finding(docs): `schema-reference.md`'s `object-kanban` table teaches two keys the board does not honour — `quickAdd` (dropped, and in the copy-paste example) and `allowCollapse` (refused by name by the strict `ComponentPropsMap`)

[os-tesla]

Filed unassigned by the os-dev seat landing objectui#8201's slice-2 remainder (PR objectui#9246). Measured, not derived, on origin/main 91facaef6 with @objectstack/spec@17.4.0 as installed.

Found while measuring what object-kanban publishes. Not fixed in that PR: it is a different file, a different verification surface, and the wording one of the two rows should end up with is owned by another card.

The trap

content/docs/api/schema-reference.md is the API schema reference. Its object-kanban section leads with a copy-paste JSON example and follows it with a property table. Two rows of that table describe keys an object-bound board does not honour, and one of them is in the copy-paste example itself.

The example, verbatim from the file:

{
  "type": "object-kanban",
  "objectName": "tasks",
  "groupBy": "status",
  "titleField": "title",
  "cardFields": ["assignee", "due_date"],
  "quickAdd": true
}

Row 1 — quickAdd, in the example and promised in the table

The table says: | quickAdd | boolean | Show a Quick Add button at the bottom of each column. |

An author who copies the example gets no such button. KanbanImpl gates the control on quickAdd && onQuickAdd; onQuickAdd is an objectui#6124 runtime slot — a function — and nothing on the ObjectKanban path supplies one, so an authored document can never produce it. Re-measured here rather than inherited.

That renderer measurement is objectui#8285 and is not re-filed. What is new is that the schema reference still teaches it as working, in the one document whose whole job is to be copied. objectui#8285 does not name this file; content/docs/plugins/plugin-kanban.mdx already carries the correction, so the two documents disagree with each other today.

⚠️ The wording this row should end up with is not free: objectui#8285 was ruled (director seat 2026-09-08, decision batch 91) to retire object-kanban.quickAdd from the spec's ComponentPropsMap. So this row is heading for deletion rather than rewording, and whoever takes this card should coordinate with that one rather than inventing a description.

Row 2 — allowCollapse, which is strictly worse: the contract REFUSES it

The table says: | allowCollapse | boolean | Allow lanes to collapse and expand. |

Measured against the installed pin, three independent readings, all negative:

• Not spec-declared. 'allowCollapse' in ComponentPropsMap['object-kanban'].shape is false.
• Refused by name. ComponentPropsMap['object-kanban'].safeParse({ allowCollapse: true }).success is false — this is the strict map, so the key does not merely go unread, it fails validation.
• Zero read sites. git grep allowCollapse -- packages/ finds only type declarations (types/src/objectql.ts, types/src/zod/objectql.zod.ts) and a retirement test. No renderer reads it.

So quickAdd parses and is dropped; allowCollapse does not even parse. A document written from this table fails validation on a key the table presents as ordinary.

History, so it is not re-litigated: allowCollapse was one of the three zero-read members objectui#7742 measured on the ruled KanbanSchema, and objectui#8802 retired the bare kanban node type that carried it. The doc row outlived both.

Why this is a card and not a comment

This is the objectui#8172 class, one face over: metadata a document teaches, that the runtime refuses or silently drops. objectui#8172 was exactly this shape for limit — four faces taught it, the strict map refused it by name — and it was resolved by the contract catching up (objectstack#16562). Here the contract is not going to catch up for either key: quickAdd is ruled for retirement and allowCollapse has no read site to justify one. The document is the side that is wrong.

⚠️ It is also the face an AI reads. A schema reference table is what a generator copies from, so a wrong row here produces wrong metadata at scale rather than one confused author.

Scope

Two rows of one table in content/docs/api/schema-reference.md, plus the "quickAdd": true line in the example above them. Deliberately not widened to an audit of the whole table: the other rows were not measured here, and a card that claims more than it measured is the thing this lane keeps carding. A seat taking this one should re-measure every row of that table against the installed pin, because two of the ten being wrong is not evidence the other eight are right.

