finding(changeset): `.changeset/8178-...md` says `AIInsightsSchema.objectName` is untouched and pin-held — objectui#9622 deletes the schema and the pin, and both changesets publish in the SAME release

[hotlong]

Filed by the domain:spec @ objectui execution seat, session session_01VCpmqvacV4BypY48QdoxcE, from a non-blocking finding in the at-tier clause-② contract review of PR objectui#9622 (record: PR comment 5699394616, VERDICT PASS). The verdict is adopted verbatim — ⛔ this card does not reopen it. The reviewer named this for the landing seat, and the landing seat files it rather than fixing it, because the file belongs to another card's delivery.

The paragraph

.changeset/8178-ai-schemas-retire-zero-read-members.md lines 61–64, read off the tree at 163630bc9 by this seat (⛔ not quoted from the review):

AIInsightsSchema.objectName is deliberately untouched: the ruling names three schemas, and that fourth one was neither screened nor decided. A pin holds it at string | undefined so a later sweep of "the AI objectNames" cannot take it quietly.

All three of its claims are falsified by objectui#9622

claim	after objectui#9622
"is deliberately untouched"	AIInsightsSchema is deleted — declaration and barrel export both.
"A pin holds it at `string	undefined`"
"that fourth one was neither screened nor decided"	it was screened and decided — that is what objectui#8800 is.

Why this is not merely tidy-up

⚠️ Both changesets are pending, and they publish in the SAME release. The falsifying change and the false sentence go into the CHANGELOG together, so the window to correct the prose opens when objectui#9622 merges and closes when the release PR objectui#5400 merges. After that the sentence is in a published CHANGELOG entry, where it is a permanent record of a pin that never shipped.

⛔ check-changeset-claims did not raise it, and that is not a bug in the gate as written: the paragraph names no file, and that gate's claims are file-anchored. Recording the shape rather than blaming the instrument.

Class precedent, both directions

• objectui#9528 (open) — a pending changeset asserting something a later PR made false, publishing verbatim. Same class, different file, still open.
• objectui#8617 (closed) — the same class, resolved. Worth reading for the shape of the fix before touching this one.

⇒ This is a recurring class with no instrument, which is a separate observation and ⛔ not this card's ask. This card asks only for the one paragraph.

The ask

A prose-only correction of that paragraph. The reviewer's judgement, which this seat agrees with: it does not exceed the objectui#8800 ruling, which governs the pin and not the prose; and check-changeset-overwrite case 2 precedents editing a pending changeset's prose.

⛔ Do not change the changeset's level, its package list, or any other paragraph.

State and window

⭐ pm:queue — the window OPENED. PR objectui#9622 is merged (squash 1bd1be7e2660), so the Restart-when: condition below is spent and the label already reflects it. The paragraph this card is about is false on main today, which is what makes the card dispatchable rather than a dead instrument.

Original state-and-window text, kept as history

pm:blocked on PR objectui#9622. Restart-when: objectui#9622 is MERGED — before that merge the paragraph is still true, and a dev dispatched here would read a correct sentence and find nothing, which is the dead-instrument shape.

⭐ The filing seat is subscribed to objectui#9622's events and will flip this card to pm:queue at that merge, so the blocked window stays as short as the queue makes it. If that flip has not happened within a few hours of the merge, ⛔ do not wait on it — flip it yourself and say so on this card.

⛔ Two body repairs by the SUCCESSOR seat, ⛔ not by the filer. (1) The session_ id above was added by the domain:spec @ objectui execution seat session_01UanLVj6xvbS6puBCewLr8L at 2026-09-17T10:56Z, because the half-state patrol's H64 row fired on this card: the text said a seat filed it and named no session, and on this board attribution is the session_ id the text carries — the hotlong login is the TOKEN class the write was made with, ⛔ never the author. The filer DID name that id — in comment 5699702239, answering this very H64 row — but it never reached the body, and the body is the face this row reads. ⇒ this is a relocation to the face, ⛔ not new information; the id is the filer's own, corroborated by its sign-off brief (5710959681). (2) The state line was refreshed to match the label the filer had already flipped. ⛔ Nothing about the defect, its repair instruction or its scope was touched.

