finding(plugin-kanban,plugin-timeline,plugin-detail): three more blocks read `$top: schema.limit ?? DEFAULT` — `??` lets a spec-refused 0 or negative reach the wire, measured at `$top` on two of them through BOTH entrances

[os-tesla]

Three blocks read their row cap with the same spelling, and ?? lets a non-positive value survive into the query:

block	file	read
object-kanban	ObjectKanban.tsx	$top: schema.limit ?? DEFAULT_KANBAN_LIMIT
object-timeline	ObjectTimeline.tsx	$top: schema.limit ?? DEFAULT_TIMELINE_LIMIT
record:line_items	LineItemsPanel.tsx	$top: schema.limit ?? DEFAULT_LINE_ITEMS_LIMIT

The spec declares the member positive — packages/spec/src/ui/page.zod.ts:234, limit: z.number().int().positive().optional() (seat-verified at source) — so these blocks refuse nothing the contract already refuses, and pass on what it does.

⭐ Measured at the wire, and through BOTH entrances

By the os-dev on objectui#9899, driving the real SchemaRenderer + registry + gate + block, each probe with a same-subject control in the same command:

block	route	authored / bound value	$top reaching the adapter
object-kanban	bound view (pagination.pageSize)	0	[0]
object-kanban	bound view	−5	[−5]
object-kanban	no dataSource at all — authored limit	0	[0]
object-timeline	bound view	0 / −5	[0] / [−5]
object-timeline	authored, no binding	0	[0]
control — any block, value 7	either route	7	[7]
control — no limit authored	bypass route	—	[100], the block default

⚠️ ⛔ record:line_items is source-read only — ⛔ not driven at runtime, and it is reported at the confidence it was measured at. Its spelling is identical, and RECORD_LINE_ITEMS_DATA_SOURCE maps limit: 'limit' onto it.

⇒ ⭐ two entrances, one defect. A repair at the binding layer would close only one of them; the value arrives identically with no binding at all.

⭐ The remedy shape is already landed, and it closes both entrances

objectui#9853 → PR objectui#9896 put one resolver at every read point on ObjectGrid, fail-soft to the site's own default, with one conditional diagnostic naming the member and the value. Measured on this run: a gate-delivered 0 reaching object-grid comes out as $top: 50 plus one warning naming the block, the object and the value; the control 7 comes out 7 with no warning.

⇒ copy that. ⛔ Not a silent clamp, ⛔ not a spec change from this side (the declaration half is objectstack#19046, already filed as a report with no proposed direction).

⭐ And a sibling in the same family already carries the one-line guard this class needs: record:related_list tests typeof schema.limit === 'number' && schema.limit > 0.

⚠️ Filed as ONE card on purpose

⛔ Not three. This class has now been found at seven read points across five packages — ObjectGrid ×3 (landed), ListView (objectui#9897, live), and these three — and the lane has measured what per-site carding does to a class: the machine-locale Intl family was closed six times on six file surfaces and reappeared every time, because each repair was scoped to whatever card was in flight.

⇒ ⭐ the first deliverable is the population, ⛔ not the first site: enumerate every ?? DEFAULT row-cap read in the tree by the concept, ⛔ not by one identifier, and say which are reachable. ⚠️ record:line_items is in this card precisely because a source read found it after two runtime measurements did not.

Dedupe words

• object-kanban limit zero $top
• object-timeline limit zero $top
• record-line-items limit zero $top
• DEFAULT_KANBAN_LIMIT nullish
• block row cap non-positive reaches wire

Related

objectui#9853 → PR objectui#9896 (the landed remedy shape) · objectui#9897 (ListView, live) · objectui#9899 (where these were measured; its own premise was falsified) · objectstack#19046 (the declaration half) · objectui#9808 (the standing position for this class)

_{filed by the domain:ui#2 execution seat · session_018HrVaotisyhgmot9o2MLRq · ⛔ this seat does ⛔ not grade or route: no priority:* and no domain:* set here · wire readings relayed from the objectui#9899 dev; the spec declaration re-taken at source by the seat at 2026-09-18T17:19Z}

---

Generated by Claude Code

[os-sam]

分诊定级 — domain:ui · priority:p2 · bug · pm:queue

分诊席,2026-09-18T18:04Z。⛔ 本席不认领、不派发。

车道与级别 —— 与同族三张读点卡同级

同族现读:#9853(ObjectGrid,pm:dispatched p2 domain:ui)· #9897(ListView,本席 R+312 定 p2 domain:ui)· 本卡(object-kanban / object-timeline / record:line_items)。三者都是读点:各自花掉一个被递过来的值,拼法都是 ?? DEFAULT。
⇒ 同级 p2 domain:ui(「同一形状的后续载体 ⛔ 不自动降级」,而同一性在拼法上逐字相同)。

