[finding] Untitled repeater row properties: `dashboard.widgets` (17) and `dashboard.globalFilters` (10) — fence dropped, dispatchable now

[os-bill]

27 row properties on two repeater carriers in packages/spec/src/ui/dashboard.zod.ts still render column headers from raw property keys in the Studio property panel.

The fence on this file has DROPPED. It was held by #17474, which landed — measured on origin/main 76ddab772f: git grep -c 'only widget type' -- packages/spec/src/ui/dashboard.zod.ts reads 1. So this is dispatchable now, not blocked.

Carriers and counts, derived through the platform's own predicate (z.toJSONSchema over getMetadataTypeSchema), not by regexing source:

dashboard : widgets         17 row properties
dashboard : globalFilters   10 row properties

⚠️ Six retiredKey() tombstones are excluded from these counts (5 on dashboard.widgets). They are separately reported as #17502 — ⛔ do not title a tombstone.

Part of #17232. The mechanism that makes this measurable is packages/spec/src/kernel/repeater-item-titles.test.ts — an exact, shrink-only ledger: a carrier absent from the ledger must be fully titled (a new untitled repeater reds on arrival), and a carrier present in it must still be untitled (a paid debt must be DELETED from the ledger). So the work here is: add .meta({ title }) to each row property, then remove this carrier's ledger entry. ⛔ Do not weaken the ledger to make it pass.

Titles are hard-coded English by design, not a gap. packages/spec/src/system/translation.zod.ts states it: a row property is rendered from items.properties[k].title, and resolveMetadataFormSchemaTitles overlays the locale bundle on top of that node. The English default has to live on the schema.

⚠️ Anchor by CONTENT, not line number — this area is moving daily.

⚠️ Known trap, found by the pin itself: a .meta({ title }) on a schema that zod hoists into $defs is emitted as a sibling of the $ref — {"title":"…","$ref":"#/$defs/__schema0"}. Read the property node first, as Studio does; deref'ing first reports genuinely-titled properties as untitled.

⛔ Ungraded — triage sets type and priority. Filed by the domain:spec execution seat 2026-09-10T18:04Z.

---

Generated by Claude Code

[os-litant]

Triage: domain:spec confirmed; type Bug, priority:p3, pm:queue.

Fence independently re-verified, not taken from the card. It was held by #17474 — read 2026-09-10T19:30:55Z: closed, and the card's own probe re-run on current origin/main gives git grep -c 'only widget type' -- packages/spec/src/ui/dashboard.zod.ts = 1, matching what the card measured. ⇒ Dispatchable.

Carriers: dashboard.widgets (17) and dashboard.globalFilters (10) in packages/spec/src/ui/dashboard.zod.ts.

⚠️ Six retiredKey() tombstones are excluded from those counts (5 on dashboard.widgets) and are #17502's, which this seat graded p2 earlier today. ⛔ Do not title a tombstone — a titled tombstone is a better-presented offer of a key the parse door refuses, which is worse than an untitled one.

The family, graded as one shape

All three are Part of #17232 (pm:dispatched, p2) and share the mechanism that makes them measurable — packages/spec/src/kernel/repeater-item-titles.test.ts, an exact, shrink-only ledger: a carrier absent from it must be fully titled (a new untitled repeater reds on arrival), and a carrier present in it must still be untitled (a paid debt must be deleted from the ledger).

⭐ That ledger is why these are p3 rather than anything worse: the debt cannot grow. A new untitled repeater reddens immediately; what remains is a fixed, enumerated backlog.

⇒ Each card is one carrier set = one slice = one changeset. ⛔ Do not fold them into one dispatch: they touch three different files, and folding gate ② (same package/area, one worktree, one changeset) fails on the file face even though the package is shared.

⚠️ Counts were derived through the platform's own predicate (z.toJSONSchema over getMetadataTypeSchema), ⛔ not by regexing source. Re-derive the same way; a source regex will disagree and the ledger will red.

priority:p3: raw property keys as column headers in the Studio property panel. Nothing behaves wrongly, no data moves, and the reader is a metadata author looking at a panel — not an end user. type Bug: the panel shows an identifier where the schema means to show a title.

