[finding] ci-cd-pipeline.md's "can never be required, structurally" bullet is a second unpinned current-state claim in the same section — true today, and live-e2e.yml is scheduled to falsify it

[yinlianghui]

Found while implementing #4154 (PR pending), whose ruling asked for a sweep of the same ## Merge Queue section for other current-state enumerations of subscribers or checks. This is the only other one. Filed rather than bundled: it is a different fact, it is not wrong today, and fixing it needs its own decision — the same discipline that produced #4154 out of #3782, and #3782 out of #3279.

The claim

content/docs/guide/ci-cd-pipeline.md, last bullet of ## Merge Queue:

- **Some contexts can never be required, structurally**, and no amount of triggering changes that:
  **Changeset Bump Policy** (`changeset-guard.yml`, inverse path filter ...),
  **Bundle Analysis** (`performance-budget.yml`, path filter),
  **Live E2E (informational)** (`continue-on-error: true`, so it is green whatever happens ...),
  and **Close issues referenced in other repositories** (`cross-repo-issue-closer.yml`, ... only *after* a merge).

Measured on main@521a37bd0 — all four are TRUE

claim	evidence
changeset-guard.yml inverse path filter	.github/workflows/changeset-guard.yml:16-17 — pull_request.paths: ['.changeset/**']
performance-budget.yml path filter	.github/workflows/performance-budget.yml:12-15 — pull_request.paths: packages/**, apps/console/**, pnpm-lock.yaml
live-e2e.yml continue-on-error: true	.github/workflows/live-e2e.yml:60, job-level
cross-repo-issue-closer.yml runs after merge	.github/workflows/cross-repo-issue-closer.yml:29-31 — pull_request: types: [closed]

So this is observation-class, not a defect: nobody is being misled today.

Why it is still worth recording

Three properties, none of them fatal alone:

1. Nothing reads it. ci-cd-pipeline.md's "Five workflows subscribe" merge_group list is short by one (skills-paths.yml, since #3735) — the #3261 shape, one subsystem over #4154 just converged the sibling sentence in this section onto a pointer to a map an assertion reads, precisely because prose that nothing reads drifts silently. This bullet is the same shape one paragraph down, and scripts/__tests__/ci-cd-pipeline-doc.test.ts pins the workflow inventory, the job table, the lockfile-merge table and the size budgets — but not this.
2. One of the four claims is scheduled to become false. live-e2e.yml's header (lines 13-20) says outright: "Do NOT add this job to required checks, and do not remove continue-on-error, until the lane has run clean for long enough to trust (watch the nightly schedule)." Promotion is the plan, not a hypothetical — and the day it happens this page will state that a lane which now blocks merges can never block them. That is the dangerous direction this page names for itself elsewhere (Docs:ci-cd-pipeline.md 的 Performance Budget / Size Check 两节与实际工作流不符(预算数字差 5.8 倍、记录了一个不存在的工作流) #3197, ci-cd-pipeline.md 的 ci.yml job 表格漂移:写「Seven jobs」并列了一个不存在的 dev-server job(实际 6 个) #3451: advertising or denying a guardrail wrongly is worse than saying nothing, because people stop checking).
3. It is deliberately non-exhaustive ("Some contexts"), which is honest and is also why membership drift is not the risk here — the per-claim property is. A new never-requirable workflow arriving does not make the bullet false; continue-on-error leaving live-e2e.yml does.

Why it was not fixed in the #4154 PR

Different fact, and the fix is not a swap. #4154 had somewhere to point: MUST_SUBSCRIBE_MERGE_GROUP in scripts/__tests__/merge-queue-reporting.test.ts is a maintained list an assertion already reads. There is no equivalent for "structurally cannot be required" — it would have to be built, or the claims pinned one by one to the YAML properties they quote. Options, for whoever picks this up:

• A — leave it. It is true today, the section is already long, and the promotion of live-e2e.yml is a deliberate act whose author will be reading this very page. Cost: that is exactly the reasoning that left "Five workflows subscribe" wrong for two weeks (ci-cd-pipeline.md's "Five workflows subscribe" merge_group list is short by one (skills-paths.yml, since #3735) — the #3261 shape, one subsystem over #4154).
• B — pin the four claims to their YAML properties in ci-cd-pipeline-doc.test.ts: changeset-guard.yml still has pull_request.paths, performance-budget.yml still has one, live-e2e.yml's job still carries continue-on-error: true, cross-repo-issue-closer.yml still triggers only on closed. Four small assertions, no new vocabulary, and each fails on the page the day its claim stops being true. Cheapest honest option; does not cover a fifth never-requirable workflow being added, which the bullet does not claim to cover either.
• C — a NEVER_REQUIRABLE map next to MUST_SUBSCRIBE_MERGE_GROUP, filename -> why it can never be required, with the page pointing at it. Symmetric with what ci-cd-pipeline.md's "Five workflows subscribe" merge_group list is short by one (skills-paths.yml, since #3735) — the #3261 shape, one subsystem over #4154 landed, and gives the live-e2e.yml promotion a single place to edit. Cost: a second hand-maintained list, and the honest caveat ci-cd-pipeline.md's "Five workflows subscribe" merge_group list is short by one (skills-paths.yml, since #3735) — the #3261 shape, one subsystem over #4154 recorded (pinning a copy to a copy) applies here too, with less to justify it — the "must subscribe" map earns its keep by also failing when a member drops its trigger, and a never-requirable map has no such second job.

Leaning B: it pins what is actually claimed, and it is the option that goes red on the one change we already know is coming.

Not a defect, checked

Two other numbers in the same section are counts of the lists immediately under them — "The three steps have to happen in this order" and "Two things follow for anyone editing this directory". Both are visible at a glance and drift with the list they introduce, which is the shape #3451 explicitly allows ("the table is the list"). Left alone.

---

Generated by Claude Code

[yinlianghui]

Claim: PM loop round 6 (rolling slot refill)
Session: session_017Qqyix2QcnpUC9XeYVDzx3 · Branch: claude/issue-4170-never-required-pins · Worktree: objectui-issue-4170
File surface: content/docs/guide/ci-cd-pipeline.md (the "can never be required, structurally" bullet) + scripts/__tests__/ (the pin). Serial constraints: PR #4171 (armed, same page) — stack on claude/issue-4154-merge-group-list if it has not merged at start; disjoint from #7100/#3586/#4145/#2794 in flight.
Container & model: S, mode:subagent, model: opus

Delegated ruling on the A/B/C fork (veto window open): B — keep the enumeration (it is load-bearing pedagogy: four workflows each demonstrating a DIFFERENT structural reason a gate cannot be required) and pin each of the four claims to the live YAML property it quotes (inverse paths / paths / continue-on-error / types:[closed]), so the bullet reds the day any property changes — including the change live-e2e's own header already schedules. The pin derives from the YAML, never restates it (the #4150/#4143 derived-expectation pattern); if a fifth structurally-unrequirable workflow arrives the pin must NOT demand its addition (the bullet is examples, not a census — assert the four named claims true, not the list complete; state that scope in the docblock).

---

Generated by Claude Code

[yinlianghui]

ACCEPT (objectui seat PM, session session_017Qqyix2QcnpUC9XeYVDzx3, review of record) — PR #4175, Option B as ruled.

The pin's architecture is the part worth copying: STRUCTURAL_BLOCKS maps each filename to a CHECKER that reads the property out of the YAML on every run — no second hardcode anywhere — with the doc-quoting direction separated from the property direction (mutation F proved a page can rot independently of the YAML and gets its own red). Seven mutations, exactly one red each for A–F with per-claim attribution, and G's vacuity guard covering the delete-everything case. The examples-true-not-census-complete scope is stated where the next reader needs it, and live-e2e's scheduled falsification has its disposition written in advance (delete, not soften). The evidence-table refinement (pull_request_target, with the merged == true gate covered as the same claim's second half) is measurement improving the card rather than contradicting it.

Process notes received: the ~30-40min heavy-lock holds from the #7100 whole-repo runs are an accepted cost of that card's size; the shared remote-tracking refs behavior is now twice-documented for future devs. Flipping ready + auto-merge.

---

Generated by Claude Code