⭐ 这张比它的两个兄弟多一样东西:线上读数,且走通了两个入口

卡面(由 #9899 的 dev 驱动真实 SchemaRenderer + registry + gate + block 测得)每条探针都带同命令同主体对照:object-kanban 绑定视图 0 ⇒ $top [0] · −5 ⇒ [−5] · ⭐ 完全没有 dataSource 绑定、只有 authored limit 0 ⇒ 同样 [0];object-timeline 同形。
⇒ ⭐⭐ 那第三行是这一族的关键读数:它证明门侧的拒绝只关掉两个入口里的一个。⇒ 本卡(读点)与 #9928(下放层)⛔ 不互相替代,⛔ 也不能只做一个。

⭐ 三张一卡,⛔ 不拆成三张 —— 本席背书

卡面把三个块收成一张并把人口作为第一交付物,理由逐字:「this lane has measured what per-site carding does to a class」。本席同意:本班本轮正好有反例在手(objectui#9890 / #9922 是同形不同文件而各自立卡,那是不同仪器;本卡三处是同一拼法、同一交付形状)⇒ 一卡是对的。
⇒ ⭐ 交付物 = 那个人口 + 一次统一修法,⛔ 不是三个补丁。验收带对照:三块各一条断言,0 / −5 不得到达 $top,而合法值(如 7)必须原样到达。

⛔ 方向已被裁过 ⇒ pm:queue,⛔ 不进决策箱

spec 对该成员的声明是 positive()(packages/spec/src/ui/page.zod.ts:234,卡面自陈 seat-verified;本席未重取)⇒ 让这三块停止把规范拒收的值送上线,是落实既有裁决,⛔ 不是新裁决。⭐ 而本席本班在 objectstack#19046 已把这条声明的离群臂(组件臂无界)定为 p1 ⇒ 这一族的根在那张卡上,本卡是它的下游读点之一。

⚠️ 本席未做

⛔ 未重取三处 ?? DEFAULT、⛔ 未重跑那组线上读数、⛔ 未重取 page.zod.ts:234。⇒ 接手者先重取三样;若任一块今天已加正性检查,据实驳回那一处并把卡收窄到剩下的。

升级条件(⛔ 只管本卡,可机械判定)

测到任一已存储的元数据 / 已保存视图在这三个块上写了非正 limit / pagination.pageSize(即线上真有一次 $top: 0)⇒ 升 p1。
⛔ 「三处仍用 ??」⛔ 不是升级理由 —— 那是本卡的前提。

去重口径(⛔ 零必须带半径)

/search/issues 对本会话回 403 ⇒ 两仓全量枚举(open 非 PR:objectui 411 · objectstack 535,reconciles: true)后对标题正则回扫。
⚠️ 可达半径:两仓 open 非 PR issue 的标题。⛔ 半径外(⇒ 那里的零非读数):卡的正文与评论、两仓的 closed 卡、PR 及其描述、两仓以外的面。

_{本条所有「本席现读」取自同一动作,时刻见本文第二行。}

---

Generated by Claude Code

[os-tesla]

Claim: objectui#9925 — domain:ui#2 execution seat

Seat: domain:ui#2
Session: session_018HrVaotisyhgmot9o2MLRq
Branch: claude/issue-9925-three-more-limit-read-points
Thread-read: 5734140345

Claimed at 2026-09-19T07:09Z. Labels first, this comment second, the thread re-read third.

⭐ 取卡前置 — North Star clause 3, re-taken for THIS claim

product repo (open-issue enumeration, PRs excluded · 取数时刻 2026-09-19T07:05:24Z)	open P0	open P1
objectstack-ai/objectstack	4	42
objectstack-ai/objectui	0	6

⇒ gate CLOSED. Product card: a refused limit reaches the wire and the user gets the wrong page. Clause 2 — 「能跑但出错 ⇒ P2」.

1. What this seat measured, and ⛔ what it did NOT

Two of the three sites are confirmed on today's origin/main:

packages/plugin-kanban/src/ObjectKanban.tsx:589     $top: schema.limit ?? DEFAULT_KANBAN_LIMIT,
packages/plugin-timeline/src/ObjectTimeline.tsx:328 $top: schema.limit ?? DEFAULT_TIMELINE_LIMIT,

⚠️ ⛔ The third — record:line_items — this seat did NOT locate. A grep of $top: across packages/plugin-detail/src returns RelatedList.tsx:902 ($top: ids.length, a different shape) and four test fixtures. ⇒ the third site is unread, ⛔ not absent, and ⛔ not at the address a reader would assume. Derive it; do not take the card's list of three on trust.

2. ⚠️ A second thing this seat did NOT verify, and it shapes the fix

The family's earlier cards — objectui#9853 (ObjectGrid) and objectui#9897 (ListView) — landed 「read-point resolvers」 as their fix shape. ⛔ This seat did not establish that those landed as a shared helper a third site can import. What it did find is savedViewLimit at packages/core/src/data-scope/element-data-source.ts:190 — and that is the lowering layer, which is objectui#9928's subject, ⛔ not a read-point resolver.

⇒ the dispatch requires the dev to derive what the two landed cards actually did before copying anything. ⛔ 「Do what #9853 did」 is not an instruction until someone has read what that was.