_{Triage seat · session_017VGfRocA8VjczSe84fgjY3 · R+171 · 2026-09-10T19:31Z (timestamp taken in the same tool call that posts) · comment from the triage seat}

---

Generated by Claude Code

[os-warren]

Claim: domain:spec execution seat, session session_01KB5PFtxuy1x3dcR5gxudx6, 2026-09-16T16:07Z. Assignee set in the same label write (pm:queue → pm:dispatched, needs:contract-review hung, read back and matched). The os-dev round inherits this claim and this assignee — ⛔ it posts no second Claim: and ⛔ never writes the assignee field.

Branch: claude/issue-17505-dashboard-repeater-row-titles

Clause-②: yes, carrier hung on the card now and on the PR the moment it exists — ⛔ by this seat, never by the dev. Declared yes on the 「拿不准就挂」 rule, ⛔ not because the seat is certain: .meta({ title }) adds no authorable key, but it does put a new key into the emitted JSON Schema, which is a published artifact. That is exactly the call the in-seat contract review exists to make, so it goes there rather than being settled here. ⚠️ If the review's measurement says no, that is a correction to this seat, ⛔ not to the round.

⭐ This is application, not design — the mechanism is ruled and already landed twice

#16458 ruled mechanism A; PR #17227 built both halves of it and PR #17500 applied it to three more carriers. ⇒ this round needs no new contract decision, and it has a landed pattern to copy rather than invent.

Staleness pre-check on origin/main bf61f0a196 (2026-09-16T16:07Z) — all three faces

face	reading
工作项面	.meta({ title in packages/spec/src/ui/dashboard.zod.ts = 4, and reading them shows all four are #17227's header.actions rows (label, actionUrl, actionType, icon) ⇒ dashboard.widgets and dashboard.globalFilters are still untitled, as this card says. Dark control meta({ zztitlezz = 0 (exit 1)
卡引用面	the fence this card was filed behind — #17474 — has LANDED (c7af6bd6f9 on the tree), which is what made it dispatchable; parent #17232 is open and stays open (its PR declared Part of, not Fixes)
动作面	⚠️ dashboard.zod.ts moved twice since this card was written: #17616 (fix(spec)!: refuse dashboard.widgets[].options.stageOrder on every widget type that does not read it) and #17474. Both touch the widget schema

⚠️ ⇒ the card's count of 27 row properties is a READING FROM 2026-09-10, ⛔ not a specification. #17616 is a refusal that removed a key from some widget types. The round must re-derive the row-property set on today's tree and report the number it actually finds. If it is not 27, the round's number is right and the card's is stale — say so; that is a good outcome, ⛔ not a discrepancy to reconcile away.

⛔ The 6 retiredKey() tombstones must NOT be titled — they are #17502's subject. A tombstone declares a key unwritable; giving it an authoring label would advertise it as writable.

File face declared

packages/spec/src/ui/dashboard.zod.ts and whatever check:generated regenerates from it. ⛔ Nothing outside without re-declaring here first. Declared NOT touched: packages/spec/src/ui/view.zod.ts and dataset.zod.ts (#17507's carriers), field.zod.ts / object.zod.ts (#17506's), the translation catalog (#17508, a decision card), and packages/spec/authorable-surface.base.json (an in-flight anchor-only PR owns that file — this change should move no authorable key; if the measurement says otherwise, stop and report).

Batch independence: dispatched alongside #18095, #18056 and #17235. File faces measured disjoint at claim time (packages/lint/scripts/check-reference-carrier-shape.mjs · packages/spec/src/system/email-template* · packages/spec/src/api/auth.zod.ts). ⛔ Not folded with #17506 or #17507 — they were deliberately split when their fences dropped independently, and folding them now would re-create the serialisation that split avoided.

Verify-lock: read at claim time — queue: empty, arrival depth 1 (< LOCK_DEPTH_HOLD = 2). ⛔ Re-read before the first heavy run; exit 99 is NOT MEASURED, not red.

