docs(changeset): date-note two pending entries whose option-description and rows refusals spec 17.3.0 lifted (objectui#10801)

[objectstack-fleet]

Fixes #10801
Clause-②: no

What this does

Two pending changesets still tell a reader that @objectstack/spec refuses options[].description and rows by name. Spec 17.3.0 declared both keys (the objectui#6140 / objectui#6153 ruling), and this repository installs 17.4.0. A pending changeset publishes verbatim into the CHANGELOG at the next release, so the text would ship stating the opposite of the installed contract.

Each file gains ONE appended dated note in the objectui#10533 form: a bold headline naming the date and the card, then what holds now, then a sentence keeping the rest as the reading of the change. The frontmatter is byte-identical (md5 of the frontmatter block, base to head: 6140 85cd680d, 7014-q1 ecedb10c, both unchanged), and zero lines are deleted (git diff is 35 insertions, 0 deletions).

File	What the note corrects (quoted, not line-addressed)
.changeset/6140-field-metadata-rows-option-description.md	"FieldSchema refuses rows BY NAME … on all four of textarea/markdown/html/richtext"; "SelectOptionSchema is strict over exactly {label, value, color, default, visibleWhen} and refuses description BY NAME"; "never in an object document sent to the platform"
.changeset/7014-q1-tier1-select-option-convergence.md	the EXTENSION paragraph's five-key claim, "refuses each of those three by name" and "must never reach an authored object document" for description; plus "exactly one key, on one face" and "SelectOption identically"

Why the 7014-q1 note also corrects "What widened". It is the same cause in the same file. SelectOptionBase derives the spec's option keys by reference, so when 17.3.0 added description the SDUI form face SelectOption gained an optional description too. The entry's "exactly one key, on one face" and "SelectOption identically" therefore undercount what ships. Nothing narrowed. The note says so in the same paragraph, and no other sentence was touched.

Placement. The claim's file surface says "appended", so both notes sit at the end of their files. In 6140 that puts the note directly after the existing 17.2.0 correction paragraph, which reads as a chronological chain. The objectui#10533 precedent (6235-mergedsort-wrap.md) put its note after the opening paragraph; only the form is copied here, not the position.

Measured, not inherited

@objectstack/spec was resolved from packages/types in this worktree: 17.4.0, and pnpm-lock.yaml has exactly one entry, 17.4.0. To date the move, 17.2.0 and 17.3.0 were installed from the registry into a scratch directory. Every probe sits beside a control that differs only by the key. Option values are at least two characters long: a one-character value fails too_small at value for an unrelated reason. The first run hit that failure and was discarded.

Probe	17.2.0	17.3.0	17.4.0
SelectOptionSchema keys	color, default, label, value, visibleWhen	+ description	+ description (catchall never, i.e. strict)
option with description	unrecognized_keys	accepted	accepted
select FieldSchema / ObjectSchema whose option carries description	unrecognized_keys at the option	accepted	accepted
option with icon (the lit control)	unrecognized_keys	unrecognized_keys	unrecognized_keys
option with disabled	unrecognized_keys	unrecognized_keys	unrecognized_keys
rows: 4 on textarea, markdown, html, richtext	unrecognized_keys	accepted	accepted
rows: 4 on text	unrecognized_keys	custom at rows	custom at rows
rows: 4 on every FieldType member	not probed	not probed	4 accept (the four above); the other 45 each add exactly one issue, custom at rows

A note on the triage reading: triage named textarea and markdown as accepting rows. The installed spec accepts it on all four multiline editor types, and the notes say four.

Form face, by type probe on the worktree sources at 17.4.0: keyof SelectOption equals label | value | color | default | description | visibleWhen | disabled | icon, and a negative control asserting the seven keys without description fails with TS2344. The convergence pin's own assertions compile (assertionFormFaceGainedOnlyDescription, assertionFormFaceKeptEveryMember, assertionMetadataFaceKeptEveryMember, assertionMetadataFaceGainedOnlyDefault). tsc -p packages/types/tsconfig.test.json --listFilesOnly lists both pin files, so the package type-check re-derives them.

Still pending: distinctive strings from both bodies (for example "refuses rows BY NAME", "The convergence is an EXTENSION", "consumed-but-undeclared field-metadata keys") have 0 hits across all 42 tracked CHANGELOG.md files. The lit control, the released entry "Adopt @objectstack/spec", hits packages/types/CHANGELOG.md.

The notes name the instruments that re-derive these facts instead of copying the answers: select-option-spec-extension-7014.test.ts covers the option key set, description accepted, icon / disabled refused, rows accepted on the four types, and custom on text / select. select-option-tier1-convergence-7014.test.ts covers the form face's gained key.