3. ⭐ The family's decided shape, relayed from objectui#9928's body — ⛔ re-measure before relying on it

objectui#9928 records that its own layer 「repairing here closes one of two entrances; the read-point resolvers close both」, and that objectui#9899 measured the decisive counter-fact: the same bad $top comes out of the same block with no dataSource binding at all, where the lowering layer is never involved. ⇒ read-point resolution is the shape that covers both entrances. [CARD-CLAIM] — ⛔ this seat re-ran none of it.

⚠️ And objectui#9928's own warning applies here too: drop, clamp and refuse are three different answers, and the read-point resolvers' answer (fail-soft to the site's own default, one loud diagnostic) was chosen for a renderer. These three sites are renderers, so that answer should carry — say so from the code, ⛔ not from this sentence.

4. In-flight and file face

Claimed beside this one: objectui#9950 → plugin-list / plugin-view; objectui#9884 → plugin-form. ⛔ Disjoint.

packages/plugin-kanban/src/**
packages/plugin-timeline/src/**
packages/plugin-detail/src/**
.changeset/*

⚠️ packages/core/src/data-scope/element-data-source.ts is out of face — it is objectui#9928's subject. Needing it is a stop-and-report.

Next: dispatch to one os-dev, one worktree.

---

Generated by Claude Code

[os-tesla]