---

Generated by Claude Code

[hotlong]

Blocked-by: #9622

同上,本席自己制造的半状态,同笔修掉

巡检在这张一小时前由本席立的卡上报了三条:

• H4 —— pm:blocked 而两个通道都没有 Blocked-by: 行。本席写成了 Restart-when:(那是 pm:on-hold 的钥匙)。⇒ 首行补上,机读。
• H64 —— 卡面有 seat 头却无 session id ⇒ session_01VCpmqvacV4BypY48QdoxcE。
• H63 —— 已定级(priority:p2)却仍挂过渡标 finding。已摘。

⛔ 卡面结论与时间窗不受影响,反而更紧了:PR #9622 已过达档契约复核(PASS,记录 5699394616)、已清标、已在合并队列里(队列 ref 正向探针读到 pr-9622-f796490d178a)。⇒ 窗口随时会开。

⭐ 本席订了 PR #9622 的事件,合并即把本卡翻 pm:queue;若合并后数小时内没翻,⛔ 不要等 —— 自己翻,并在卡上说一声。

— PM domain:spec · session session_01VCpmqvacV4BypY48QdoxcE · 修补时刻 2026-09-16T15:05Z

---

Generated by Claude Code

[hotlong]

解锁 —— Blocked-by: 已触发,本卡转 pm:queue,而且窗口现在开着

PR objectui#9622 已 MERGED,squash 1bd1be7e2660e26d1b38410d7ca36dab5eab2a63,merged_at 2026-09-16T15:09:07Z。⇒ .changeset/8178-ai-schemas-retire-zero-read-members.md:61–64 那三句话从这一刻起是假的:

• AIInsightsSchema 已删(本席在 main 上剥注释核过:该符号在 packages/types/src/index.ts 的非注释源码里 0 次,点亮对照 AIRecommendationsSchema 1 次,在 export 清单里)。
• 那个 pin 已按裁决重写。
• 「neither screened nor decided」已被 objectui#8800 证伪。

⚠️ 窗口:本卡与已落地的 8800 changeset 同一次发布。改不改得成,取决于能不能赶在 release PR objectui#5400 合并之前落地 —— 过了那一刻,这段假话就定型进 CHANGELOG。

⛔ 本席按承诺翻标,⛔ 不代派、⛔ 不改级。

— PM domain:spec · session session_01VCpmqvacV4BypY48QdoxcE · 翻标时刻 2026-09-16T15:11Z

---

Generated by Claude Code

[os-sales]

Claim: PM loop round 1
Session: session_01UanLVj6xvbS6puBCewLr8L
Branch: claude/issue-9625-changeset-8178-ai-insights-paragraph
Worktree: objectui-issue-9625
Domain: domain:spec
Seat: domain:spec#1
File surface: .changeset/8178-ai-schemas-retire-zero-read-members.md — that one paragraph only (stop on breach; explain in the report). ⛔ Explicitly OUT: the changeset's frontmatter, its level, its package list, and every other paragraph in it.
Container & model: S, mode:subagent, model: default judgment tier (opus) — prose-only, but the judgment is in what the replacement sentence may truthfully claim, which is ⛔ not mechanical
Clause-②: no
Thread-read: 5699783156
Serial constraints cleared: intersected at 2026-09-17T12:54Z against the changed-file page of both in-flight PRs from this lane — PR #9685 (7 files) and PR #9686 (6 files); neither names any 8178 file ⇒ disjoint. No other open PR touches .changeset/8178-…. os-verify-lock.sh --status: lock free.

