docs(changeset): correct two now-false sentences in pending @object-ui/types changesets

[os-warren]

Fixes #9042

Two pending changesets on main carry sentences that are false today and would ship
verbatim into packages/types/CHANGELOG.md at the next @object-ui/types publish. This
corrects both. Markdown only, both files under .changeset/; no source, no accept set.

Site 1 — 8992-user-actions-collapse-and-docblock.md, one word

the other ten defaulted keys to the other **seven** defaulted keys. One line.

The file contains two whole-word occurrences of "ten" and only one is wrong; the other
(all ten defaulted keys read description = undefined after the strip, line 57) is correct
and is untouched — verified after the edit, not just before.

Re-derived on this branch's base rather than taken from the card. Two independent
derivations of the resolved dependency, @objectstack/spec@17.4.0:

derivation	members	defaulted	not defaulted	the three are defaulted	"the other" defaulted
published json-schema/ui/ListView.json	11	10	1 (buttons)	yes	7
runtime, over the installed package's Zod shape	11	10	1 (buttons, def type optional)	yes	7

The seven are sort, search, filter, refresh, rowHeight, addRecordForm, editInline — exactly
the seven the file's own item 2 already enumerates, which is the self-contradiction the card
identified.

The post-strip reading behind the occurrence that must NOT move was also measured, through
packages/types/dist built on this branch:

POST-STRIP description===undefined count: 10 [addRecordForm, editInline, filter, group,
                                              hideFields, refresh, rowColor, rowHeight,
                                              search, sort]
POST-STRIP description KEPT count:        1 [buttons]
CONTROL descOf(spec.buttons) = "Custom action button IDs to show in the toolbar"
CONTROL descOf(spec.group)   = "Allow users to change record grouping fr..."

The two CONTROL lines are the live half: the reader returns real strings when a description
is present, so the ten undefined readings are a measurement rather than a broken accessor.

Site 2 — 8913-object-kanban-columns-declared.md, one paragraph

The paragraph was true when written and was invalidated when objectui#8990 landed. The
headline sentence is the wrong one, so deleting the trailing objectui#8990 clause would
have left both false claims standing.

Re-derived on current main:

claim in the old paragraph	measured now
"groupBy is required on this face"	FALSE — groupBy?: string in packages/types/src/objectql.ts; the Zod twin's node reports def type optional
"is inert … no document that passes this schema reaches it"	FALSE — a schema-valid lane-less board reaches the bare-string arm
"the requiredness is tracked as objectui#8990"	stale — that card is closed / completed

The reachability is a real parse plus the renderer's own dispatch predicate, evaluated on the
parsed document rather than read off the source:

SUBJECT  lane-less + bare-string columns : success=true
CONTROL  no record source (must FAIL)     : success=false issues=["custom@"]
CONTROL  mixed columns array (must FAIL)  : success=false issues=["invalid_union@columns"]
CONTROL  numeric lane id (must FAIL)      : success=false issues=["invalid_union@columns"]
CONTROL  groupBy present, bare strings    : success=true

PARSED groupBy value: undefined
renderer guard  schema.columns && length>0 : true
renderer guard  typeof columns[0] === str  : true
renderer guard  !schema.groupBy            : true
=> BARE-STRING ARM REACHED BY A SCHEMA-VALID DOCUMENT: true

Three refusals with distinct issue codes and a second success: the instrument discriminates,
so success=true on the subject is a reading and not a schema that accepts everything.

The rewrite says the arm is now reachable and cites objectui#8990 as the change that made it
so rather than as pending work. It keeps the original reason the arm is declared (refusing an
arm the protocol names would be a second narrowing) — that reason did not move; what moved is
that the arm is exercised rather than dormant. It also points at
packages/plugin-kanban/src/__tests__/laneLessBoard-8990.test.tsx, whose existence was checked.
The sibling 8990-object-kanban-groupby-optional.md already records the same reachability from
its own side, so after this change the two pending entries agree instead of contradicting.

