finding(ci): FOUR places in the tree tell authors the four shard jobs are NOT required checks — all were true when written and are FALSE today, and acting on any one blocks every pull request in the repository

[os-try-charles]

⚠️ STAND-IN TRIAGE, declared. Graded by the domain:devx @ objectui execution seat (session_013VGeMu3p6qEFWR6K6GGLaW, R58) under 〈多仓协调〉规则 4's stand-in clause; triage seat objectstack#6015 read 🔴 空缺 at 2026-09-14T12:00Z (updated_at 05:50:57Z). ⛔ I stop the moment it has a holder.

Surfaced by the objectui#9499 dev as to file, with its dedup words and its probe. Adopted and filed here.

The two sentences

file	says
scripts/dependabot-merge-gate.mjs (header)	the required set is "demonstrably not carrying the shards today"
content/docs/guide/ci-cd-pipeline.md	it is "a repository-settings surface nothing here can read"

Both are false today — one command, with a firing control

GET /repos/objectstack-ai/objectui/rules/branches/main returns a required_status_checks rule containing exactly:

Lint · Type Check · Build & E2E ·
Test (shard 1/4) · Test (shard 2/4) · Test (shard 3/4) · Test (shard 4/4) ·
Build Docs · Changeset Declaration
strict_required_status_checks_policy: true

⇒ the four shard contexts are required, and the surface is readable — by an ordinary token, through a documented endpoint.

CONTROL, same command, same token, one argument changed: GET /rules/branches/zzz-no-such-branch-9499 returns 200 []. ⇒ the reader distinguishes "a branch with rules" from "a branch with none", so the non-empty answer is a reading and ⛔ not a default. Corroborated by a second endpoint, GET /rulesets/11776024 (enforcement: active, source_type: Repository).

⭐ Both were verified twice: by the objectui#9499 dev, and independently by this seat before filing.

⛔ Why this is dangerous rather than untidy

An author who trusts either sentence renames or re-shards the test jobs — exactly what objectui#9499's repair requires — and then:

• every pull request in the repository blocks forever on four contexts that can never be produced again;
• every merge-group build burns the merge queue's check_response_timeout_minutes: 60 and fails.

⛔ There is no ordering in which nothing blocks: the old and new context names can never both be satisfiable. And only the maintainer can edit the ruleset — this seat's token reads it but GET branches/main/protection answers 403 Resource not accessible by integration ⇒ the person who breaks it cannot unbreak it themselves.

⭐ Both sentences were TRUE when written — which is the actual defect class

The evidence behind them was objectui#4959: a merge landed while all four shards were in_progress, which is only possible if the shards were not required. ⇒ ⛔ nobody wrote a falsehood.

The ruleset's updated_at is 2026-08-24 — after that reading. The world changed and the prose did not.

⇒ this is AGENTS.md #9's exact shape: a fact derived once, written down, and never re-derived. ⭐ And it is the second instance this seat has filed today from the same class (objectui#9493 is the other) — which suggests the class, not the instances, is what deserves attention.

Suggested acceptance

1. Both sentences either corrected or replaced by a pointer to the endpoint — ⭐ prefer the pointer. Restating an endpoint's answer in prose is what created this card; a sentence that says "read GET /rules/branches/main" cannot go stale.
2. ⚠️ If a figure or a list is quoted anyway, it carries when and how it was derived, so the next reader can check it instead of trusting it.
3. ⛔ Do not "fix" this by changing the ruleset to match the prose. The ruleset is correct; the prose is wrong.
4. Every zero carries a control proven able to return non-zero in the same command; name the population in words.

⚠️ Serial note: content/docs/guide/ci-cd-pipeline.md is pinned by scripts/__tests__/ci-cd-pipeline-doc.test.ts, and scripts/dependabot-merge-gate.mjs by scripts/__tests__/dependabot-merge-gate.test.ts. ⛔ Both will need to move with the prose. ⚠️ Whoever implements objectui#9499's fork will be editing these same two files — coordinate, or land this one first.

Blocks nothing, but is a precondition for safely executing objectui#9499.

---

⛔ CORRECTION — 2026-09-14T16:4xZ, by the seat that filed this card

⚠️ "Two places" is wrong. Measured on origin/main at 3adc50eec8: FOUR independent carriers, plus three quotations of them. I found this while verifying the file surface before dispatching the card, ⛔ not after shipping a fix — but the undercount was published and is corrected here rather than silently amended.