Gates (head 00a082a68, hand-derived from package.json and .github/workflows/; each exit captured by redirect-then-status)

Command	Verdict line	Exit
node scripts/check-changeset-presence.mjs	✅ No source or published contract of a released package changed in this range, so no changeset is owed.	0
node scripts/check-changeset-no-major.mjs	✅ No changeset declares a major bump.	0
node scripts/check-changeset-overwrite.mjs	⚠️ This change touches 2 changeset(s) it did not add (report-only). Declared packages are identical at base and head for both files, which is the prose-correction case its header calls legitimate.	0
node scripts/check-changeset-claims.mjs	✅ No pending changeset names a file this change touches.	0
node scripts/check-pending-changeset-literals.mjs	✅ No test source names a pending changeset.	0
node scripts/check-spec-symbol-derivation.mjs	✅ spec member citations: 1411 sources + 184 documentation pages; nothing cites a key its spec symbol does not declare.	0
node scripts/check-control-bytes.mjs	✅ check-control-bytes: OK	0
node scripts/check-changeset-fixed.mjs	✅ All workspace packages are in the changeset fixed group.	0
node scripts/check-new-cross-file-line-citations.mjs	VERDICT new-cross-file-line-citations: 0 new citation(s)	0
node scripts/check-installed-spec-pin-claims.mjs	OK (it excludes .changeset/ by design, so it is run for completeness only)	0

Tests, under the shared verify lock, from the repo root: the two .changeset/** readers node scripts/markdown-test-inputs.mjs --list names (check-changeset-no-major.test.ts, dollar-dialect-alias-census.test.ts), plus the changeset-gate suites that walk the real directory (check-changeset-presence, -claims, -overwrite, check-pending-changeset-literals, changeset-polarity-census, body-dialect-census, check-control-bytes) and the three @object-ui/types pins the notes cite or sit beside (select-option-spec-extension-7014, select-option-tier1-convergence-7014, field-metadata-rows-option-description-6140): Test Files 12 passed (12), Tests 427 passed (427), lock VERDICT command-exit 0, at 00a082a68.

No build: the diff touches no package source, and no gate above reads dist/.

Acceptance notes

• Observation, not carded, owner: none. The SelectOptionBase docblock in packages/types/src/select-option.ts lists "WHAT COMES FROM THE SPEC, unchanged" as label, value, color, default. description also arrives from the spec by reference now. The list is incomplete, not false: nothing there says description is refused. The same docblock deliberately declines to re-enumerate the keys, so this is polish.
• Scanned the other pending changesets, line-joined, for the same refusal claim about description or rows. The only other carrier is the sibling 7014-select-option-spec-claim-correction.md, which already carries its correction.

Session of this run: https://claude.ai/code/session_01DuWo5bdP9SdVebamn99GGk (the dispatching domain:ui seat #1; dev subagent).

---

Generated by Claude Code

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 00a082a686ce195841e304eb9ee4fd4fbdc5c0c3

Isolated, adversarial read of PR objectui#10828 for card objectui#10801. Inputs: the card and its three comments, the PR body, file list, diff and the head's check-runs (REST), the objectui#10533 precedent and the objectui#6140 / objectui#6153 rulings on GitHub, and the objectui tree at refs/review/pr-10828 (fetched read-only) against origin/main 12809a595 (merge-base 95bad1236; both files and packages/types/src/select-option.ts are byte-identical between the merge-base and current main, so the diff below is the whole change). The dev report on the card was treated as a set of claims to check, not as evidence.

Diff: 2 files, 35 insertions, 0 deletions, one commit. Each file gains exactly one appended paragraph.

① Derived judgments

Installed spec, read from the head. packages/types/package.json declares "@objectstack/spec": "^17.4.0"; pnpm-lock.yaml resolves exactly one version, 17.4.0 (integrity sha512-bOtSFih0…TmqoA==). I installed @objectstack/spec 17.2.0, 17.3.0 and 17.4.0 from the registry into a scratch directory; npm's 17.4.0 integrity is identical to the lockfile's, so the probes below ran on the bytes objectui installs. The registry's 17.x line is 17.0.0, 17.1.0, 17.2.0, 17.3.0, 17.4.0 with no 17.2.x patch, so "17.3.0 declared" is the exact boundary, not an approximation.