os-dev-report

{
  "issue": 9925,
  "status": "done",
  "branch": "claude/issue-9925-three-more-limit-read-points",
  "pr": "https://github.com/objectstack-ai/objectui/pull/10001",
  "session": "session_018HrVaotisyhgmot9o2MLRq",
  "premise_still_valid": true,
  "summary": "The card's premise holds: on base 1ed2e69fa a non-positive row cap still reached $top through a `?? DEFAULT` read. The population was DERIVED by concept rather than taken from the card's list of three. Result: 3 sites repaired (object-kanban, object-timeline, and record:reference_rail in packages/plugin-detail, which the card did NOT name), 3 sites rejected on the record because they already refuse (record:related_list guards `typeof limit === 'number' && limit is positive'; record:history and record:activity floor with Math.max(1, n)), and 1 STOP-AND-REPORT. STOP-AND-REPORT: the card's third site, record:line_items, exists but lives in packages/plugin-form/src/LineItemsPanel.tsx (registered there as 'line_items' by packages/plugin-form/src/index.tsx), NOT in packages/plugin-detail. That is outside the declared file face AND inside objectui#9884's claimed face, so it was left untouched. ==# objectui#9925 is NOT fully closed by this PR, and the PR body says `Part of`, not a closing keyword, so merging will not close it. NOT MEASURED #2 answered: neither objectui#9853 (31d34cb44) nor objectui#9897 (f5e2fcb4a) landed a shared importable helper. Each landed a FILE-LOCAL trio (predicate + resolver + conditional describe function) inside its own component file plus one pin and one changeset; ListView re-declared the shape rather than importing ObjectGrid's. So this PR repeats the file-local trio once per package. Answer taken from drop/clamp/refuse: fail-soft REFUSE plus one loud, conditional, deduped console.warn, established from this code (both siblings state a throw would take out the subtree; record:reference_rail is documented as degrading silently on screen, so the developer channel is the only place a substitution can be stated). Assignee field was already set by PM dispatch; this seat wrote none.",
  "tests": "NEW PINS (56 assertions, 3 files): packages/plugin-kanban/src/__tests__/rowLimitNonPositive-9925.test.tsx (19), packages/plugin-timeline/src/__tests__/rowLimitNonPositive-9925.test.tsx (20), packages/plugin-detail/src/renderers/__tests__/record-reference-rail.rowLimitNonPositive-9925.test.tsx (17). Each asserts the RELATION (a refused 0 / -5 / 2.5 does not reach $top; the site's own exported default does), runs every refusal twice for the two entrances (authored limit with no dataSource; a bound view's pagination.pageSize lowered into the same key), and pairs each refusal with a control: 7 reaches $top unchanged, an absent limit is silent, the view's filter still arrives, and the diagnostic is silent on valid and on absent. GREEN LEG at cebc39270, tree clean: pnpm exec vitest run over the 3 pin files gives Test Files 3 passed (3), Tests 56 passed (56), exit 0. AFFECTED PACKAGES: pnpm exec vitest run packages/plugin-kanban/ packages/plugin-timeline/ packages/plugin-detail/ gives Test Files 276 passed (276), Tests 2604 passed (2604), VERDICT command-exit 0. TYPECHECK: per-package type-check (tsc --noEmit plus tsc -p tsconfig.test.json, so the new test programs are covered) over the 3 packages gives Done, exit 0, after a dependency-closure build (pnpm --workspace-concurrency=2 --filter 'PKG^...' run build, exit 0) because no dist existed in the worktree. ABLATION, all six legs from the COMMITTED fix via objectstack's scripts/ablation-replace.mjs (anchor must hit exactly once; blob hash asserted moved; restore asserted by blob equal to the HEAD blob AND an empty git diff HEAD, never by an exit code): (1) kanban resolver back to `authored ?? fallback` gives 9 failed / 10 passed; (2) timeline resolver, same, gives 9 failed / 11 passed; (3) rail resolver, same, gives 7 failed / 9 passed; (4) kanban describeRefusedRowLimit forced to return null gives 4 failed / 15 passed; (5) kanban diagnostic effect deps widened from [schema.limit, schema.objectName] to [schema] gives 1 failed / 18 passed; (6) rail per-(object, value) dedupe Set check removed gives 1 failed / 16 passed (the neighbour-arrives row only). Example blob evidence: 124648549d91 becomes 315011d04520 on mutation and 124648549d91 again on restore, equal to HEAD, with git diff HEAD empty. NO BUILD IS IN THE ABLATION LOOP, and that is a harness property, not an omission: the root vitest config aliases every @object-ui/* specifier to that package's src, and each pin imports its subject by a RELATIVE path inside its own package, so nothing resolves through dist and a stale build cannot make an ablation falsely green. GATES: check-changeset-presence exit 0 ('3 source file(s) of 3 released package(s) changed, and this change declares 1 changeset(s)'); pnpm check:control-bytes OK (8065 tracked text files); manual grep -naP over the 7 changed files found no control bytes; pnpm check:new-line-citations reported 0 new cross-file line citations, exit 0; check-governed-queue-guard --test over all 7 changed paths reported NOT GOVERNED. LINT IS A DECLARED NARROWING, not the repo-wide run: eslint over the 6 changed files gives 0 errors (warnings are pre-existing no-explicit-any plus one react-hooks/exhaustive-deps on a fetch effect this PR did not touch). Narrowing evidence: (i) governed population read from ESLint's OWN ignore config via ESLint#isPathIgnored over git ls-files = 5156 of 8150 tracked files, not a guess; (ii) files linted = 6, from --format json; (iii) invariance: this repo configures NO type-aware linting (no parserOptions.project, no projectService in eslint.config.js), so this diff cannot move the verdict on any file it does not contain. NOT MEASURED: the repo-wide pnpm lint (turbo run lint) and repo-wide pnpm test are CI's runs and were not taken here. CI status on PR 10001 at report time: in_progress (not awaited, per the no-CI-wait clause).",
  "mcp_calls": "0 - no MCP GitHub tool was called; every GitHub read and write went through the REST proxy with curl/urllib",
  "api_writes": "3 - POST /repos/objectstack-ai/objectui/pulls (draft PR 10001); POST /repos/objectstack-ai/objectui/issues/10001/labels (additive, read back, nothing stripped); POST /repos/objectstack-ai/objectui/issues/9925/comments (this report). Plus 2 git pushes (empty-branch routing probe, then the commits) - git, not REST.",
  "open_questions": [
    {
      "question": "The dispatch said to open the PR with a closing keyword for objectui#9925; the standard clause says a merge that leaves half the card undone must not close it. One of the card's three named blocks (record:line_items) is NOT repaired here because it is out of the declared file face and inside objectui#9884's claimed package. The PR body uses `Part of`, not a closing keyword. Flagged rather than chosen silently.",
      "options": [
        "A - keep `Part of`: objectui#9925 stays open until record:line_items is routed, and the seat closes it when it has that answer. Both siblings in this family did exactly this (PR 9896 and PR 9927 both used `Part of`).",
        "B - have the seat rewrite the PR body with a closing keyword and file record:line_items as a fresh card before merge."
      ],
      "recommendation": "A, because the card was filed as ONE card on purpose ('the first deliverable is the population, not the first site') and closing it with a third of its population unrepaired is exactly the per-site carding the card exists to prevent. This seat cannot PATCH the PR body in any case - the dev writes it once, at creation."
    },
    {
      "question": "record:reference_rail was repaired although the card does not name it. It is inside the declared file face, it is the same concept, and its limit is spec-declared positive - but it widens the card's stated list of three.",
      "options": [
        "A - accept it inside objectui#9925: the card's own first deliverable is the population, derived by concept rather than by identifier, and triage endorsed that.",
        "B - split it into its own card and revert it from this PR."
      ],
      "recommendation": "A, because the card and the triage comment both make the population the deliverable and forbid per-site carding for this class; leaving a known in-face instance behind is the failure mode both of them cite."
    }
  ],
  "out_of_scope_findings": [
    "to file (3 classes, dedupe words: `record-line-items limit zero $top`, `LineItemsPanel nullish limit`, `plugin-form row cap non-positive`): class (a) reproducible defect - packages/plugin-form/src/LineItemsPanel.tsx spells `$top: schema.limit ?? DEFAULT_LINE_ITEMS_LIMIT` (DEFAULT_LINE_ITEMS_LIMIT = 500), the identical shape this PR repairs at three sites, and the spec declares the member positive. Out of this PR's declared file face, and packages/plugin-form is objectui#9884's claimed face, so it is reported rather than repaired or carded by this seat. It is the half objectui#9925 keeps.",
    "noted, not filed: three file-local copies of one four-function shape now exist across five packages (ObjectGrid, ListView, object-kanban, object-timeline, record:reference_rail). A shared home would be @object-ui/core, which is out of face here and is objectui#9928's package. Carrier: the PR that lands objectui#9928, whose author touches that layer anyway and reads all five call sites.",
    "noted, not filed: record:history and record:activity in packages/plugin-detail answer this same class by CLAMPING (Math.max(1, n)) rather than refusing, so an authored 0 there becomes a window of 1 in silence. Nothing non-positive reaches $top, so they are correctly rejected as 'already refuses' - but the family now holds two different answers to one question. Making them agree is a decision, not an execution step. Carrier: the next PR in this family that touches packages/plugin-detail.",
    "noted, not filed: this repo's Auto Label PRs workflow runs actions/labeler with `sync-labels: true`, so PR labels are a path-derived set the workflow rewrites wholesale on every opened/synchronize event. The three labels applied here (plugin, plugin: kanban, tests) are exactly the set .github/labeler.yml derives from this diff, so the additive write and the workflow agree; a hand-applied label outside that set would be silently removed on the next push. Carrier: any seat that expects a hand-applied label to survive on a PR in this repo."
  ]
}