⭐ This is the card's own defect class, committed by the card. A count derived once, written down as authoritative, never re-derived. Recording it as such.

The population, re-derived, named in words

Every occurrence of either phrase in the tree, git grep -n -i over origin/main:

#	file:line	phrase	kind
1	scripts/dependabot-merge-gate.mjs:57	not carrying the shards	independent claim
2	content/docs/guide/ci-cd-pipeline.md:138	nothing here can read the current state of it either	independent claim
3	content/docs/guide/ci-cd-pipeline.md:2789	repository-settings surface nothing here can read	independent claim
4	.github/workflows/dependabot-auto-merge.yml:47	repository-SETTINGS surface nothing here can read or change	independent claim ⭐ not in the original body
5	content/docs/guide/ci-cd-pipeline.md:2655	quotes #2	quotation
6–7	scripts/check-required-check-set.mjs:37,41	quote #2 and #4	quotation, inside an inventory docblock

CONTROL, same command, one argument changed: git grep -n -i "zzz-no-such-phrase-9502" origin/main returns 0 files. ⇒ the six non-zero readings are readings and ⛔ not a grep that matches everything.

⚠️ "carrying the shards" is 1 occurrence in 1 file — that half of the original body was right. The undercount is entirely on the "nothing here can read" half, which I reported as one file and is four.

⭐ The bigger correction: the repo ALREADY reads this surface

scripts/check-required-check-set.mjs exists on main and its own docblock opens:

the merge queue's required-check set is off-repo configuration, and this is the only thing in the tree that looks at it.

It takes a live read of ruleset 11776024, exits 2 when the reading could not be taken (⛔ never a pass) and 3 on breach, is wired as pnpm check:required-check-set (package.json:90), has a standing caller (.github/workflows/required-check-set-patrol.yml) and an ablation-driven test.

⇒ "nothing here can read the current state of it" is false twice over: the endpoint is readable, and a gate in this repository already reads it on a schedule.

⚠️ And the author of that gate knew. check-required-check-set.mjs:34–44 inventories the three carriers of the claim by name and closes:

The WRITE half of all three is still true and this file does not touch it.

⇒ ⛔ this is not an oversight to sweep up. A prior author drew a deliberate read/write line and left the read-half prose standing. Any repair must either honour that line or argue against it — ⛔ not delete the sentences as though nobody had considered them.

⛔ GOVERNED FORK — AGENTS.md:338, ⛔ NOT in this card's scope

check-required-check-set.mjs:42 names a third carrier the original body missed entirely:

origin/main:AGENTS.md:338
- ruleset 的具体配置(谁可绕过、required checks 清单)本文不写 —— 从仓内读不到,别照抄任何推断。

AGENTS.md is governed surface. ⇒ ⛔ it is out of scope for this card and for any dev dispatched on it, and it must not be edited in the same pull request.

⚠️ It is also not obviously false in the same way. Its parenthetical covers 谁可绕过 (bypass actors) alongside the required-checks list, and GET /rules/branches/main — the endpoint this card's reading used — does not carry bypass actors. So the AGENTS.md sentence may be partly true on a leg the other three do not have. ⛔ Ruling that is a separate, governed card, and this seat is ⛔ not ruling it here.

⇒ the honest state after this card lands is three carriers corrected and one governed carrier left standing, deliberately and in writing.

---

Generated by Claude Code

[os-try-charles]

⭐ A second card in this lane has been reasoning on the false premise — and it has cost 20 days of a hold

Found while running this lane's unlock scan, ⛔ not searched for.

objectui#6319 (pm:on-hold since 2026-08-25) is held on the reasoning that its only remedy is "a repository setting, ⛔ not code" — objectui#6245's option 1, "require branches to be up to date before merging" — and that nobody had applied or could apply it.

That setting reads strict_required_status_checks_policy: true today, in the same GET /rules/branches/main response this card is about.

⇒ ⭐ this card's two false sentences are not just a documentation defect — they are a premise other cards have been reasoning from. objectui#9499 hit the same surface from the opposite direction (it needs that ruleset edited and cannot do it); objectui#6319 hit it by concluding the surface was out of reach and going to sleep for three weeks.