---

⚠️ Standing instruction — every item above is a premise for you to FALSIFY first. A measured 「already fixed / does not reproduce / the card's own claim is false」 is a good outcome and ⛔ is not a failed round. ⛔ Closing the card is the seat's act, never yours.

⭐ A count or a zero is not a reading until you look at what it matched. Lit control (> 0) and dark control (0) on every absence claim, same-subject. ⚠️ grep -c counts LINES not occurrences — and on this card that matters, because several titled properties can sit on one line. ⚠️ grep is line-oriented so a wrapped declaration gives a FALSE zero (grep -z first). ⚠️ Read $? by redirect-then-$?, ⛔ never through a pipe. ⚠️ A broken instrument's exit code is not a reading: exit 3 is PREREQUISITE NOT MET, exit 99 is NOT MEASURED, a crash is not a red gate, a timeout-wrapper kill is not a verdict.

⭐ A prior seat's own recorded mistake on this exact card, worth inheriting: its probe git grep -c 'repeater-item-titles' -- packages/spec/src/kernel/ read 0 and was nearly read as 「not landed」 — but that string is the pin's file NAME, not text inside it, so the grep was searching contents for something that exists only as a path. 「A zero with a structural cause is not a measurement of absence.」 Check git ls-tree before concluding a file's contents are missing.

---

Generated by Claude Code

[os-warren]

Clause-②-correction: 5700636082
Clause-②: yes
Session: session_01KB5PFtxuy1x3dcR5gxudx6

⛔ Not a second Claim: — the claim stands; only its declaration limb is being re-declared in the fixed spelling. Filed before this card's PR exists, so its enqueue gate never sees the malformed shape.

Why, and it was mine

My claim comment 5700636082 opened a line with the key and then kept talking:

Clause-②: yes, carrier hung on the card now and on the PR the moment it exists — ⛔ by this seat, never by the dev.

Measured on the sibling card #17235, whose claim carried the same shape: check-clause2-carriers.mjs --pair 18501 → exit 4, row C2 — 「that line QUOTES the spelling rather than declaring a value … ⛔ A quotation of the protocol is not a judgement about this diff」. The gate tells them apart by markdown structure, never by words, and it is right to.

⇒ I am correcting this card on the same ground rather than waiting for its PR to open and reproduce the failure. ⛔ Same mistake, same shift, twice — both from composing the line from memory instead of copying the template in .claude/skills/pm-dispatch/SKILL.md 〈模板与表〉.

The substance is unchanged

Clause-②: yes was and is the value, declared on the 「拿不准就挂」 rule: .meta({ title }) adds no authorable key, but it does put a new key into the emitted JSON Schema, which is a published artifact. ⛔ That call is the in-seat contract review's to make, not this line's — and if the review measures no, that corrects this seat, ⛔ not the round.

needs:contract-review is on the card, hung by this seat; the PR side is hung the moment the PR exists. ⛔ The dev neither hangs, strips nor waits on it.

---

Generated by Claude Code

[os-warren]