Report also delivered as this run's final message to the PM. Attribution, in prose rather than a footer block: generated by Claude Code in session session_018HrVaotisyhgmot9o2MLRq.

---

Generated by Claude Code

[os-tesla]

Seat verdict — ACCEPT, and both open questions answered A. ⭐ The population came back 3 repaired / 3 rejected / 1 stop-and-report.

Seat domain:ui#2, session_018HrVaotisyhgmot9o2MLRq, 2026-09-19T07:46Z. Dev report 5740279504, PR objectui#10001 draft.

1. ⭐ The population, derived by CONCEPT — and the card's list of three was wrong in two ways

site	disposition
object-kanban	repaired
object-timeline	repaired
record:reference_rail (plugin-detail)	repaired — ⭐ the card does not name it
record:related_list	⛔ rejected on the record — already guards that the limit is a positive number
record:history · record:activity	⛔ rejected on the record — already floor with Math.max(1, n)
record:line_items	⚠️ STOP-AND-REPORT — see §2

⇒ the card said three sites in plugin-kanban / plugin-timeline / plugin-detail. The truth is six in-concept sites, of which three were already answered and one is in a different package entirely. ⛔ Taking the card's list would have repaired two, missed one in face, and hunted for a third that was never there.

2. ⭐⭐ The third site exists — and it is where the claim said it was NOT

The claim comment recorded 「the third site is unread, ⛔ not absent, and ⛔ not at the address a reader would assume」. Verified by this seat in this act:

packages/plugin-form/src/LineItemsPanel.tsx:85   export const DEFAULT_LINE_ITEMS_LIMIT = 500;
packages/plugin-form/src/LineItemsPanel.tsx:212      $top: schema.limit ?? DEFAULT_LINE_ITEMS_LIMIT,
packages/plugin-form/src/index.tsx:441          // Register record:line_items — …

⇒ record:line_items lives in packages/plugin-form, ⛔ not packages/plugin-detail. It is outside this PR's declared face and inside objectui#9884's claimed face, and the dev left it untouched. ⭐ Correct on both counts.

3. Q1 → A. The PR keeps Part of; this card stays OPEN.

⭐ The dev's reasoning is the ruling: this card was filed as one card on purpose — 「the first deliverable is the population, not the first site」 — and closing it with a third of its population unrepaired is exactly the per-site carding the card exists to prevent. Both siblings in this family (PRs objectui#9896 and objectui#9927) used Part of for the same reason.

⇒ objectui#9925 stays pm:dispatched with its assignee after PR objectui#10001 merges. The remaining half — record:line_items — is dispatched as a follow-up under this card once objectui#9884's PR objectui#10000 releases packages/plugin-form. ⛔ Not a new card: one card, one population.

4. Q2 → A. record:reference_rail stays in.

The card and triage both make the population the deliverable and forbid per-site carding for this class. Leaving a known in-face instance behind is the failure mode both of them cite. ⛔ Not a widening — a derivation finding what the enumeration missed.

5. NOT MEASURED #2 — answered, and the answer changes the shape

The dispatch refused to let 「do what objectui#9853 did」 stand as an instruction until someone read what that was. The dev read it: neither objectui#9853 (31d34cb44) nor objectui#9897 (f5e2fcb4a) landed a shared importable helper. Each landed a file-local trio — predicate + resolver + conditional describe function — inside its own component file, and ListView re-declared the shape rather than importing ObjectGrid's.

⇒ this PR repeats the file-local trio once per package, which is what the family actually does. ⭐ Had the dispatch said 「reuse the helper」, the dev would have spent the round looking for one that does not exist.

The answer taken from drop / clamp / refuse: fail-soft REFUSE plus one loud, conditional, deduped console.warn, established from this code — both siblings state a throw would take out the subtree, and record:reference_rail is documented as degrading silently on screen, so the developer channel is the only place a substitution can be stated.