⚠️ ⛔ I did NOT unlock objectui#6319 — whether the flag actually closes its residual window is unmeasured (the merge-queue interaction, and which field the 2026-08-24 updated_at refers to, are both unknown). The reading is recorded there as a question, ⛔ not as a resolution.

⇒ this strengthens the acceptance already on this card: repair 1 says prefer the pointer over restating the endpoint's answer. ⭐ A sentence saying "read GET /rules/branches/main" would have let objectui#6319's filing seat check in one command instead of concluding the surface was unreadable.

⚠️ Carrier added: whoever takes objectui#6319, and whoever takes objectui#9499 — both need this card's repair to reason correctly about that surface.

---

Generated by Claude Code

[claude]

⚠️ STAND-IN TRIAGE, declared. Graded by the domain:devx @ objectui execution seat (session_013VGeMu3p6qEFWR6K6GGLaW, R59) under 〈多仓协调〉rule 4's stand-in clause. Triage seat objectstack#6015 read 🔴 空缺 again at 2026-09-14T16:32Z (updated_at still 05:50:57Z, no later round-open marker). ⛔ I stop the moment it has a holder. ⛔ Graded, not claimed, not dispatched — the triage seat's scope is 扫/分类/打标签/拆跨域/查重,⛔ 永不认领派发.

Ruling — pm:queue (priority:p2 retained)

Authority, quoted from the dispatch protocol's classification table:

pm:queue 也收恢复不变量的 finding、test-only pin,与实现未被裁错的说明书脱节(修文档)。

That is this card exactly. The ruleset is correct and was not mis-ruled; two sentences describing it are out of date. The repair direction is therefore fixing prose, and the card already carries a named landing point for each sentence, a suggested acceptance, and a reading with a firing control (GET /rules/branches/zzz-no-such-branch-9499 → 200 [], so the non-empty answer is a reading and not a default). There is nothing here to ask a maintainer.

⚠️ Two things whoever takes it must carry

1. The two pinning tests move with the prose — scripts/__tests__/ci-cd-pipeline-doc.test.ts and scripts/__tests__/dependabot-merge-gate.test.ts. Both are named in the card body; neither is optional.
2. ⛔ Do not change the ruleset. The card says it and it bears repeating: the ruleset is right, the prose is wrong. This seat's token can read GET /rules/branches/main but GET /branches/main/protection answers 403 Resource not accessible by integration — the ruleset is a maintainer-only surface and is out of scope for any dev on this card.

Serial surface — measured, clean

GET /pulls/{n}/files over all 8 open PRs in this repo: none touches scripts/dependabot-merge-gate.mjs, content/docs/guide/ci-cd-pipeline.md, or either pinning test. Control on the same command: the identical filter returned a non-empty collision set for #9488 and #8941 (both on scripts/check-eager-closure-budget.mjs) ⇒ the filter discriminates, so this zero is a reading and not a dead scanner.

⭐ Why this is worth taking before its priority suggests

This card is a precondition for safely executing objectui#9499 — a p1 currently sitting in the maintainer decision box. #9499's repair renames or re-shards the four Test (shard N/4) jobs, and both of these sentences tell an author those contexts are not required. An author who trusts either one blocks every pull request in the repository on four contexts that can never be produced again. ⛔ I have not raised the priority — inventing a level is not this seat's call — but the ordering is stated here so it is not lost.

---

Generated by Claude Code

[claude]

Dispatch order — R59

Claimed by the domain:devx @ objectui execution seat (#5748), PM session session_013VGeMu3p6qEFWR6K6GGLaW. Assignee baozhoutao is this seat's shared identity — the claim is this comment plus the session ID.

Base: origin/main @ 3adc50eec8.

Branch: claude/issue-9502-required-set-readable-pointers

⚠️ Read the CORRECTION block at the bottom of the card body first. The original body said "two places". It is four, I published that undercount, and I corrected it before writing this order. The corrected population is what you are executing.

The surface — verified by me on origin/main @ 3adc50eec8, ⛔ not inferred

Independent claims to repair — 3 files, 3 lines:

file:line	the sentence
scripts/dependabot-merge-gate.mjs:57	"demonstrably not carrying the shards today"
content/docs/guide/ci-cd-pipeline.md:138	"nothing here can read the current state of it either."
content/docs/guide/ci-cd-pipeline.md:2789	"repository-settings surface nothing here can read"
.github/workflows/dependabot-auto-merge.yml:47	"repository-SETTINGS surface nothing here can read or change"

(That is 4 lines in 4 files — ci-cd-pipeline.md carries two of them.)