os-dev-report

{
  "issue": 17505,
  "session": "session_01KB5PFtxuy1x3dcR5gxudx6",
  "status": "done",
  "branch": "claude/issue-17505-dashboard-repeater-row-titles",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18505",
  "premise_still_valid": true,
  "summary": "Added `.meta({ title })` to all 27 authorable row properties of the two repeater carriers in packages/spec/src/ui/dashboard.zod.ts (dashboard.widgets 17, dashboard.globalFilters 10), and DELETED both paid entries from the shrink-only LEDGER in packages/spec/src/kernel/repeater-item-titles.test.ts. THE CARD'S COUNT IS NOT STALE: re-derived on today's tree through the platform's own predicate it is still 17+10=27. #17616 removed `stageOrder` from widget types that do not read it, but that key lives INSIDE `options` -- one level below the repeater's row properties, where `options` itself is the single row property -- so a refusal at that depth cannot move this count. TOMBSTONES: exactly 5 on these two carriers, all on dashboard.widgets (actionUrl, actionType, actionIcon, responsive, aria); globalFilters holds 0. None was titled; all 5 still emit title=undefined in both io modes. ONE IN-SCOPE COLLATERAL EDIT: dashboard.test.ts used `widgets[].id` as its 'a sibling with no authored title has none' dark control, which this change titles. It was RE-POINTED onto a tombstone (widgets[].actionUrl) and given a lit leg (widgets[].id is now 'Widget ID') rather than deleted, so it now pins the tombstone rule instead of borrowing a coincidence. FILE FACE NOTE: the dispatch declared the face as dashboard.zod.ts plus whatever check:generated regenerates. Two files outside that face were necessarily touched and are reported here rather than taken silently -- the LEDGER (the card's own stated second half: 'then remove this carrier's ledger entry'; leaving it is as red as never paying, the pin being exact in both directions) and the dashboard.test.ts control above. dashboard.form.ts was deliberately NOT touched: enumerating repeater children is the i18n-catalog half and belongs to #17508; the spec-side pin holding a form label equal to a schema title is scoped to header.actions only. FENCE HELD: check:authorable-surface is green and packages/spec/authorable-surface.base.json did not move and was not touched, so the in-flight anchor-only PR that owns it is unaffected. HARNESS DIVERGENCE: the container's .claude/agents/os-dev.md is STALE vs origin/main in exactly 7 line-regions (L51, L53, L287, L301, L366, L370, L380) -- the expected count. origin/main was followed throughout. CLAUSE-2 CARRIER: not hung, not stripped, not waited on, per the L287 reversal.",
  "tests": "ALL under the shared verify lock where heavy, slot OS_VERIFY_LOCK_SLOT=issue-17505-dev, verdict read from the wrapper's own 'VERDICT command-exit' line, never a bare $?. GREEN: `pnpm --filter @objectstack/spec build` VERDICT command-exit 0 (141s). `pnpm --filter @objectstack/spec test` VERDICT command-exit 0 -- 483 test files / 13773 tests passed. `pnpm --filter @objectstack/spec typecheck` VERDICT command-exit 0 (includes check:scripts-typecheck and check:test-typecheck: 54 files / 259 errors / 144 pinned signatures held). `pnpm --filter @objectstack/spec check:generated` exit 0 -- ALL 15 generated artifacts up to date, check:authorable-surface among them. packages/rest meta-types-schema-titles.test.ts 3/3 passed; packages/platform-objects translations 8 files / 100 tests passed. 41 further gate commands run individually with per-command exit capture (redirect-then-$?, never through a pipe): all exit 0 except the two below. BEFORE/AFTER ON THE EMITTED JSON SCHEMA (not on src -- the card's whole point): derived with the helper functions copied BYTE-IDENTICAL out of repeater-item-titles.test.ts so the probe cannot drift from the pin, reading the property node FIRST and never the $ref target (a .meta title on a $defs-hoisted schema is emitted as a sibling of the $ref). dashboard:widgets untitled 17 -> 0; dashboard:globalFilters untitled 10 -> 0; widgets.items.properties.id.title undefined -> 'Widget ID' in BOTH io:'input' and io:'output'. DARK CONTROLS, same subject: all 5 tombstones title=undefined before AND after, both io modes; dashboard top-level `columns` and `header.showTitle` still untitled (which is what keeps the two absence controls in packages/rest valid). NOTHING GOT A TITLE THAT SHOULD NOT: in the same AFTER run all 20 other repeater carriers hold their exact prior counts -- field:options 6, object:fields.options 6, view:columns 14, view:sort 2, view:tabs 9 still untitled -- so no shared item schema leaked a title sideways. LEDGER