Both files are still PENDING — verified with live controls

• Present in .changeset/ on origin/main: both, listed straight out of the origin/main tree.
• Absent from all 42 tracked CHANGELOG.md files: 0 files for each of six distinctive
  sentences from the two changesets.
• Live control on that same sweep, so the 0s are a reading and not a dead query:
  ObjectKanban hits 20 files, object-kanban 17, UserActionsConfigSchema 5.
• packages/types/CHANGELOG.md is 4550 lines; controls Minor Changes (27) and
  ListColumnSchema (6) hit inside that file specifically.

Neither has been released, so the correction belongs here and not in a docs-only CHANGELOG PR.

Does this PR owe a changeset of its own?

Answered by running scripts/check-changeset-presence.mjs, not by guessing:

Compared the working tree with 7f27bc543 (merge-base with origin/main): 2 file(s) changed,
0 of them published source of a package the release covers, 0 of them a manifest whose
published contract moved, 0 under a package changesets ignores, 0 changeset(s) added.
No source or published contract of a released package changed in this range, so no changeset
is owed.

Exit 0. The guarded population is a released package's published, executable source — its
src/, its index.html, files its package.json publishes verbatim — minus documentation.
.changeset/ is under no package, so a markdown-only diff there is outside the population
entirely and owes nothing.

Gates run locally

gate	exit	note
check-changeset-presence.mjs	0	no changeset owed, quoted above
check-changeset-no-major.mjs	0	no major declared
check-changeset-fixed.mjs	0	all workspace packages classified
check-changeset-overwrite.mjs	0	report-only; see below
check-governed-queue-guard.mjs --test	0	NOT GOVERNED, 2 paths vs 5 surfaces; control AGENTS.md returns exit 3 GOVERNED

check-changeset-overwrite.mjs does report this PR, expectedly and non-blockingly — it
reports any change that modifies a changeset it did not add:

0 changeset(s) added, 2 modified, 0 deleted.
   M  .changeset/8913-object-kanban-columns-declared.md
          declared at base: @object-ui/types: minor
          declares now:     @object-ui/types: minor
   M  .changeset/8992-user-actions-collapse-and-docblock.md
          declared at base: @object-ui/types: minor
          declares now:     @object-ui/types: minor

That gate's own header names this exact case as legitimate — "factual corrections to prose" —
and its narrower signal, a lost package declaration, does not fire: declared at base and
declares now are identical on both files, which is the 18-for-19 shape its history measured.

Labels

⛔ needs:contract-review is deliberately NOT hung on this PR. The card carries
Clause-②: no and carries no such label; h31ContractReviewCarrierSplit in
scripts/pm/check-half-states.mjs returns a finding only when the two carriers disagree
(cardGated && barePrs.length > 0, or !cardGated && gatedPrs.length > 0) and returns null
when both are absent. Hanging it here would manufacture the second, more dangerous half of that
split. Verified by reading the function, and the card's labels were re-read immediately before
opening this PR.

The declaration holds against the diff: 2 paths changed, 0 outside .changeset/, 0 not
markdown — no accept set moves and no exported symbol changes.

Sibling sweep

Fixed here: the two files named on the card. Nothing else is touched.

Reported, not fixed:

• 7 pending changesets carry a tracked as #N forward reference whose card is now
  closed/completed (#6011, #7170, #7188, #7708, #8166, #8983, #8990). One of those is
  this PR's own site 2. The other six read as a citation whose implication of pending work is
  stale; ⛔ I did not verify whether each surrounding claim was actually falsified, and none of
  them carries a falsified headline claim of the kind site 2 had. They are a report, not a fix,
  and not this PR's one-line class.
• Checked and found still true on main: 9012-core-spec-floor.md (core declares
  @objectstack/spec: ^17.3.0), 8604-record-details-columns-enum.md
  (columns?: '1' | '2' | '3' | '4' in record-components.ts, while the dashboard columns
  one level away stays number), and 8990-object-kanban-groupby-optional.md.
• No other pending changeset asserts groupBy requiredness or the inert/unreachable claim
  (1 match each, both this PR's file), against controls of 25 files mentioning groupBy and 67
  mentioning inert. No other pending changeset carries the miscount class (the other ten,
  defaulted keys, eleven members, UserActionsSchema — 1 file each, this PR's), against
  controls of 43 files containing all ten and 393 containing keys.

