334 与 objectui#11417。

Governing text

• scripts/check-eager-closure-budget.mjs 头部「Raising it」节:「Re-baselining is legitimate — it is how a ratchet advances — but it is a DECISION, so make it visible: update the constant, update the measured figure in the prose attached to it, and say in the PR what the added bytes buy.」
• 同文件「RAISED A THIRD TIME, ON EXPLICIT MAINTAINER RULING (objectui#11088; objectui#11073)」:每次抬高都出自维护者明确裁定。
• 先例裁定 5890854293(objectui#11088,本趟已取回):「The ceiling and the headroom assertion … are raised to the new baseline in the same commit …, the PR body states what the 413.8 KB gzip buys …, and two payback cards are filed in the same stroke. The upgrade lands now. Precedent: objectui#5531」。

前提(各带复核命令)

• P1:本 head 的 Bundle Analysis 因 eager-closure 预算失败。复核:GET /repos/objectstack-ai/objectui/commits/HEAD_SHA/check-runs?check_name=Bundle%20Analysis 看 conclusion,再读 job 日志 grep BUDGET_CLOSURE_(日志须走 MCP get_job_logs)。
• P2:main 上同一检查是绿的。复核:对 main 现 head 发同一请求(席位读到 d39ac2e5 为 success)。
• P3:增量全在 vendor-objectstack。这是开发 agent 的实测,席位未重测。复核:两棵树各自构建 console,比对 apps/console/dist/eager-closure.json。

选项 × 真实代价

选项	做什么	用户可感知的后果
A 借额度先落地	本 PR 同一提交里抬高总预算、vendor-objectstack 分块预算与 BASELINE,余量照旧是半个回归量;PR 写明多出的字节换来什么;同笔立回收卡(objectstack 侧:spec shared 入口在导出不变时涨了 152 KB gzip;是否要 objectui 侧的卡由实测决定)	首屏今天就多 179 KiB;编辑校验立刻与服务端一致;9 张卡解锁。回收卡落地后,预算按回收量降回
B 先瘦身再升级	PR 停在草稿;先在 objectstack 侧把 spec 入口减下来,发新版本后再升级	首屏不变大;但 console 继续按 17.6.0 校验,作者写出服务端会拒的元数据,保存时才报错;9 张卡继续等
C objectui 侧懒加载	先把 spec 校验移出首屏再落地	能救回多少未测量;17.5.0 那次读源码的结论是大头在 spec 入口里,console 侧懒加载救不回大部分

业务含义直译:A = 先赊账把新货上架,账单和还款计划同时写好;B = 等供应商把包装改小再上货,这期间货架上是旧版;C = 自己拆包装重装,省多少不知道。

四轴(业务立场)

• 项目长远合理性:A 是带具名还款卡的借贷,先例两次都还了(objectui#11101、objectstack#20646 落地后预算两次下调);B 让 console 与服务端的校验长期不一致;C 是消费端绕行,收益未测。
• 实际业务拉动:9 张卡在等这次升级(实测于本卡正文与分诊更正);你点名「11717 最优先」。
• 防 AI 犯错:A 让 AI 在 console 写元数据时当场被拒(响亮);B、C 拖期间,AI 在 console 里写出的是服务端保存时才拒的文档(编辑期静默)。
• 创业阶段不扩散:A 只按裁定改既有门禁的参数,不新增门禁也不加豁免;C 要多一套懒加载结构。

os-decision-facets
① 项目长远合理性:A 借额度带具名回收卡,先例两次回收兑现;B 拉长前后端校验分歧;C 消费端绕行,收益未测。
② 实际业务拉动:9 张卡等升级(实测);维护者点名最优先。
③ 防 AI 犯错:A 编辑期响亮拒收;B/C 期间 console 静默接受服务端会拒的文档。
④ 创业阶段不扩散:A 只改既有门禁参数,不加门禁不加豁免;C 新增懒加载结构。
Prior rulings read: eager-closure budget raise → objectui#11088 5890854293(A),objectui#5531(gate header 引用,未逐字取回); ADR none; thread: 本卡 6024525892 6020673439

推荐:A,回退 B。只看①选 A;②③④ 是否翻转:否。
置信缺口:摇树后留在首屏的具体模块未测量,所以回收卡能收回多少未知;首屏是否真需要这些 schema 也未测量。

裁后执行

• 裁 A:
  • 本席续派同一个开发 agent:合并最新 main;在一个提交里改 MAX_EAGER_CLOSURE_GZIP_BYTES、BASELINE、PER_CHUNK_BASELINE,头部按 11088 的格式写明分块表、换来什么与回收卡号;
  • 本席立回收卡:objectstack 侧 spec shared 入口体积;objectui 侧仅在实测指向时立;
  • 契约复核(达档)覆盖落地 head 后,翻 ready、入队。
• 裁 B:本卡转 pm:blocked,Blocked-by: 指向新立的 objectstack 瘦身卡;PR 保持草稿。
• 裁 C:本席续派开发 agent 先测量可懒加载的量,按读数回报。

本笔状态:pm:dispatched → needs-user-decision。认领、assignee 与 PR 都保留;PR 保持草稿。在途的达档契约复核照常进行。

---

Generated by Claude Code