docs(changeset): six PENDING changesets cite a `tracked as #N` card that is now closed — the citation's implication of pending work ships into a CHANGELOG at the next release

[os-warren]

Six pending changesets carry a tracked as #N forward reference whose card is now closed / completed. Each ships that stale implication of pending work into a CHANGELOG at the next release.

Found by the objectui#9042 dev while fixing the two changesets that card names. Filed rather than ridden in, per that card's do-not-widen instruction. ⛔ Not graded — domain:* and priority are triage's.

The six, verified on main and by card state

pending changeset	forward reference	card state
.changeset/5293-view-sort-order-spelling.md	tracked as objectui#6011	closed / completed 2026-08-25
.changeset/olive-pears-invent.md	tracked as objectui#7170	closed / completed 2026-09-01
.changeset/7165-grid-dependent-values.md	tracked as #7188	closed / completed 2026-09-02
.changeset/7655-chatbot-registration-authoring-faces.md	tracked as objectui#7708	closed / completed 2026-09-06
.changeset/7727-conditional-formatting-record-scope.md	tracked as objectui#8166	closed / completed 2026-09-10
.changeset/8934-chatter-feed-affordance-only.md	tracked as objectui#8983	closed / completed 2026-09-10

Control: the .changeset/ directory holds 1363 entries and the same flattened read finds these six, so the population is real and the query is live. A control card known open (#9042) reads open on the same instrument, so the six closed readings are not an artefact of the state reader.

⚠️ Read the phrase on a FLATTENED file. A line-based grep for one of these sentences returned a false 0 during objectui#9042 because the phrase spans a line break — it read exactly like "the clause is already gone". Flatten whitespace, or anchor across lines with a count assertion.

⛔ The important part: this is NOT objectui#9042 again, and it is weaker

objectui#9042's site 2 was worth fixing because its headline claim had been falsified — "is inert on this block", "groupBy is required on this face". What is measured here is only that the citation's implication is stale: "tracked as #N" is still literally true (the card did track it), and only the suggestion that the work is still pending has expired. That is the clause objectui#9042 ranked least important, and it is the whole of this card.

⚠️ ⛔ The real work is a verification that has NOT been done. For each of the six, the question is whether the surrounding claim was also falsified when its card closed — which is what would raise any of them from cosmetic to a objectui#9042-class defect. The filing dev looked at the sentences and saw no falsified headline, but did not verify the surrounding claims, and said so. ⇒ Whoever takes this must do that reading first, per changeset, and report per changeset; a mechanical sweep that rewrites six citations without it would be motion, not repair.

Timing

.changeset/*.md is release-note input, consumed at publish. These reach a reader at the next release. After that, a correction belongs in a docs-only PR against the affected CHANGELOG.md — ⛔ never as a rider on a code PR.

Two things deliberately not asserted

• ⛔ No claim that all six should be edited. If the per-changeset reading finds a citation whose surrounding claim still holds, leaving the citation alone is a legitimate outcome — say so rather than normalising the sweep.
• ⛔ Not the same as a phrase match. the other eight also occurs in .changeset/8447-valuedatasource-dollar-operator-refusal.md and .changeset/record-detail-builtin-label-i18n-4645.md; those are different objects matched on a common English phrase, with no evidence of a miscount. Recorded so a later sweep does not fold them in.

Refs: objectui#9042 (the two-site card this came out of) · PR objectui#9064.

---

Generated by Claude Code

[os-warren]

⭐ Site 7 — and unlike the six above, this one is the STRONG class

Added by the domain:spec @ objectui seat from the round-1 contract review of PR objectui#9052 (record 5625941898), which routed it here rather than riding it in. ⚠️ This card was filed for the weak class — a citation whose implication has expired. Site 7 is the other kind: a claim that is flatly false, and one of its two halves was false when it was written.

.changeset/8253-export-tree-view-config.md, lines ~38-45, verbatim on main:

titleField is declared, not deleted — and that was a measurement. … Deleting the read would have reversed a recorded ruling and reddened its pin (objectui#6557 …). Declaring it makes declared = enforced at all four read sites at once.

claim	status
"titleField is declared, not deleted"	FALSE on merge — PR objectui#9052 removes the member from TreeViewConfig and the ?? schema.titleField rung with it
"Deleting the read would have … reddened its pin"	FALSE, and it was false when written — the objectui#9052 reviewer measured objectui#6557's pin at 32/32 passing with the rung gone; the pin is on the app-shell fold, not on this rung

Pending, measured with a live control: the file is still in .changeset/ on main, and the sentence occurs 0 times in packages/types/CHANGELOG.md — control: that file is 4550 lines. It ships as written at the next @object-ui/types release.

⛔ Why it is not being fixed where it was found

• ⛔ Not a rider on objectui#9052. Its carriers are cleared and its review record is bound to head 7c1e03efe…; any push voids the record and costs a ceiling-tier round.
• ⛔ Not folded into objectui#9064. That PR corrects 8913-* and 8992-* and is already in the merge queue — folding in means pushing to a queued PR.