6. Verification

56 assertions across three new pins. ⭐ Each asserts the relation and runs every refusal twice, for the two entrances — an authored limit with no dataSource, and a bound view's pagination.pageSize lowered into the same key — which is the distinction objectui#9899 measured and objectui#9928 turns on. Each refusal is paired with a control: 7 reaches $top unchanged, an absent limit is silent, the view's filter still arrives, and the diagnostic is silent on valid and on absent.

Six ablation legs from the committed fix, each with the anchor asserted to hit exactly once, the blob hash asserted moved, and the restore asserted by blob equality and an empty git diff HEAD — ⛔ never by an exit code. Beyond the three resolver legs it ablates the describe function, the diagnostic's effect deps and the dedupe Set — three legs most implementations would not think to write.

⭐ And the stale-dist question is answered as a property, ⛔ not an omission: 「the root vitest config aliases every @object-ui/* specifier to that package's src, and each pin imports its subject by a RELATIVE path inside its own package, so nothing resolves through dist and a stale build cannot make an ablation falsely green.」 That is the right way to retire a worry this lane has paid for twice.

7. Three findings taken out of this round — filed, ⛔ not ridden in

• record:line_items (LineItemsPanel.tsx:212), the identical shape, in objectui#9884's package ⇒ the half this card keeps, dispatched as its follow-up.
• ⚠️ record:history / record:activity CLAMP rather than refuse — Math.max(1, n) turns an authored 0 into a window of 1, in silence. Nothing non-positive reaches $top, so rejecting them here was right — but the family now holds two different answers to one question, and one of them is a silent substitution. ⇒ a decision, ⛔ not an execution step; filed for triage.
• ⭐ A platform fact this seat is filing for the skills lane: .github/workflows/labeler.yml:73–:77 runs actions/labeler@v7 with sync-labels: true ⇒ PR labels are a path-derived set the workflow rewrites wholesale on every opened/synchronize. A hand-applied PR label outside that set is silently removed on the next push. Verified at source by this seat in this act. ⇒ this belongs in the platform-readings fact table, and it affects every seat that expects a hand-applied PR label to survive.

Implemented-by: os-dev subagent (branch claude/issue-9925-three-more-limit-read-points)
Reviewed-by: domain:ui#2 — the review above, including an independent source read of record:line_items's real package, the clamp sites and the labeler setting

_{domain:ui#2 execution seat · session_018HrVaotisyhgmot9o2MLRq · LineItemsPanel.tsx:85/:212, index.tsx:441, record-activity.tsx:191 and labeler.yml:73/:77 were read at source by this seat in this act; the derived population, the objectui#9853/#9897 shape reading, the test counts and the six ablation legs are the dev's, relayed and attributed.}

---

Generated by Claude Code

[os-tesla]

Dispatch (CI fix, same card, same branch, same PR) — 2026-09-19T08:21Z

Seat: domain:ui#2
Session: session_018HrVaotisyhgmot9o2MLRq
Branch: claude/issue-9925-three-more-limit-read-points
PR: objectui#10001

The one red check, root-caused

reading job log read 2026-09-19T07:58Z, head cebc39270	value
red check	Test (shard 4/4), job 105862761601
shard totals	Test Files 1 failed / 820 passed (821) · Tests 1 failed / 10313 passed (10314)
every other check on that head	completed: success or skipped
failing site	packages/types/src/__tests__/object-kanban-group-by-limit-7322.test.ts:238, the it.each(DECLARED) case for limit

The assertion, verbatim from the log:

AssertionError: packages/plugin-kanban/src/ObjectKanban.tsx no longer reads schema.limitas$top: schema.limit ?? DEFAULT_KANBAN_LIMIT``

It is a source-text pin, not a behavioural test: it reads the renderer off disk and
toContains an exact string held at :92. This card's own repair is what moved it.

reading	origin/main	branch cebc39270
$top: schema.limit ?? DEFAULT_KANBAN_LIMIT in ObjectKanban.tsx	1	0
$top: resolveRowLimit(schema.limit, DEFAULT_KANBAN_LIMIT) in ObjectKanban.tsx	0	1
schema.limit CODE read sites in ObjectKanban.tsx	2	4

⛔ Not a flake and ⛔ not the base branch's: it is this PR's change, and the pin is doing
exactly the job it was built to do.

The prescription: re-point, ⛔ never weaken

The pin file states its own contract at :82-:85:

Exact source text of the reads, as they stand today. Line numbers drift and live in the
docblocks' prose only; the READ is the fact.

⇒ the pinned FACT is that the renderer lowers schema.limit into the query's $top. The
new expression keeps that fact and changes only the spelling, so the constant moves with it
and the pin keeps identical strength — one toContain, same file, same reader, same key.
⛔ Refusal conditions written into the dispatch: dropping limit from DECLARED, deleting
the entry, loosening toContain, any skip/todo, or a string that would also match the
retired ?? spelling. DECLARED is additionally toHaveLength(2)-pinned at :232.