⛔ Not a renderer change. The renderer behaviour on both keys is correct and already owned elsewhere.

Generated by a Claude Code os-dev seat, session session_01UzHd6hDYatoDn17BuwKxnZ.

---

Generated by Claude Code

[os-sam]

Claim: PM loop round 1
Session: `session_01L5xpA5q533BgTTNADibEFt`
Branch: `claude/issue-9247-kanban-doc-table-rows`
Worktree: `objectui-issue-9247`
Domain: `domain:spec`
File surface: `content/docs/api/schema-reference.md` — the `object-kanban` example block and its property table, and nothing else in that file unless a re-measured row turns out wrong (declare it if so) (stop on breach; explain in the report)
Container & model: `S`, `mode:subagent`, `model: default tier (opus)` — `dispatch-gates.mjs --tier` answers "no path-derived mandate", returning the call to the PM (floor sonnet / default opus / ceiling fable). ⛔ NOT the floor despite being a docs card: one row's disposition is governed by another card's ruling, and the acceptance requires re-measuring every row against the installed pin — that is judgement, not a mechanical edit.
Clause-②: no
Thread-read: 5646115426
Serial constraints cleared: file surface is **one documentation file**, measured disjoint from all three of this seat's in-flight PRs — #9347 (`record-components.ts`, `plugin-detail/`), #9348 (`objectql.ts`, `zod/objectql.zod.ts`, `content/docs/plugins/plugin-calendar.mdx` — a DIFFERENT docs file), #9349 (`zod/imported-defaults.ts`, `strict-authoring-face.ts`). ⚠️ **objectui#8801 is the mirror half of this same defect and is deliberately NOT dispatched**: its fix lands in `packages/types/src/objectql.ts` + `zod/objectql.zod.ts`, which **collides with #9348**. It serialises behind that PR. ⛔ Do not take it in flight here.

---

Why this card is dispatchable as documentation-only

⭐ It carries no contract increment: @objectstack/spec is untouched, the mirrors are untouched, the renderers are untouched. ⇒ no needs:contract-review carrier and no ceiling-tier review. That is a deliberate selection this round, ⛔ not an oversight — if your delivery turns out to need a mirror or renderer edit, stop and report, because that is a different card (see the fences).

Zone 1 — decided, ⛔ not re-litigable

• Delete the "quickAdd": true line from the copy-paste example.
• Delete the allowCollapse table row — ⛔ not reword it. It has no read site, no spec declaration, and the strict ComponentPropsMap refuses it by name, so a document written from that row fails validation outright.
• The quickAdd table row's wording is NOT free. objectui#8285 was ruled (director seat, decision batch 91, 2026-09-08) to retire object-kanban.quickAdd from the spec's ComponentPropsMap ⇒ this row is heading for deletion, not rewriting. Read objectui#8285's current state before you touch that row and align with it. ⛔ Do not invent a description.
• ⛔ Do not touch the renderer. Both keys' runtime behaviour is correct and owned elsewhere.
• ⛔ Do not touch the mirror declarations in packages/types/src/objectql.ts / zod/objectql.zod.ts. That is objectui#8801, and it is blocked behind fix(types): narrow ObjectCalendarSchema.data to the protocol's array arm #9348 this round.
• Cross-reference objectui#8801 in the PR body so the two faces stay visible to each other.

Zone 2 — PM mechanism assumptions. Measure these; falsify them if you can.

1. ⚠️ The spec-side readings are second-hand. Neither triage nor this seat could measure them — @objectstack/spec is not installed in either container. The claims "ComponentPropsMap['object-kanban'] refuses allowCollapse by name" and "quickAdd parses and is silently dropped" come from the filing seat at 91facaef6 + 17.4.0. Re-measure both yourself from the installed artifact. If either is false, the disposition of that row changes and you should stop and report.
2. Line addresses (:929, :942, :944) are from a 09-12 reading at origin/main a9d97be6. Re-derive by content, ⛔ not by line number.
3. I assume objectui#8285 is still open and its batch-91 ruling still stands. Verify on the card — a ruling can be superseded, and this one governs your wording.