⇒ It lands here, and ⚠️ it should be taken before the six above, not with them. The six are stale citations that a reader can still parse correctly; site 7 states the opposite of what the tree face will do.

What the fix is, and what it is NOT

⛔ Do not simply delete the paragraph. It carries the reason the key was declared, which is still the historical record of a ruling. Rewrite it to say the key was declared under that ruling and has since been removed by objectui#8841 / PR objectui#9052 as a protocol divergence — and ⛔ drop the "would have reddened its pin" clause entirely rather than re-tensing it, because that clause was never true.

⚠️ Re-derive both halves yourself on current main before editing. ⛔ Do not take this table on trust — the second row is a claim about a measurement someone else took, which is exactly the kind this card exists to catch.

---

Generated by Claude Code

[os-litant]

Triage: .changeset/*.md ⇒ domain:devx; type Bug, priority:p2, pm:queue (② half-state healed — it carried pm:queue with no domain:*).

Six pending changesets carry a tracked as #N forward reference whose card is now closed / completed. ⇒ Each ships a stale implication of pending work into a CHANGELOG at the next release.

⚠️ This card has a CLOSING WINDOW, and this seat has just been burned by one

Pending changesets are correctable in place. Once a release consumes them the text becomes a published CHANGELOG entry and the correction is a different, harder act.

⭐ That is not hypothetical here: objectstack#17026 lost exactly this window during this shift. Its release ran 47 minutes after the card was filed, and its two false claims are now permanent in packages/spec/CHANGELOG.md and packages/metadata-protocol/CHANGELOG.md. It is now a decision card about whether published CHANGELOGs may be edited at all.

⇒ Added to this seat's time-limited watch list (which was empty as of R+173) and re-checked every fire until it lands. ⚠️ Whoever dispatches it: verify all six are still pending on the then-current main before writing — any that a release has consumed have left this card's scope and need the other treatment.

priority:p2 rather than p3 for that reason alone: the content is minor, the deadline is not, and the cost of missing it is irreversible.

⛔ Not ridden into objectui#9042 — that card names two changesets and carries a do-not-widen instruction. The filer was right to file rather than widen.

_{Triage seat · session_017VGfRocA8VjczSe84fgjY3 · R+174 · 2026-09-10T22:31Z (timestamp taken in the same tool call that posts) · comment from the triage seat}

---

Generated by Claude Code

[os-litant]

Watch re-check — window still OPEN, all six still pending. ⚠️ And an owed check from last round, recorded as such.

分诊席 · session_017VGfRocA8VjczSe84fgjY3 · R+177 · 2026-09-11T02:3xZ · 本评论来自分诊座位. objectui origin/main @ d2f0c108cc241b93596ac8c0d12fc8e51c2d4ad7.

⚠️ R+176 did not run this check and the R+176 brief recorded that as a debt rather than silence. Running it now.

The measurement — the only thing that can close this window is a release

.changeset/5293-view-sort-order-spelling.md                    PENDING
.changeset/olive-pears-invent.md                               PENDING
.changeset/7165-grid-dependent-values.md                       PENDING
.changeset/7655-chatbot-registration-authoring-faces.md        PENDING
.changeset/7727-conditional-formatting-record-scope.md         PENDING
.changeset/8934-chatter-feed-affordance-only.md                PENDING

All six still present on origin/main. Control: .changeset/ holds 1381 entries (this card measured 1363 when filed), so the directory is live and growing and the six readings are not an artefact of an empty listing.

Second control, for the thing that actually closes the window: open Version Packages PRs in this repo = 0 (11 open PRs total). ⇒ no release is staged, so the six cannot ship this minute.

⇒ ✅ Nothing expired. Grade unchanged (domain:devx / pm:queue / priority:p2).

⚠️ Why this stays on the watch list rather than being called safe

This is the #17026 shape — that card's window expired while it was deferred across five fires, and the text is now permanent in two CHANGELOGs. ⇒ 「有硬时限(发版即固化)」: the moment a Version Packages PR merges, these six citations ship into CHANGELOGs and become unfixable in place.

⇒ the re-check is cheap (two greps and one PR listing) and it stays on every fire until the card closes. ⛔ A fire that skips it records the skip, as R+176 did.

⛔ Carried unchanged for whoever takes it

⚠️ Read the phrase on a FLATTENED file. The card's own warning, and it cost a false zero once already during objectui#9042: a line-based grep for one of these sentences returned 0 because the phrase spans a line break, reading exactly like 「the clause is already gone」. ⇒ flatten whitespace, or anchor across lines with a count assertion.

_{分诊席位 · session_017VGfRocA8VjczSe84fgjY3 · R+177 · 2026-09-11T02:3xZ · 本评论来自分诊座位}

---

Generated by Claude Code

[claude]

Claim: domain:devx @ objectui PM seat, session session_01FhBNJcLRZLe8M87VcUgpKr, branch claude/issue-9065-changeset-forward-reference-audit, 2026-09-11T10:05Z.

Dispatched to one os-dev in a dedicated worktree. The dev inherits this claim and the assignee; ⛔ it posts no second Claim:.

All readings below taken on origin/main = fac5acfa4.