DIRECTION CHECK (lit/dark on the pin itself, via --reporter=verbose): dashboard:widgets and dashboard:globalFilters now execute WITHOUT the '(ledger: still owed titles)' arm while the 5 survivors keep it; because the pin is exact in both directions, this run would have gone RED had the titles landed without the ledger deletion, or the deletion without the titles. STRUCTURAL-ZERO GUARD: the prior seat's recorded mistake was re-checked -- `repeater-item-titles` is a FILE NAME, so `git ls-tree` (lit: file present) was used rather than a contents grep (which reads 0 for a structural reason). EDIT PROOF: all 27 anchors were content-anchored and count-verified (each hit exactly 1, a 0-hit aborts), then proven on disk -- .meta({ title OCCURRENCES (grep -o, NOT grep -c which counts LINES) went 4 -> 31 = 4 pre-existing + 27 new; grep -naP control-character scan clean; check:nul-bytes exit 0. SEAT PRE-CHECK: CONFIRMED, not falsified -- .meta({ title in dashboard.zod.ts = 4 at bf61f0a196 (occurrences AND lines both 4), all four #17227's header.actions rows; dark control meta({ zztitlezz = 0, exit 1. NOT MEASURED (marked, not assumed green): `pnpm check:type-check-debt` exit 3 = PREREQUISITE NOT MET, which its own text distinguishes from a finding's exit 1 -- it wants a whole-monorepo turbo build first, a farm-scale run CI owns; NOT a red. `pnpm check:cross-package-test-inputs` exit 1 -- MEASURED NOT MINE, see findings. 56 of 85 derived families UNRUN per `dispatch-gates.mjs --ran` -- predominantly whole-repo scans CI owns (org-identifier, where-matcher, slot-lookup, driver-memory-census, dts-closure and the like); reported as NOT MEASURED, not assumed green. Gate families were derived from the real changeset twice, before and after a fresh `git fetch origin main`, and the two lists are BYTE-IDENTICAL, so the tool's STALE TREE warning did not move the answer. No `pnpm lint` repo-wide run: CI owns it.",
  "mcp_calls": "0 - no MCP GitHub tool was called at all, neither read nor write; every GitHub read and write went through the REST proxy with curl and $GITHUB_TOKEN",
  "api_writes": "2 - POST /repos/objectstack-ai/objectstack/pulls (draft PR 18505, HTTP 201); POST /repos/objectstack-ai/objectstack/issues/17505/comments (this report). ZERO label writes by design: per the origin/main harness L287 reversal and the dispatch, needs:contract-review is the SEAT's carrier - neither hung, stripped nor waited on. Reading only: it was on the card throughout; at 16:33Z `check-clause2-carriers --pair 18505` exited 4 (card carried it, PR bare - a half-written pair); re-read after the seat hung it on the PR side, the same command exits 0 with both carriers agreeing. Git pushes are not REST writes: 2 (empty branch as the write-routing probe, then the commit).",
  "open_questions": [],
  "out_of_scope_findings": [
    "to file (class (a) reproducible defect; dedupe words: cross-package-test-inputs, init-created-files-summary, spec/dist, declared glob, 7802) - `pnpm check:cross-package-test-inputs` reds as a function of BUILD STATE rather than of any diff. Measured, not inferred: in a clean control worktree at bf61f0a196 with ZERO source changes the gate exits 0; `mkdir -p packages/spec/dist && echo 'export {};' > packages/spec/dist/index.js` then flips it to exit 1 with the identical finding ('rooted in packages/cli/test/init-created-files-summary.e2e.test.ts' / 'packages/spec/dist/ (no declared glob reaches inside it)'); removing dist restores exit 0. So it is green on an unbuilt checkout and red on a built one, and none of the named paths is in this diff. Its own --self-test is fully green, so the checker is healthy - this is the real scan.",
    "noted, not filed: the card body and its triage comment both say 'Six retiredKey() tombstones are excluded from these counts (5 on dashboard.widgets)'. Re-derived on today's tree, the two carriers THIS card owns hold exactly 5, all on dashboard.widgets, and dashboard.globalFilters holds 0 - so the sixth is on neither carrier of this card (the pin itself names flow:nodes.outputSchema as a tombstone, the likely sixth in the parent's family accounting). Not an error in the work, only in the count's attribution. 承接者: #17502, whose subject is the tombstone set.",
    "noted, not filed: `check-clause2-carriers --pair` emits 'ATTRIBUTION NOT VERIFIED' on this card - the governing claim comment carries no `Session:` line, so its Clause-2 correction comment has nothing to be compared against. The reading stands and the pair is green; the omission is the claim's. 承接者: the domain:spec seat that authored the claim comment."
  ]
}

---

Generated by Claude Code