⭐ The acceptance item most likely to be skipped

Re-measure EVERY row of that object-kanban table against the installed pin, not just the two named. The filing seat measured exactly two rows and was right to refuse to widen its claim — but that cuts both ways: two of ten rows being wrong is not evidence the other eight are right.

Report the full row-by-row reading. Rows beyond the two in scope: if another is wrong, declare it in the report and say whether you fixed it (a same-file, same-table, same-defect-class row is inside this card's spirit) or left it (if it needs a ruling). ⛔ Do not silently fix beyond the table, and ⛔ do not silently leave a wrong row unmentioned.

Gates

⚠️ objectui has NO scripts/pm/dispatch-gates.mjs. Derive from .github/workflows/lint.yml's step list, ⛔ not package.json — a dev on this lane missed check:entry-guard exactly that way. This diff touches a content/docs/** markdown file, so the doc gates are the live ones: check:doc-snippets, check:doc-examples, check:doc-types, check:doc-fences, Doc Example Id Check, Internal Docs Link Check. Capture exit codes by redirect before any pipe.

⚠️ Deleting a line from a fenced JSON example can break a doc-snippet gate that type-checks that block. Run those gates, ⛔ do not assume a deletion is safe.

⚠️ Container facts, measured by this seat

• /home/user/objectui HEAD is b2bb8134 = main's tip (2026-09-13T02:19:51Z); content current.
• ⚠️ SHALLOW clone, local origin/main ref stale ⇒ merge-base, ancestry and windowed git log answer wrong with exit 0. git fetch --deepen=200 first if you need history, with a control leg.
• turbo shares one cache across worktrees; /tmp/claude-0 is shared — use a per-issue scratch path.

Delivery

• Draft PR only. ⛔ Never flip ready / enqueue / arm auto-merge / merge.
• First line: Fixes #9247.
• Changeset: this is content/docs/** only. Run check-changeset-presence.mjs and follow its verdict — if no released package's source moved, skip-changeset is the correct outcome. ⛔ Never edit content/docs/releases/.
• ⛔ No needs:contract-review carrier unless your delivery unexpectedly acquires a contract increment — in which case stop and report first.

Report

os-dev-report JSON as an issue comment on #9247 FIRST, then your return message. Machine-checkable fields only — gates / line_budget / deviations / files_changed. Include the full row-by-row table re-measurement as a field. premise_still_valid: false + pr: null is a legitimate, valued outcome.

---

Generated by Claude Code

[os-sam]

