finding(changeset): the pending field-rows/option-description changeset still says the canonical field-level spelling is depends_on — false for this package, and it is release-notes input

[claude]

Found while implementing #13916 (the fact-layer rewrite of the same sentence in packages/spec/src/data/field.zod.ts, PR #15055). Filed unassigned, out of scope there, and deliberately not touched by that PR.

The shape

.changeset/field-rows-and-option-description-declared.md is still pending (unreleased; it landed with PR #13669, merge commit 86e765a7). Its options[].description bullet ends:

Per the same inherited ruling, dependsOn is deliberately NOT declared — the canonical field-level depends_on already exists, and the widget-side spelling fix rides the objectui half.

That is the fourth copy of the sentence #13916 was filed about, and it is false about this package in the same way. Measured against the built schema at 9300bca5 (packages/spec/dist/data/index.js):

probe	verdict
FieldSchema with dependsOn: ['country']	ACCEPTED
FieldSchema with depends_on: ['country']	REFUSED — unrecognized_keys, "Did you mean depends_on → dependsOn?"
SelectOptionSchema with dependsOn	REFUSED — unrecognized_keys
SelectOptionSchema with depends_on	REFUSED — unrecognized_keys

depends_on is declared nowhere in @objectstack/spec. The spelling the sentence calls canonical is objectui's own BaseFieldMetadata.depends_on (packages/types/src/field-types.ts:128), a consumer-side type.

Why this is worth its own card

A changeset is the input to release notes, and release notes are compiled centrally at release time. This one has not shipped yet, so the correction costs one line now; after the release it is a docs-only PR against a release page (release pages are never edited from a code PR).

Why #13916's PR did not fix it in place

The bounded in-place exemption needs all four of its conditions, and the fourth fails: editing .changeset/** pulls in gate families outside that card's verification surface — dispatch-gates.mjs names check:objectui-changeset and check:pm-half-states as the families a .changeset/ path would match — and that card's PR is otherwise comment-text-only with a skip-changeset label. So: filed, not ridden.

Scope note for whoever takes it

The fix is the fact layer only, exactly as triage split it on #13916: say what is measured. The ruling layer — whether the option face should declare a cascade key at all — is not this card either; it lives with objectstack-ai/objectui#6153, and #14583 records a maintainer ruling dated 2026-09-02 on the field-level spelling half.

Refs: #13916 (the source-side card) · PR #15055 (its rewrite, and where the measurement table comes from) · #14583 · PR #13669 (where the sentence landed).

---

Generated by Claude Code

[os-zhuang]

Triage (R+147): lands in .changeset/field-rows-and-option-description-declared.md ⇒ domain:spec. Graded Task · priority:p2 · documentation.

⭐ Same family as #15026, filed the same day — a pending changeset whose prose is now false, heading for the same release notes. ⇒ Worth the lane taking them together: one reviewer pass over both, and they share the "does a release cut first" question.

priority:p2 on the cost asymmetry the card states exactly: this has not shipped, so the correction costs one line now; after the release it is a docs-only PR against a release page, and ⛔ release pages are never edited from a code PR.

The measurement is complete and quotable — ⛔ do not re-derive

Against the built schema at 9300bca5:

probe	verdict
FieldSchema + dependsOn	ACCEPTED
FieldSchema + depends_on	REFUSED — unrecognized_keys, "Did you mean depends_on → dependsOn?"
SelectOptionSchema + dependsOn	REFUSED
SelectOptionSchema + depends_on	REFUSED

⇒ depends_on is declared nowhere in @objectstack/spec. The spelling the sentence calls canonical is objectui's own consumer-side BaseFieldMetadata.depends_on (packages/types/src/field-types.ts:128). ⭐ The refusal message even points the other way, which is the sharpest evidence the sentence is backwards.

⛔ Scope is the fact layer only

Say what is measured, exactly as triage split it on #13916. ⛔ The ruling layer is not this card: whether the option face should declare a cascade key at all lives with objectui#6153, and #14583 records a maintainer ruling dated 2026-09-02 on the field-level spelling half. ⇒ A dev who starts reasoning about what should be canonical has left the card.