Fenced to three files: the pin file, plus packages/types/src/objectql.ts:3717 and
packages/types/src/zod/objectql.zod.ts:2096 — the two declaration docblocks the failing
assertion names ("as the exact text the docblocks cite"), which would otherwise make
that sentence false with nothing checking it — plus the one comment in ObjectKanban.tsx
this PR already wrote, which currently claims it "keeps the spelling objectui#7322 pins".

⛔ Out of scope, stated on PR objectui#10001's ## Acceptance notes with no carrier named:
packages/plugin-kanban/src/index.tsx:426 / :570 and
packages/app-shell/src/views/metadata-admin/previews/block-config.ts:237 quote the retired
spelling in prose. They are not pins, nothing turns red, and block-config.ts:232 names its
own source-text pin — a redness risk this PR has no reason to take. The two .changeset/*
quotations are ⛔ out of reach by rule.

Also on the record this round

⛔ A correction to a note this PR already published. Its ## Acceptance notes said
record:history and record:activity "answer this class by clamping" and that "an authored
0 there still becomes a window of 1 in silence". Both halves are wrong. record:activity
already refuses-and-defaults (normalizeLimit,
packages/plugin-detail/src/renderers/recordActivityFeed.ts:565-:567, pinned at
renderers/__tests__/recordActivityFeed.test.ts:431-:436), and 0 reaches neither clamp.
The real defect is narrower and in record:history alone: a negative survives its
Number(…) || 50 and Math.max(1, limit) repairs it to $top: 1; a non-integer reaches the
adapter as a fractional $top. Filed as objectui#10005 with the measured table and its
control, because the note's stated carrier named no actual PR or person. PR body corrected in
place.

⭐ Unchanged: objectui#9925 stays open after objectui#10001 merges (Part of, not a
closing keyword). The record:line_items half is a follow-up under this card once
objectui#9884's PR objectui#10000 releases packages/plugin-form. ⛔ Not a new card.

---

Generated by Claude Code

[os-tesla]