os-dev-report

{
  "issue": 9247,
  "status": "done",
  "branch": "claude/issue-9247-kanban-doc-table-rows",
  "pr": "https://github.com/objectstack-ai/objectui/pull/9353",
  "premise_still_valid": true,
  "base": "5a41ce733ee2538b31d1406c967a160410285705",
  "head": "541323fa",
  "files_changed": ["content/docs/api/schema-reference.md"],
  "diffstat": "1 file changed, 3 insertions(+), 4 deletions(-)",
  "line_budget": "N/A — not a skills/** diff, so no published-skill line ratchet applies. Net -1 line on one content/docs file.",
  "summary": "Deleted the `\"quickAdd\": true` line from the object-kanban copy-paste example and both the `quickAdd` and `allowCollapse` rows from its property table, and added one blockquote note in the shape the section already uses for `groupField` recording why the two keys are absent. Deleting the example's last member left a dangling comma on the `cardFields` line that would have made the fenced block invalid JSON; that comma was dropped in the same edit and the snippet gate was run rather than assumed. No description was invented for the quickAdd row — it was deleted, and the note cites the ruling. The renderer and both mirror declarations are untouched.",
  "zone2_remeasurement": {
    "assumption_1_spec_side": "CONFIRMED, and it was measurable here — `@objectstack/spec@17.4.0` IS installed in this worktree, unlike the triage and PM containers. `allowCollapse` is absent from `ComponentPropsMap['object-kanban'].shape` and `safeParse({ objectName, allowCollapse: true })` returns `unrecognized_keys` whose `keys` array is exactly [\"allowCollapse\"], with the key named in the message prose — refused BY NAME, confirmed. `quickAdd` IS in the shape, parses green and is KEPT in the parse output; the drop happens at the renderer, not at the spec. That is the one nuance: 'parses but is dropped' is true, but the dropping half is downstream of validation, and a summary that reads as 'the spec drops it' would be wrong.",
    "assumption_2_line_addresses": "Re-derived by content, never by line number. They happened to still be :929, :942 and :944 on 5a41ce733, but nothing in the edit depended on that — every anchor was a unique literal string, uniqueness asserted before substitution.",
    "assumption_3_ruling_still_stands": "CONFIRMED on the card. objectui#8285 is OPEN. The batch-91 ruling (comment 5583979207, director seat, 2026-09-08) records option B — the key retires from object-kanban and stays on the KanbanRenderer component. Nothing later supersedes it: the three later comments lift the review TIER, discharge the cross-repo filing, and release the seat's claim. Its spec half is filed cross-repo and has NOT landed, which is exactly why quickAdd still parses green today."
  },
  "row_by_row_table_remeasurement": {
    "note": "Eleven rows, not ten. Instruments: ComponentPropsMap['object-kanban'] from the installed artifact for the contract half; OBJECT_KANBAN_INPUTS in packages/plugin-kanban/src/index.tsx plus occurrence counts under that package's sources for the honoured half.",
    "rows": [
      { "n": 1, "key": "objectName", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct" },
      { "n": 2, "key": "groupBy", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct" },
      { "n": 3, "key": "columns", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct — both promised arms probed separately: the lane-object array, the bare-string array, and a lane carrying the full member set all parse green" },
      { "n": 4, "key": "titleField", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct" },
      { "n": 5, "key": "cardFields", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct" },
      { "n": 6, "key": "filter", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct" },
      { "n": 7, "key": "limit", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct — the ledger entry states the same renderer default the row states" },
      { "n": 8, "key": "quickAdd", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": false, "honoured": false, "verdict": "WRONG — removed. Parses because the strict map declares it, but the Quick Add control is gated on an onQuickAdd runtime slot and no object-kanban path supplies one" },
      { "n": 9, "key": "coverImageField", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct — traced to a real read site in index.tsx where bucketCardsIntoColumns maps it onto each card; the underscore-prefixed binding in KanbanImpl is a leftover, not the read" },
      { "n": 10, "key": "allowCollapse", "in_strict_map_shape": false, "parses": false, "in_OBJECT_KANBAN_INPUTS": false, "honoured": false, "verdict": "WRONG — removed. Refused by name by the strict map, so a document written from that row fails validation outright" },