Measured, one probe beside one control per key, on all three versions (SelectOptionSchema, FieldSchema, ObjectSchema from @objectstack/spec/data, options with a two-character value):

Probe	17.2.0	17.3.0	17.4.0
SelectOptionSchema shape keys	color, default, label, value, visibleWhen	+ description (six)	+ description (six); catchall never on all three
option +description	unrecognized_keys[description]	ACCEPTED	ACCEPTED
select FieldSchema, option +description	unrecognized_keys at options.0	ACCEPTED	ACCEPTED
ObjectSchema, select option +description	unrecognized_keys at fields.status.options.0	ACCEPTED	ACCEPTED
option +icon (control)	unrecognized_keys[icon]	unrecognized_keys[icon]	unrecognized_keys[icon]
option +disabled	unrecognized_keys[disabled]	unrecognized_keys[disabled]	unrecognized_keys[disabled]
rows: 4 over all 49 FieldType members	49 refused by name unrecognized_keys[rows]	4 accepted (textarea, markdown, html, richtext); 45 add exactly one issue, custom at rows	same 4 / 45 split

The bare controls (option, select field, object, each field type without the key) parse on every version. Every spec claim in both notes and in the PR's measurement table matches this table. The PR's aside that a one-character option value fails too_small at value also reproduces on 17.4.0.

6140 note, sentence by sentence. "17.3.0 declared rows and option description" true. "this repository resolves 17.4.0" true. "17.2.0 refuses both by name, as above" true (rows by name on all 49 types, description by name). "From 17.3.0 SelectOptionSchema declares description, so an option carrying it is accepted and a select field or an object whose options carry it parses whole" true. "icon is still refused by name" true. "accepts rows on the four multiline editor types" true from 17.3.0. "on 17.4.0 every other field type refuses it by a cross-field refinement (issue code custom at rows), not by name" true (45 of 49, zero unrecognized_keys). "both may be authored there, rows on those four types only" true. "select-option-spec-extension-7014.test.ts in @object-ui/types re-derives both boundaries" true: the pin at packages/types/src/__tests__/select-option-spec-extension-7014.test.ts asserts the six-key set, description accepted on an option and through a select field, icon / disabled refused by name, rows accepted on the four types and refused with custom (not unrecognized_keys) on text / select, and the four inert editor keys still refused by name.

7014-q1 note, sentence by sentence. "17.3.0 declared it" true. "strict over six keys, description among them" true. "an option carrying it is accepted and a field whose options carry it parses whole" true. "disabled and icon are still refused by name" true. "'must never reach' now holds for disabled and icon only" true of the three keys that paragraph names. "the spec's new key reached both faces with no edit here" true by construction: SelectOptionBase extends Omit of the spec's SelectOption on visibleWhen (select-option.ts:69), the spec's SelectOption is z.input of SelectOptionSchema, whose 17.4.0 declaration carries description: z.ZodOptional of z.ZodString (absent at 17.2.0), and the form face SelectOption extends Omit of SelectOptionBase on value (form.ts:412). "the SDUI form face SelectOption gains an optional description too (nothing narrowed)" true and pinned: select-option-tier1-convergence-7014.test.ts records PRE_CONVERGENCE_SelectOption without description and asserts assertionFormFaceKeptEveryMember, assertionFormFaceGainedOnlyDescription, assertionDescriptionIsOptional and assertionMetadataFaceGainedOnlyDefault; tsconfig.test.json includes src/**/*.test.ts and is chained from the package type-check script, and Type Check is green on the head. So "exactly one key, on one face" and "SelectOption identically" do undercount, and the note is right to say so. The last sentence naming the two pins is true.

Standing sentences. Every sentence the addition makes obsolete is named by the note rather than left under the umbrella: 6140's "refuses rows BY NAME", "is strict over exactly {…} and refuses description BY NAME", "must never reach authored object metadata" and "never in an object document sent to the platform" (all named as "the sentences above that call these keys NOT a spec key and keep them out of an object document sent to the platform"); 7014-q1's whole EXTENSION paragraph for description, plus "exactly one key, on one face" and "SelectOption identically". Sentences left standing that remain true at head: 6140's "the four inert editor keys stay deliberately undeclared and are pinned so" (field-types.ts:278-282; pin lines 172-176), the 17.2.0 re-measurement paragraph (dated by version), "RichTextField and TextAreaField now read through the declared types" (only a comment mentions as any at head); 7014-q1's derivation and value widening paragraphs. Residual nit, not a defect: 6140's 17.2.0 paragraph still says the keys were aligned with "a @objectstack/spec schema member that does not exist" in the present tense; the new headline "the spec has since declared both keys" is the sentence that dates it.