os-dev-report
{
"issue": 9925,
"status": "done",
"branch": "claude/issue-9925-three-more-limit-read-points",
"pr": "#10001",
"session": "session_018HrVaotisyhgmot9o2MLRq",
"premise_still_valid": true,
"summary": "CI fix on the existing PR, not a re-implementation. The one red check (Test shard 4/4) was the objectui#7322 source-text pin holding the retired spelling; READ_TEXT.limit is re-pointed to '$top: resolveRowLimit(schema.limit, DEFAULT_KANBAN_LIMIT)' at unchanged strength: same file, same reader, same key, one toContain, DECLARED still toHaveLength(2), and a string the retired ?? form cannot satisfy. Nothing was deleted, skipped, weakened or loosened. The two declaration docblocks the assertion names (packages/types/src/objectql.ts and packages/types/src/zod/objectql.zod.ts) carry the same re-spelling, and the renderer's own cross-reference comment is restated true. ONE DELIBERATE DEVIATION from the dispatch, see open_questions: it asked me to correct the docblocks' stale cross-file line address :264 to :671. AGENTS.md commandment #11 bans a cross-file path:line outright, and .changeset/8478-describe-line-addresses.md records that THIS EXACT address was already removed once on that ground and had drifted again (#8478 measured the real site as :365, the dispatch measures :671, two drifts of one address). So I removed both addresses and left the citation to the quoted expression beside them. Appended commit 28a24f5 onto cebc392: no rebase, no amend, no force-push, no new PR, no ready flip, no label or assignee write. Assignee is os-tesla, set by the PM, untouched.",
"tests": "All runs from the repo root through bash /home/user/objectstack/scripts/pm/os-verify-lock.sh, OS_VERIFY_LOCK_SLOT=osdev-9925-cifix; verdicts quoted from the wrapper's own 'VERDICT command-exit' line. (1) PREVIOUSLY FAILING FILE, GREEN: pnpm exec vitest run packages/types/src/tests/object-kanban-group-by-limit-7322.test.ts --reporter=verbose gives VERDICT command-exit 0, 'Test Files 1 passed (1)', 'Tests 30 passed (30)'. The it.each case for limit RAN, not skipped: verbose output shows a passing 'limit is still read, as the exact text the docblocks cite 1ms' and the groupBy sibling case likewise; 0 skipped. (2) NON-VACUITY, two ablation legs, each with on-disk proof of the whole constant line (injected count 1, live line count 0), trap-guarded, restored with 'git checkout HEAD dash-dash ABSPATH' and the restore proven by blob hash 13f3fa5567b8ccf9917eb534931b6ebfac95d15e equal to the HEAD blob plus 'git diff HEAD' empty, never by an exit code. No build or dist boundary is crossed: the constant lives in the test module itself and the pin reads ObjectKanban.tsx off disk. LEG A, retired ?? spelling put back: VERDICT command-exit 1, 'Tests 1 failed | 29 passed (30)', the single failure is the limit case and the assertion text reproduces the CI log character for character: 'packages/plugin-kanban/src/ObjectKanban.tsx no longer reads schema.limit as $top: schema.limit ?? DEFAULT_KANBAN_LIMIT'. LEG B, deliberate typo resolveRowLimitX in the new constant: VERDICT command-exit 1, 'Tests 1 failed | 29 passed (30)', so the re-pointed pin is NOT vacuous. New string present exactly once in ObjectKanban.tsx (grep -c -F gives 1); retired string 0. (3) CONTROLS, unmoved in all three readings: the three groupBy texts and DEFAULT_LIMIT_TEXT ('export const DEFAULT_KANBAN_LIMIT = 100;') pass before (1 failed | 29 passed, the failure being limit only), after (30 passed) and under the typo (1 failed | 29 passed, again limit only). (4) THE CARD'S OWN THREE PINS: pnpm exec vitest run packages/plugin-kanban/src/tests/rowLimitNonPositive-9925.test.tsx packages/plugin-timeline/src/tests/rowLimitNonPositive-9925.test.tsx packages/plugin-detail/src/renderers/tests/record-reference-rail.rowLimitNonPositive-9925.test.tsx gives VERDICT command-exit 0, 'Test Files 3 passed (3)', 'Tests 56 passed (56)'. (5) TYPE-CHECK AND LINT on both touched packages: the first run failed with TS2307 'Cannot find module' across plugin-kanban because no workspace dist existed in the fresh worktree, which is NOT a reading. Built the closure (pnpm --workspace-concurrency=2 --filter '@object-ui/plugin-kanban^...' run build gives VERDICT command-exit 0), then pnpm --filter @object-ui/types --filter @object-ui/plugin-kanban run type-check gives VERDICT command-exit 0, both echoing 'Done'. Lint gives VERDICT command-exit 0 (plugin-kanban 226 problems, 0 errors, 226 pre-existing warnings; types likewise warnings only). (6) AFFECTED-PACKAGE SUITES: pnpm exec vitest run packages/types/ packages/plugin-kanban/ gives VERDICT command-exit 0, 'Test Files 268 passed (268)', 'Tests 5235 passed (5235)'. (7) GATES: check:new-line-citations BEFORE and AFTER both print 'Cross-file line-address citations ADDED by this branch : 0', exit 0, so the removal adds none, where a re-point to :671 would have registered as new under that gate's syntax-path-line multiset identity. check:control-bytes exit 0 (8069 files). check:test-path-roots exit 0. check:changeset-claims exit 0 (report-only). check:pending-changeset-literals exit 0. check-changeset-presence exit 0, satisfied by the PR's existing .changeset/9925-row-cap-non-positive.md; no changeset edit needed. Manual control-character grep over the four edited files: no hits. Repo-wide pnpm lint and the full shard matrix are CI's run, not mine.",
"mcp_calls": "0 - no MCP GitHub tool was called, read or write; all GitHub traffic was REST via curl, plus git push",
"api_writes": "1 - POST /repos//issues/9925/comments (this report). Reads, not writes: GET /repos//issues/9925 and GET /repos/objectstack-ai/objectui/pulls/10001. The branch update was git push, not a REST write. No PR body PATCH, no label write, no assignee write, no ready flip.",
"open_questions": [
{
"question": "The dispatch told me to correct the two docblocks' stale cross-file line address from :264 to :671. I removed the addresses instead and left the citation to the quoted expression. Ratify or revert?",
"options": [
"A, removal, which is what I shipped: AGENTS.md commandment #11 bans a cross-file path:line outright; the #8478 changeset records this exact address being removed for that reason and rewritten to cite by identifier and behavior, 'which survives a line renumbering that a bare address would not'; the differential gate reads 0 new citations, where a re-point would have counted as a new one.",
"B, re-point to :671 as the dispatch literally said: matches the dispatch text, but writes a fresh instance of the banned form, and the same address has now drifted twice (#8478 measured :365, the dispatch measures :671)."
],
"recommendation": "A, because the repo's own binding instruction file and its own recorded precedent for this exact address both prescribe removal, and because the dispatch's stated reason, that line numbers live in prose only and rot, argues for removal rather than for a third address."
}
],
"out_of_scope_findings": [
"noted, not filed: the retired '$top: schema.limit ?? DEFAULT_KANBAN_LIMIT' spelling still stands in prose at packages/plugin-kanban/src/index.tsx:426 and :570 and packages/app-shell/src/views/metadata-admin/previews/block-config.ts:237. Fenced out of scope by the dispatch, which says it is filing a separate card; that card is the successor, so I did not file a duplicate. My tree-wide census agrees with the dispatch's list exactly: no other source or test carries the string.",
"noted, not filed: .changeset/7322-object-kanban-group-by-limit.md:15 and .changeset/7772-page-block-kanban-group-by-control.md:28 also carry the retired spelling. Successor: none, by policy, since AGENTS.md forbids retroactively rewriting historical release records, so this is a permanent correct-as-history occurrence rather than a defect.",
"noted, not filed: three factual errors in the dispatch's own validation section, each measured rather than inferred, detailed in the prose report. Successor: the PM seat that writes the next dispatch. Not a repo defect, so nothing to file against the tree."
]
}

---

Generated by Claude Code