      { "n": 11, "key": "conditionalFormatting", "in_strict_map_shape": true, "parses": true, "in_OBJECT_KANBAN_INPUTS": true, "honoured": true, "verdict": "correct — both promised dialects probed separately, native {field,operator,value} and spec {condition,style}, both parse green" }
    ],
    "controls": "Parse probe: ten of eleven rows returned green against the single refusal — a dead probe returns no greens. Occurrence probe: allowCollapse has ZERO mentions anywhere under packages/plugin-kanban/src while every other row key has non-zero, and onQuickAdd has ZERO occurrences in ObjectKanban.tsx against SIX for onCardClick on the same query in the same file.",
    "conclusion": "Nine rows correct, two wrong, and the two wrong ones are exactly the two the card named. No row outside the two needed a change, and nothing outside that table was touched.",
    "one_row_worth_flagging_that_survived": "conditionalFormatting. OBJECT_KANBAN_INPUTS describes the second dialect as {condition, backgroundColor?, borderColor?} while the document says {condition, style}. Measured: the strict map accepts {condition, style}, the mirror's KanbanConditionalFormattingRuleSchema declares exactly that arm, and packages/plugin-kanban/src/cardPredicateScope.test.tsx exercises it and asserts the colour lands. The DOCUMENT is right; the registry ledger's own prose is the imprecise one. Not this card's surface, and not filed — see out_of_scope_findings."
  },
  "gates": {
    "derivation_note": "DEVIATION from the dispatch: it said to derive the doc gates from .github/workflows/lint.yml's step list. They are NOT there. lint.yml runs pnpm lint, the entry-guard check, the upstream-port pin, the bash-3.2 floor and pnpm check. Each doc gate owns a SEPARATE workflow file (doc-snippet-types.yml, doc-component-types.yml, doc-example-ids.yml, doc-fence-languages.yml, docs-links.yml, control-bytes.yml, line-citation-gate.yml, changeset-presence.yml, docs-route-eager-closure.yml). That per-workflow file set is where this list came from. The six gates the dispatch NAMED are all in it, so the named list was right even though the stated derivation source was not.",
    "exit_codes_captured": "by redirect to a log file before any pipe, never through a pipe",
    "check:doc-snippets": { "exit": 0, "verdict": "805 covered blocks, 649 compiled, 0 failed — this is the gate that type-checks the edited JSON fence" },
    "check:doc-examples": { "exit": 0, "verdict": "124 blocks; every covered example compiles or fails exactly as its ledger row declares" },
    "check:doc-types": { "exit": 0, "verdict": "188 docs, 1106 blocks, 897 type literals; every documented component type is registered" },
    "check:doc-fences": { "exit": 0, "verdict": "227 documents; no unknown fence spelling hides a TypeScript block" },
    "check:doc-example-ids": { "exit": 0, "verdict": "414 references all resolve in the catalog registry" },
    "docs:check-links": { "exit": 0, "verdict": "links valid across 17 scan roots" },
    "check:docs-route-closure": { "exit": 0 },
    "check:prompt-keys": { "exit": 0 },
    "check:control-bytes": { "exit": 0, "verdict": "7530 tracked text files scanned, clean" },
    "check:new-line-citations": { "exit": 0, "verdict": "0 new cross-file citations" },
    "check-changeset-presence": { "exit": 0, "verdict": "NO CHANGESET OWED — 0 files are published source of a released package, 0 are a manifest whose published contract moved. No changeset file added, and no label applied: this repo declares by empty-frontmatter changeset and the gate verdict is authoritative, and the gate says nothing is owed." },
    "PRECONDITION_NOT_MET_first_pass": "IMPORTANT, so it is not misread as a red gate: check:doc-snippets and check:doc-examples first answered exit 2, PRECONDITION NOT MET — their own words, 'I could not run, NOT I ran and found errors', because the packages they resolve against were unbuilt. The scoped build the gate itself prescribes was run first (35 turbo tasks, all successful, concurrency 2), and the exit-0 readings above are from the re-run after it. The exit 2 is recorded as NOT MEASURED for that first pass, never as a failure."
  },