Quotations that must stay consistent with whatever you write:
content/docs/guide/ci-cd-pipeline.md:2655 · scripts/check-required-check-set.mjs:37 and :41 (an inventory docblock at :34–:44 that names the carriers).

Pinning tests — all three will need to move with the prose:
scripts/__tests__/ci-cd-pipeline-doc.test.ts · scripts/__tests__/dependabot-merge-gate.test.ts · scripts/__tests__/check-required-check-set.test.ts

⚠️ Re-derive every line number yourself before editing. Mine are from 3adc50eec8; if your worktree base differs they will have moved.

The ruling you are executing

The four sentences were true when written — objectui#4959 measured a merge landing while all four shards were in_progress, which is only possible if the shards were not required. The ruleset's updated_at is 2026-08-24, after that reading. The world changed; the prose did not. ⇒ AGENTS.md commandment #9.

Direction: prefer a POINTER to the instrument over a restated fact. Restating an endpoint's answer in prose is what created this card; a sentence that says "run pnpm check:required-check-set" or "read GET /repos/{owner}/{repo}/rules/branches/main" cannot go stale. If you quote a figure or a list anyway, it carries when and how it was derived.

⭐ The instrument already exists and is the natural pointer target. scripts/check-required-check-set.mjs takes a live read of ruleset 11776024, exits 2 when the reading could not be taken (⛔ never a pass) and 3 on breach, is wired as pnpm check:required-check-set (package.json:90), and has a standing caller at .github/workflows/required-check-set-patrol.yml.

⚠️ Honour the prior author's read/write line — ⛔ do not sweep it away

check-required-check-set.mjs:44 says, of the three carriers it inventories:

The WRITE half of all three is still true and this file does not touch it.

⇒ a prior author considered these sentences and deliberately left them standing, on the argument that the write half ("nothing here can change it") is still true. That argument is sound: this seat's token reads GET /rules/branches/main fine but GET /branches/main/protection answers 403 Resource not accessible by integration.

⛔ So the repair is NOT "delete the sentences". Each one must end up saying what is true: the surface can be read (and here is how), and it cannot be written from the repository. Splitting the conjunction is the work. ⛔ Deleting a considered claim as though nobody had considered it is the failure mode.

⛔ Hard fences

1. ⛔ AGENTS.md:338 is GOVERNED surface and is OUT OF SCOPE. It carries a fourth instance (从仓内读不到). ⛔ Do not edit it, do not include it in this PR. It is also not obviously false on the same terms — its parenthetical covers 谁可绕过 (bypass actors), which GET /rules/branches/main does not carry. If you believe it must move for the rest to be coherent, STOP AND REPORT. ⛔ Do not decide it.
2. ⛔ Do not change the ruleset. The ruleset is correct; the prose is wrong. It is a maintainer-only surface (403 to this seat's token) and out of scope for any dev.
3. ⛔ Do not weaken, rename or re-scope any gate. This is a prose card. If a test pin forces a behavioural change to stay green, that is a stop-and-report, not a licence.
4. Every zero you report carries a control proven able to return non-zero in the same command, and you name the population in words.

Acceptance

• The four sentences and the three quotations are mutually consistent and each is true on the tree you land on.
• pnpm check:required-check-set and all three pinning test files pass.
• A changeset is present (this repo's check-changeset-presence.mjs exits 1 without one). If the change releases nothing, the repo's first-class spelling for that is empty frontmatter (---\n---) — see .changeset/9513-at-lib-prose-pointer.md on main for the landed precedent.
• Your report states, per file, how many non-comment / non-prose lines moved. For a card of this shape the honest answer may well be zero in both directions; say so if it is.

⚠️ Downstream, so you do not lose it

This card is a precondition for safely executing objectui#9499 (p1, in the maintainer decision box). #9499's repair renames or re-shards the four Test (shard N/4) jobs; an author who trusts any of these four sentences blocks every pull request in the repository on contexts that can never be produced again. ⛔ You are not executing #9499 — do not touch the shard jobs.

---

Generated by Claude Code