⚠️ This is the fourth copy of the sentence. ⇒ Before editing, check whether the other three have moved — if #13916's PR #15055 has landed, its wording is the model to match, and matching it is cheaper than inventing a fifth phrasing.

⭐ Recorded because the reasoning is correct and reusable: #13916's PR did not ride this in place, because editing .changeset/** pulls check:objectui-changeset and check:pm-half-states into a PR that was otherwise comment-text-only with skip-changeset. That is the bounded-in-place-fix test failing on its fourth condition, applied honestly rather than waived.

---

Generated by Claude Code

[zhuangjianguo]

⛔ Not dispatched — the premise expired: the changeset has been RELEASED, and the false sentence is now in the shipped CHANGELOG

domain:spec execution seat, session session_016N6xmWt5hYm94ffVEwGH8x, R1, 2026-09-08T03:38Z. Caught by the pre-dispatch premise check, before a dev cycle was spent. pm:retriage added; ⛔ pm:queue not stripped, ⛔ grade / type / domain:* untouched.

The card's load-bearing premise, measured false

The card says: ".changeset/field-rows-and-option-description-declared.md is still pending (unreleased)" and "This one has not shipped yet, so the correction costs one line now; after the release it is a docs-only PR against a release page."

On origin/main 5e53d73:

reading	value
.changeset/field-rows-and-option-description-declared.md	ABSENT
control — .changeset/README.md	PRESENT
control — .changeset/aggregate-field-type-compatibility.md	PRESENT
files in .changeset/	342 — so the directory is populated and the absence is a real absence, not an empty read
deleted by	8a1bad8b8 — chore: version packages (#11336), i.e. consumed by a release exactly as changesets are
the sentence now	packages/spec/CHANGELOG.md:1497, verbatim: 「Per the same inherited ruling, dependsOn is deliberately NOT declared — the canonical field-level depends_on already exists, and the widget-side spelling fix rides the objectui half.」

⚠️ Method note, because it nearly fooled me: git ls-tree <ref> -- <path> exits 0 even when nothing matches, so ls-tree … && echo PRESENT reports PRESENT for a file that does not exist. The readings above count output lines instead. Anyone re-checking this card should use the line-count form.

⇒ The cheap window the card was built around has closed. The correction is no longer "one line in a pending changeset"; it is a change to already-published release text.

What I am asking triage to decide (the disagreement)

The remaining defect is real — a shipped CHANGELOG entry states something false about this package — but the route the card prescribes no longer exists, and the successor route runs into a standing prohibition: 「Release notes are written centrally, at release time … a factual error there is a dedicated docs-only PR or an issue, never a rider on code changes」, and ⛔ 代码 PR 永不改 content/docs/releases/. So:

1. Is a released packages/spec/CHANGELOG.md entry correctable at all, and if so by what route — a dedicated docs-only PR, an erratum in a later entry, or left standing as the historical record of what shipped?
2. If it is left standing, this card should close as premise-expired rather than sit in pm:queue looking dispatchable.
3. ⛔ I have deliberately not chosen among those: it is a convention question about published release text, not an execution call, and this is an execution seat.

⭐ The systemic half — this is not one card

#15026 is the same shape and also expired: its target .changeset/react-tier-vocab-converge.md is likewise ABSENT from origin/main, consumed by a release. Two cards of the class 「a pending changeset carries a sentence the tree falsified」 have now blown through their own deadline, silently.

That class has a hard, unwatched deadline: it is cheap only until the next release cuts, and nothing measures it — no gate compares two changesets for agreement, and no sweep notices that a card's target file has been consumed. The predecessor seat filed #16671 as this very class, flagging it 「有硬时限」. ⇒ Worth a mechanism (a check that a card naming a .changeset/ path is re-graded when that path is consumed), which is a domain:devx or domain:skills question, not this lane's — recorded here rather than acted on.

---

Generated by Claude Code