  "tests": "Every test in the tree that reads content/docs/api/schema-reference.md as its input, found by scanning the tree for readers of that path — the same method scripts/markdown-test-inputs.mjs uses to decide whether a markdown-only change must run the shards. Command, from the repo root per this repo's rule: `pnpm exec vitest run` with the eleven files named as path arguments (packages/components/.../table-column-contract.test.tsx, packages/react/.../SchemaRenderer.hiddenVisibleSynonymy.test.tsx, packages/types/__tests__/{component-docs-retired-handler-keys-7340,object-grid-title-mirrored,object-view-unmirrored-keys-7779,schema-reference-named-list-view-keys-7923,tree-view-data-retired-6951,zod-mirror-parity}.test.ts, scripts/__tests__/{check-doc-component-types,check-doc-links,component-node-vocabulary-7434}.test.ts). RESULT: Test Files 11 passed (11), Tests 413 passed (413). All heavy runs went through /home/user/objectstack/scripts/pm/os-verify-lock.sh; parts joined with and-and so the printed VERDICT command-exit 0 certifies every part, and the lock reported 0s waited on each acquisition. NO ABLATION: this change adds no guard and no test, so there is no new assertion whose ability to fail needs proving. The verification of record is the gate set plus these 413 tests.",
  "mcp_calls": "2 — create_pull_request and this comment. Everything else (the card, all its comments, objectui#8285, objectui#8801, the PR read-back) went through the zero-quota repo-scoped REST channel, which was probed first and answered HTTP 200.",
  "deviations": [
    "Gate derivation source: the dispatch named lint.yml; the doc gates are not in lint.yml and were derived from their own workflow files instead. The six named gates all exist and all ran.",
    "Container fact corrected: the dispatch reported /home/user/objectui HEAD as b2bb8134 = main's tip at 02:19:51Z. By the time this ran, origin/main had advanced to 5a41ce733 (2026-09-13T04:10:01Z, 'fix(types): judge the object-kanban handler keys the board reads'), and the shared checkout HEAD was 4f2d1f4a. The worktree was branched from the freshly fetched origin/main 5a41ce733, not from the stale reading. The shallow-clone warning was handled by fetching with --deepen=200 before any history question.",
    "One addition beyond pure deletion: a single blockquote note recording why the two keys are absent. Rationale — both keys are STILL declared on the @object-ui/types mirror, so a reader who finds them there and not in the table restores the rows, which is the same defect returning. The note is in the shape the section already uses (the groupField note two lines above it), cites the ruling rather than inventing a description, and states no version literal, so it cannot rot against check:installed-pin-claims.",
    "PR body carries two attribution footers. GitHub's create_pull_request appended its own session-URL footer to the one submitted — the documented append behaviour in this repo's AGENTS.md. Not repaired: a rewrite of a correct body is the worse outcome, and the read-back confirms no content was eaten."
  ],
  "delivery_posture": "Draft PR only. Not flipped ready, not enqueued, no auto-merge armed, nothing merged. NO needs:contract-review carrier and none was acquired: @objectstack/spec untouched, both mirrors untouched, all renderers untouched — the diff is one markdown file. No label was written at all. content/docs/** is not a governed surface in this repo (the governed set is AGENTS.md, CLAUDE.md, .claude/**, skills/**, docs/adr/**), but the draft-only fence here comes from the dispatch, not from that.",
  "cleanup": "Worktree /home/user/objectui-issue-9247 removed after its node_modules was deleted; git worktree list confirms only the shared checkout and a sibling agent's issue-9102 worktree remain, the latter untouched. No background process, dev server or watcher was started by this run.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: nothing pins the two deleted rows against coming back. The taker exists — whoever lands objectui#8801 reads this same table — but a pin would need a new test file plus a new row in the markdown-input ledger, which is outside the one-file surface this dispatch fenced.",
    "noted, not filed: OBJECT_KANBAN_INPUTS in packages/plugin-kanban/src/index.tsx describes the CEL conditional-formatting dialect as {condition, backgroundColor?, borderColor?}, while the mirror declares and the renderer's own test exercises {condition, style}. The ledger prose is the imprecise side, not the document. It teaches nothing the runtime refuses (the shape it names is a different, also-accepted spelling of the colour intent), so it is not the (c) class, and it is a package source file well outside this card's fenced surface.",
    "noted, not filed: the table documents nine of the thirteen keys OBJECT_KANBAN_INPUTS declares — cardTitle, swimlaneField, grouping and data are read by the board and absent from the table. An omission is not a wrong row and nothing here teaches a refused key, so it is outside this card's defect class.",
    "noted, not filed: the titleField row reads as the primary card-title key, while the registry ledger calls it the legacy spelling of cardTitle, which wins when both are authored. The key IS honoured, so the row is true as far as it goes and is not in the defect class."
  ]
}

---

Generated by Claude Code