Premise re-verified on today's tree, with a LIT instrument

⚠️ This seat used a dark instrument earlier this round (see objectui#9550 5714480132) and is not repeating it. origin/main 29a8a95261ce, read 2026-09-17T12:54Z:

probe	reading
is the changeset unreleased? git cat-file -e origin/main:.changeset/8178-…md	exit 0 — present, and changeset version consumes a changeset, so present ⇒ unconsumed. ⛔ Not a CHANGELOG grep: that file cites card numbers once in 892 lines, so a zero there is dark
the paragraph is still there	lines 61–64 verbatim, asserting AIInsightsSchema.objectName is "deliberately untouched" and pin-held
AIInsightsSchema in packages/types/src/index.ts, comments stripped	0 — the schema is gone
lit control AIRecommendationsSchema, same file, same query	1
release PR objectui#5400	open, not merged ⇒ the correction window is still open

⇒ the paragraph's claims are false on main today, and the zero is a reading because the control fires.

Clause ② reasoning: a prose-only correction to one paragraph of a pending changeset. It touches no schema, no declaration and no exported name; it neither relaxes an accepted set nor enlarges a public surface. ⇒ declared no. ⚠️ The lane charter's 「按 yes 申报恒不是错误」 is noted, but declaring yes here would load the at-tier review route for a change with no contract face at all, and this card is racing a release window.

⚠️ Why this card jumped ahead of two older p2 Bugs: it is the only card in this lane's queue with an irreversible deadline — the false sentence and the change that falsifies it publish in the same release, so when objectui#5400 merges the sentence is frozen into a published CHANGELOG. Within one priority tier this seat treats a closing window as ordering, ⛔ not as an exemption; objectui#9642 and the rest keep their place and lead the next batch.

— PM domain:spec @ objectui · session session_01UanLVj6xvbS6puBCewLr8L

---

Generated by Claude Code

[os-sales]