The PR body of #9017 carries the same "the other ten" sentence. It is merged, so its body is
cosmetic and reaches no CHANGELOG; ⛔ I did not edit it, and it is not required.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01Jmxdo7bmeqCQHLSfmLVX9w

---

Generated by Claude Code

[os-warren]

Provenance — landing on a Clause-②: no declaration, ⛔ no contract-review carrier owed

Head e70158bc7ac340852f2b04f97f0516cf2e9eff01, 2 files, +15/−6, both under .changeset/.

⛔ No in-seat contract review is owed here and none was run. The declaration on card #9042 is Clause-②: no, which is what decides whether the PR must pass in-seat contract review; both carriers are correctly absent (card and PR), so --pair agrees at absent rather than at present. That is the dual-carrier rule working, ⛔ not a skipped gate.

Landing pre-checks — measured, and one of them is HONESTLY NOT MEASURED

check	reading
② check-clause2-carriers.mjs --pair 9064	rc 0 — declaration readable, both carriers agree
② check-widening-tells.mjs --declaration no --files	rc 0, but ⛔ NOT MEASURED — see below
③ every check on the head, ⛔ not a required subset	33 check runs, 30 success + 3 skipped, 0 failed, 0 pending; total_count 33 = array length 33
governed surface	--pr objectstack-ai/objectui#9064 → 0 of 2 paths hit the register ⇒ NOT governed
mergeability	mergeable: true, mergeable_state: clean

⚠️ The widening-tells gate returned rc 0 and told me its own rc 0 is uninformative, in its own words: "⛔ NOT MEASURED is not a clean reading — no tell could have fired on these files whatever they contain: no declared surface covers it (2)." .changeset/*.md is not a declared surface. ⛔ So I am not reporting that gate as clearing this PR. What the no actually rests on is that the diff is markdown-only under .changeset/ — 2 files, 0 paths outside it, 0 non-markdown — which is checkable by inspection and which I confirmed on the diff.

Both edits verified by content, ⛔ not by the report

Site 1 is one line, and the occurrence that must NOT move is visible in the same hunk's context, untouched:

-   supplied descriptions the other ten defaulted keys did not have.
+   supplied descriptions the other seven defaulted keys did not have.

with all ten defaulted keys read description = undefined sitting four lines above it, unchanged. Eleven members, buttons the only undefaulted one ⇒ ten defaulted; the three this PR's sibling moved are among them ⇒ the other ones are seven — which is also exactly what the file's own item 2 enumerates.

Site 2 is the paragraph, and the rewrite is better than what I specified. I asked for the headline sentence to be corrected; it also kept the historical claim true rather than deleting it — "groupBy was required on this face when this entry was written" — then names the mechanism (effectiveColumns takes the string branch under if (!schema.groupBy)), cites objectui#8990 as the change that made the arm reachable rather than as pending work, preserves the unchanged reason the arm is declared, and points at the pin carrying a firing control.

Two corrections to my own dispatch boilerplate, both from the dev, both right

1. check-widening-tells.mjs does not exist in objectui. My brief said a Clause-②: no puts it live here; it lives in objectstack and is wired into objectstack's lint.yml (objectui has 0 wiring hits against a live control of 2 for check-half-states). It is a real gate for this seat's landing decision, run from the objectstack clone — as above — but ⛔ it is not a red a dev in objectui can ever wait for, and telling one to expect it is a defect in the brief. Corrected.
2. h31ContractReviewCarrierSplit fires only when the two carriers DISAGREE, and returns null when both are absent — which the dev established by reading the script rather than inferring it. That confirms the correction I posted on card docs(changeset): two PENDING changesets carry now-false sentences — a wrong key count in 8992-…md and a paragraph in 8913-…md invalidated by objectui#9021; both ship into the types CHANGELOG at the next publish #9042 and confirms the original Claim: clause was wrong.

⭐ And one method note worth keeping: the dev's own line-based grep for the target sentence returned a false 0 before the edit, because the phrase spans a line break — it read exactly like "the clause is already gone". What proved presence was a multi-line anchor assertion with a count. That is the reassuring-output hazard the brief warns about, caught by the person it would have fooled.

PR flipped ready and enqueued. ⛔ Base is not hand-merged — the queue rebuilds on current main itself.

---

Generated by Claude Code