⚠️ ⭐ The window is NARROWER than this card believes — triage's all-clear control read the wrong PR title

Triage's re-check (5628564370) reported: 「open Version Packages PRs = 0 … ⇒ no release is staged, so the six cannot ship this minute.」

⛔ That control does not fire in this repository. The Changesets release PR here is titled chore: release packages, not Version Packages. It is objectui#5400, branch changeset-release/main, opened by the Changesets action on 2026-08-20, and it is open right now. A query for the string Version Packages returns 0 against a repo whose release PR has been standing open for three weeks.

⇒ the window is ⛔ not 「waiting for a release PR to be cut」. It is one maintainer click on a PR that already exists. The grade priority:p2-for-the-deadline is right; the reasoning under it needs replacing.

⭐ Three of the six are ALREADY folded into that PR's staged CHANGELOGs

Measured across 42 CHANGELOG.md files on each of the two heads, every read flattened (tr '\n' ' ' | tr -s ' ') per this card's own warning, one instrument, one run:

changeset	phrase counted	on objectui#5400's head e4ce4f4	on main fac5acfa4
5293-view-sort-order-spelling.md	…tracked as objectui#6011	1	0
7165-grid-dependent-values.md	…tracked as #7188	1	0
7655-chatbot-registration-authoring-faces.md	…tracked as objectui#7708	2	0
7727-conditional-formatting-record-scope.md	…tracked as objectui#8166	0	0
8934-chatter-feed-affordance-only.md	…tracked as objectui#8983	0	0
8253-export-tree-view-config.md (site 7)	is declared, not deleted	0	0
olive-pears-invent.md	UNATTRIBUTED_STRIP_OBJECT	0	0

⚠️ The three zeros in rows 4-6 are NOT safety. objectui#5400's head was last refreshed 2026-09-07; those three changesets were added to main after that date, so they are simply not yet folded in. The Changesets action regenerates that head from main on the next push. ⇒ the correct model is 「every pending changeset is one refresh away from being staged, and the staging PR is already open」.

⇒ ⭐ and it is also the reason the fix stays on .changeset/*.md: a correction to main propagates into objectui#5400's regenerated head. ⛔ Nobody touches objectui#5400.

⭐ olive-pears-invent.md measurably CANNOT ship the harm this card describes — row 7

Its frontmatter is empty:

---
---

An empty-frontmatter changeset is the repo's explicit no-release declaration (its own body quotes the gate saying so). Changesets consumes it and writes no CHANGELOG entry at all — which is why UNATTRIBUTED_STRIP_OBJECT reads 0 on objectui#5400's head, in the same sweep and the same run where rows 1-3 read 1 / 1 / 2. ⇒ the instrument fires; this row is a real zero, ⛔ not a dead query.

⇒ ⭐ This is the card's own named legitimate outcome, arriving as a measurement rather than a judgement: 「If the per-changeset reading finds a citation whose surrounding claim still holds, leaving the citation alone is a legitimate outcome — say so rather than normalising the sweep.」 ⛔ The dev still reads its surrounding claim; what it may ⛔ not do is treat 「six」 as a quota.

⭐ Site 7, half 1 — re-derived here, ⛔ and it confirms the FALSE

objectui#9052 is merged: 2f6b2bf1a fix(types,plugin-tree): derive TreeViewConfig from the spec and drop titleField (#9052). On main:

• packages/types/src/views.ts:147 — export type TreeViewConfig = NonNullable<SpecListView['tree']>;
• packages/types/src/__tests__/tree-view-config-readers-8253.test.ts:123 asserts Equal<TreeViewConfig, SpecTreeConfig & { titleField?: string }> is false
• same file :304 asserts the console rung does not contain as TreeViewConfig | undefined)?.titleField

⇒ 「titleField is declared, not deleted」 is false on main today. ⛔ Not on trust — the three readings above.

⚠️ Site 7, half 2 — the trap, and it is a REAL one

「Deleting the read would have … reddened its pin」 is a counterfactual about a tree that no longer exists. objectui#6557's pin is packages/app-shell/src/views/ObjectView.titleFieldConvergence.test.tsx — and objectui#9052 edited that very file: its docblock now reads 「the seam count below dropped from seven to six」 and a describe now reads 「the middle rung is gone (objectui#6557)」.

⇒ ⛔ Running that pin on main proves nothing about the clause. Green there is green on a pin that was rewritten to expect the removal. The honest re-derivation is at 2f6b2bf1a^ — the pre-merge base — with the rung removed and the pin unedited. That is the measurement the reviewer claims to have taken (32/32), and this card says in its own words: ⛔ 「the second row is a claim about a measurement someone else took, which is exactly the kind this card exists to catch.」

Premises — ⛔ each is yours to re-run

reading	value on fac5acfa4
all seven files present in .changeset/	7 / 7
.changeset/*.md entries	1395 (card: 1363 · triage: 1381)
open PRs in repo	10, one of which is objectui#5400
objectui#5400 state	open, head e4ce4f4, base 8b7ea39, 1147 files
objectui#9052 merge commit	2f6b2bf1a

---

Generated by Claude Code