Still pending. 42 tracked CHANGELOG.md files at head; 0 hits for "refuses rows BY NAME", "The convergence is an EXTENSION", "consumed-but-undeclared field-metadata keys", "Converge the two named select-option types" and "Dated note, 2026-09-27"; the lit control "Adopt @objectstack/spec" hits packages/types/CHANGELOG.md and others. Both files are still in .changeset/ at head, beside the already-corrected sibling 7014-select-option-spec-claim-correction.md.

Frontmatter. md5 of the frontmatter block, computed by me at merge-base, origin/main and head: 6140 85cd680d4d100a0510146a1d8032b965 at all three; 7014-q1 ecedb10c5189b2c8841ea736b67c3cc6 at all three. Diff hunks start at body lines 35 and 37; 0 deletions.

② Semver level

Unchanged. 6140 stays '@object-ui/types': minor and '@object-ui/fields': patch; 7014-q1 stays '@object-ui/types': minor. No major in either file, nor anywhere in the diff. Changeset Bump Policy, Changeset Declaration, Changeset Fixed Group Check and Changeset Overwrite Report are all success on the head. Nothing about what is released moves.

③ Boundary flags

Form vs the objectui#10533 precedent. 6235-mergedsort-wrap.md on main: a bold "⚠️ Dated note, DATE — headline — objectui#NNNN." then what holds now, then "The rest of this entry is kept as the reading of this change". Both notes here copy that form exactly. Two stated departures: (a) position, appended at the file end rather than after the opening paragraph (in 6140 this reads as a chronological chain after the 17.2.0 correction; in 7014-q1 it follows the closing export sentence); (b) the precedent also re-tensed each falsified sentence ("At this change …"), which this PR does not do, so a top-down reader meets the present-tense refusals before the ⚠️ marker. Both are named in the PR body and the report, and (b) is offset by the note quoting the sentences it overrides instead of only the umbrella, which is the residual the objectui#10533 review itself recorded as non-blocking. Nit, not a defect.

Clause-②: no is honest. The diff is two Markdown bodies, 0 deletions, frontmatter byte-identical; no type, schema, export, runtime path or acceptance set moves. The prose is brought to the boundary the installed spec already draws.

Dev's deviations. (1) Widening the 7014-q1 note to "exactly one key, on one face" and "SelectOption identically": correct and necessary, since the form face gained description by the entry's own derivation and both sentences would otherwise publish false in the same entry the note sits in; same file, same paragraph, same cause. (2) "Four types, not two": the triage note named textarea / markdown; measured, rows is accepted on all four multiline editor types on 17.3.0 and 17.4.0, so four is the true reading and two would have shipped an undercount. (3) Placement at the file end: judged above.

PR body. Checked against the tree and the check-runs: the quoted sentences in the "What the note corrects" table are verbatim from the files; the frontmatter md5 prefixes match; "35 insertions, 0 deletions" true; the 17.2.0 / 17.3.0 / 17.4.0 table matches my measurements cell for cell; the keyof SelectOption eight-key claim follows from the sources and the pinned assertion; the 42-file pending check reproduces; check-changeset-overwrite.mjs does call a prose correction with identical declarations legitimate (its case 2), and check-installed-spec-pin-claims.mjs does exclude .changeset/ (:393); the SelectOptionBase docblock does list only label, value, color, default as "from the spec" with description absent, and deliberately declines a longer enumeration. Nothing false found. One loose phrase: "pnpm-lock.yaml has exactly one entry, 17.4.0" (the lockfile carries a packages entry and a snapshots entry for the same 17.4.0; one version, which is the point). Local gate and test counts in the body are the implementer's words; CI is the evidence used here.

Check-runs on the head (read after they settled, 2026-09-27 ~18:35Z). 41 check-runs: 38 success, 3 skipped (Test (coverage), Test (coverage shard …/4), dependabot), 0 failure. Successes include Build & E2E, Type Check, Lint, Test shards 1-8, Test (dist pins), Spec Main Shape Gate, Control Byte Scan, Line Citation Gate, Governed Surface Queue Guard and the four changeset gates. The PR is still a draft.

Measured vs reasoned. Measured: every spec cell above (three registry installs, integrity-matched to the lockfile at 17.4.0), the lockfile and manifest versions, the frontmatter md5s at three refs, the diff stat, the CHANGELOG grep and its lit control, the 17.x version list, and the check-runs. Reasoned from source, not executed: the form-face keyof SelectOption claim (from select-option.ts, form.ts, the spec's .d.mts at 17.2.0 vs 17.4.0, and the convergence pin's type assertions that CI type-checks). Not re-run: any repo gate family.

Implemented-by: claude/issue-10801-changeset-boundary-correction
Reviewed-by: session_01DuWo5bdP9SdVebamn99GGk

VERDICT: PASS

---

Generated by Claude Code