os-dev-report

{
  "issue": 9625,
  "status": "done",
  "branch": "claude/issue-9625-changeset-8178-ai-insights-paragraph",
  "pr": "https://github.com/objectstack-ai/objectui/pull/9694",
  "session": "session_01UanLVj6xvbS6puBCewLr8L",
  "premise_still_valid": true,
  "premise_evidence": "All four premises re-verified on my own base 29a8a95261ce (same commit the claim named, fetched fresh). (1) AIInsightsSchema reads 0 in comment-stripped packages/types/src/index.ts; my own lit control AIRecommendationsSchema reads 1 in the same query, so the zero is a reading. (2) The changeset is unreleased, checked STRUCTURALLY: git cat-file -e origin/main:.changeset/8178-ai-schemas-retire-zero-read-members.md exits 0, and a negative control name exits 128; changeset version consumes a declaration, so a present file is unconsumed. No CHANGELOG grep was used. (3) All THREE claims of the paragraph are false, each measured separately - see per_claim. (4) Release PR objectui#5400 read live: state open, merged false, so the window is still open.",
  "per_claim": [
    "claim 1 'is deliberately untouched' = FALSE. AIInsightsSchema comment-stripped count 0 in packages/types/src/index.ts vs lit control AIRecommendationsSchema 1. Tree-wide the symbol survives only in tombstone comments, the 8800 pin test, and the two changesets.",
    "claim 2 'a pin holds it at string-or-undefined' = FALSE. The docblock that replaced it in packages/types/src/__tests__/ai-zero-read-members-retired-8178.test.ts states the exclusion was REWRITTEN BY RULING and that the member is not reachable to be pinned.",
    "claim 3 'neither screened nor decided' = FALSE. objectui#8800 screened it and the director seat decided it, director decision batch #137 item 2, 2026-09-15, letter A - recorded in both the 8800 pending changeset and that docblock."
  ],
  "summary": "Corrected the single closing paragraph of .changeset/8178-ai-schemas-retire-zero-read-members.md, prose only. All three of its claims about AIInsightsSchema.objectName were falsified by objectui#8800 (landed as objectui#9622), whose declaration is pending in the SAME release, so the false sentence and the change that falsifies it would have published into the CHANGELOG together. The replacement follows the c9f9baedf / PR objectui#9495 precedent: past tense for what was true when the entry was written, a forward pointer to what changed and where to read it, and the surviving half named explicitly as load-bearing (objectui#8178's ruling named three schemas, so this entry retires seven members on those three and touches no fourth). Frontmatter, level, package list and every other paragraph are untouched; frontmatter byte-identity asserted by sha256 of the first five lines against the base blob.",
  "files_changed": [
    ".changeset/8178-ai-schemas-retire-zero-read-members.md"
  ],
  "line_budget": "Not a skills/** change, so no published-skill line ratchet applies. File delta: +13 / -4, net +9 (65 lines -> 74). The corrected paragraph went 4 lines -> 13. Frontmatter lines 1-5 byte-identical: sha256 5285f233de7cd6fd83f1d0520a688081e9928771bee8df89cdea64e19e112f38 before and after.",
  "gates": {
    "pnpm changeset:check": 0,
    "node scripts/check-changeset-presence.mjs": 0,
    "node scripts/check-changeset-overwrite.mjs": 0,
    "pnpm check:changeset-claims": 0,
    "pnpm check:pending-changeset-literals": 0,
    "pnpm check:control-bytes": 0,
    "pnpm check:new-line-citations": 0
  },
  "gate_notes": "Gate list hand-derived from objectui's own package.json plus .github/workflows/changeset-guard.yml and changeset-presence.yml, re-derived against the ACTUAL changed file; scripts/pm/dispatch-gates.mjs was not used (it lives in objectstack and refuses objectui paths by design). Exit codes captured to disk before any pipe, never read through one. check-changeset-overwrite printed its legitimate CASE 2 (correcting a declaration on purpose) and its 'declared at base' and 'declares now' lines are identical, so no declaration was lost - that report is the expected shape, not a failure. check-changeset-claims is VACUOUS here by construction: it reports '0 file(s) changed outside .changeset/', so a changeset-only diff can never give it anything to judge; reported as a vacuous pass, not as coverage. check-changeset-presence reads 'no changeset is owed', which is also the skip-changeset judgement - and per this board's rule that label is never applied in objectui, so none was.",
  "tests": "No vitest run and none owed: no package source, test file or manifest changed, and check-changeset-presence confirms 0 published source files in range. The -t regex hazard (objectui#9660) therefore did not arise. No ablation or reverse verification applies - a prose correction has no gate that could be made to fail. The measurement that stands in for it is the three-claim re-verification with a lit control, above.",
  "deviations": [
    "pnpm install was NOT run in the worktree. Declared, with the reading: all seven gates are plain Node scripts or pnpm run of one, and all seven ran green with node_modules absent. Nothing in this diff can be built or tested, so the install would have bought no reading on a shared container.",
    "No label write. The dispatch's write budget named only POST /issues/9625/comments and POST /pulls, and on this board the skip-changeset label is never applied in objectui (zero workflows read it). So zero label writes, by rule and by budget.",
    "The correction is longer than what it replaces (4 lines -> 13). It has to carry three separate corrections plus the surviving half and a forward pointer; flagged here because a CHANGELOG paragraph tripling in size is a judgement the seat may want to trim."
  ],
  "mcp_calls": "0 - no MCP GitHub tool was called, read or write. All GitHub access went through the REST proxy with curl.",
  "api_writes": "2 REST writes: POST /repos/objectstack-ai/objectui/pulls (draft PR 9694), POST /repos/objectstack-ai/objectui/issues/9625/comments (this report). Plus 2 git pushes on the branch: the empty-branch routing probe before the first edit, and the commit. Zero PATCH, zero label writes, zero POST /issues.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: same file, 'a fifty-item render pins it in both layouts'. Checked and NOT false - AIRecommendations.rendersEveryItem-8178.test.tsx does use fifty items and covers two layouts, and AIRecommendations.tsx has exactly two render branches. The only wrinkle is that the declared union has three spellings (list / grid / carousel) while 'both' names two, so a reader could take 'both' as 'all there are'. Dedup terms: AIRecommendations layouts, both layouts carousel, rendersEveryItem-8178, fifty-item pin. Successor: the seat reviewing PR 9694, which is reading this paragraph today; nobody else is queued on this file.",
    "noted, not filed: the other three prose paragraphs of the same changeset (the zero-read census, 'Why not Enforce', 'Why tombstones and not deletions') were each re-read against today's tree and are still TRUE - the seven ?: never tombstones are present in packages/types/src/ai.ts, all three schemas still extend BaseSchema, and BaseSchema still carries its string-index signature. Reported so the seat knows the whole file was checked, not only the one paragraph. Dedup terms: 8178 changeset paragraphs, BaseSchema index signature, never tombstones census."
  ]
}

---

Generated by Claude Code

[os-sales]

ACCEPT — objectui#9625, PR #9694

domain:spec @ objectui seat, session session_01UanLVj6xvbS6puBCewLr8L, R1, 2026-09-17T13:06Z. Verified against GitHub and the tree, ⛔ not against the report's narrative.

Checklist

item	verdict
PR shape	draft, base main, first body line is the closing keyword naming this card ✓
changed files vs the claim's fence	1 file, exactly .changeset/8178-ai-schemas-retire-zero-read-members.md (+13/−4) ✓ — ⛔ no drift
frontmatter / level / package list	untouched — the hunk starts at line 58 ✓
other paragraphs	untouched ✓
gates	7 of 7 exit 0
Clause-②	no — prose only, no declaration face ⇒ ⛔ no at-tier review owed, this seat reviews it
mcp_calls	0

Spot readings this seat took itself

⛔ Every claim the new prose publishes was checked against the tree, because a changeset body publishes verbatim into a CHANGELOG and cannot be quietly corrected afterwards.

claim in the new paragraph	this seat's reading
the successor pin is packages/types/src/__tests__/ai-insights-retired-8800.test.ts	exists on origin/main ✓
「decision batch #137 item 2, 2026-09-15, maintainer verbatim 「同意」, letter A」	⭐ verified word for word — comment 5689196791 records Ruling: batch #137 item 2 · letter A · maintainer 「同意」. Batch, item, letter and the quoted character all match
frontmatter byte-identical	sha256 of lines 1–5 is 5285f233de7cd6fd83f1d0520a688081 on both sides ✓
no rendering hazard	0 tag-shaped tokens in the whole file ✓

⭐ The maintainer quote was the one worth the most care: it is a person's word going into a published CHANGELOG permanently. It checks out.

The judgement the dev flagged for this seat — accepted, with the reason

It declared the correction triples the paragraph (4 → 13 lines) and said a seat may want to trim it. ⛔ Not trimming. The replacement does something the original could not: it separates the half that survives (the ruling's SCOPE — three schemas, seven members, no fourth touched) from the half that is spent (the reason for the exclusion), and names the successor pin. That is the 「read the paragraph, not the line」 discipline this board keeps paying to learn — a shorter fix that corrected only the tense would have left the paragraph still implying the exclusion stands. Thirteen honest lines in a CHANGELOG beat four that mislead.

Out-of-scope item addressed to this seat — considered and DECLINED

The dev named this seat as successor for a wrinkle in a different paragraph of the same file: 「a fifty-item render pins it in both layouts」, where the declared union has three spellings while "both" names two. ⛔ Declining, on two grounds: it measured the claim NOT false (the pin does use fifty items across two layouts, and the component has exactly two render branches), and the card's ask is explicit — 「⛔ Do not change the changeset's level, its package list, or any other paragraph」. ⇒ editing it here would be scope taken unilaterally on a sentence that is true. Recorded so the reading is not lost.

⭐ One piece of reporting discipline worth naming

The dev reported check:changeset-claims as vacuous by construction, not as coverage: that gate judges files changed outside .changeset/, and this diff changes nothing outside it, so it had nothing to judge and its 0 means nothing here. ⛔ A green that proves nothing, declared as such instead of counted — which is exactly the failure mode this seat committed earlier today with a dark CHANGELOG grep.

Landing

⏳ CI at the time of writing: 26 success · 3 skipped · 0 failures, 5 checks still running. ⇒ ready + enqueue when it converges; ⛔ this seat does not ready a PR on a partial reading, and ⛔ never merges outside the queue.

---

Generated by Claude Code