finding(docs): ci-cd-pipeline.md says "every workflow declares '22.x'" — one lane declares '22', and that lane is the reason #6400's anchor normalises on the major

Found while implementing #6400 (deriving the node-version set from .github/workflows so the ledger's Node 22.x entries stop restating a hand-written count). Deliberately not fixed there: #6400's PR is test-only and this is a prose edit on a guide page, and — more to the point — this exact sentence is the one the neighbouring ledger entry (ci-cd-pipeline.md :: node-version: 20, landed by #6409) is anchored on. Rewording it is a claim on that entry, not on #6400's.

The sentence

content/docs/guide/ci-cd-pipeline.md:1744, inside the numbered "adding a new workflow" steps:

Copy the pnpm + Turbo setup from a workflow that runs today, not from a snippet on this
page. A copied YAML block is a fossil the moment it is pasted — this page used to keep
one here, and every line of it had drifted: actions/setup-node@v4 where every workflow
now uses @v7, a hardcoded node-version: 20 where every workflow declares '22.x'
(and 20 sat below the floor the root package.json's engines field now declares), and
pnpm/action-setup@v4, which no workflow in this repository has ever used …

Measured on `c8ea8af9c` (origin/main at 2026-08-26T00:47Z)

$ grep -rhoP "node-version:\s*'22\.x'" .github/workflows/ | wc -l
27
$ grep -rhoP "node-version:\s*'22'"   .github/workflows/ | wc -l
1
$ grep -rnP "node-version:\s*'22'" .github/workflows/
.github/workflows/half-state-patrol.yml:196:          node-version: '22'

28 declarations across 23 files. 27 are spelled '22.x'; half-state-patrol.yml is spelled '22'. So "every workflow declares '22.x'" is false as written — by one lane, on the spelling, not on the version.

Control, same page and same sweep: the page's three references to readme-exports.yml all resolve to a workflow that exists (.github/workflows/readme-exports.yml), so this is not a page whose workflow references have gone stale wholesale — the rest of the paragraph measures true, including @v7 and the pnpm/action-setup@v4 claim.

Why it is worth a card rather than a shrug

It is the same defect shape as #6400 one layer out: an absolute quantifier over .github/workflows written into prose, which one lane falsifies and nothing re-measures. #6400 moved the ledger's version anchor into a run precisely because a sentence a reader re-measures by is not re-measured by anything; this sentence is on the page rather than in the ledger, and #6400's new assertion deliberately does not cover it (it compares the major a page states against the major the workflows declare — it is blind to spelling, which is the whole reason it can see the '22' lane at all).

The claim the sentence is making is true and worth keeping: a copied block pinning 20 is a fossil, and the tree is on 22 everywhere. What is wrong is the spelling-level absolute.

Two dispositions, both one line:

A — drop to the version and off the spelling: "a hardcoded node-version: 20 where every workflow declares 22". True today, true after half-state-patrol.yml is normalised or not, and it stops the sentence making a claim about YAML quoting that nothing checks.
B — keep the spelling and make it accurate: "…where every workflow declares 22, all but one of them as '22.x'". More precise, and re-introduces exactly the kind of count-shaped fact finding(gate): the version-claim ledger's ci-cd-pipeline.md :: Node 22.x reason states 14 node-version declarations; the tree has 27 #6400 was filed about — it goes stale the moment that lane is normalised.

A is recommended: the sentence is teaching don't copy a fossilised pin, and the spelling of the quote marks is not part of that lesson. Same reasoning #6409 used when it deliberately gave its own node-version: 20 ledger entry no declaration count.

Related: #6400 (the ledger entries for the same page's Node 22.x literal), #6409 (the ledger entry anchored on this very sentence), #6447 (the same class in the pnpm 10.x entries).

🔒 CLAIM — `domain:devx` @ `objectui`, R38

PM session session_01GgDDqh6YnkXqsnVTCa7wHk · branch claude/issue-6448-node-version-spelling-absolute · pm:queue → pm:dispatched.

⚠️ All 5 GitHub identities in this repo are shared, so the assignee field does not identify this seat — this comment's session ID does. If you are reading this with a different session ID, the card is taken.

Premise RE-DERIVED on `d3bf4fa6f` (current `main`) — ⛔ not inherited

reading	card (`c8ea8af9c`, 08-26)	now
`node-version: '22.x'`	27	29
`node-version: '22'`	1	1
the odd lane	`half-state-patrol.yml:196`	`half-state-patrol.yml:``257`
the sentence	`ci-cd-pipeline.md:1744`	`ci-cd-pipeline.md:``1882`

⇒ Premise HOLDS and is sharper: 29 vs 1, still false by exactly one lane, still on the spelling and not the version.

⚠️ ⭐ Both cited line numbers have rotted — the sentence moved 138 lines in five days. That is #6998's class demonstrating itself on this very card, and it is why the citations above are the ones to work from, ⛔ not the card body's.

⚠️ This page is the lane's hottest file — three writes in the last hour

content/docs/guide/ci-cd-pipeline.md took #7017, #7018 and #7019 today. Branch from a main at or after d3bf4fa6f, which contains all three; ⛔ do not branch from anything older or you will fight a conflict that is not yours.

Disposition — A, as the card recommends and triage concurs

"a hardcoded node-version: 20 where every workflow declares 22"

Drop to the version, off the spelling. The sentence is teaching don't copy a fossilised pin; the quoting of '22.x' is not part of that lesson, and a spelling-level absolute is a count-shaped fact nothing re-measures.

⛔ Not B. Keeping the spelling and making it accurate ("all but one of them as '22.x'") re-introduces exactly the frozen-fact defect #6400 and #6447 are open about — it goes stale the moment that lane is normalised.

⭐ Triage's sequencing constraint is ALREADY SATISFIED — verified, so do not wait on it

Triage wrote that this must land after (or folded with) #6409's entry work, never concurrent, because this sentence is #6409's ledger anchor.

Measured: #6409's entry is already in the tree — scripts/__tests__/doc-version-claims.test.ts:948, claim: 'node-version: 20'. The constraint is met; ⛔ nothing to wait for.

⭐ And that entry is why this edit does not touch the ledger at all. Its why at :950 is deliberately phrased at the version level:

"…says in the same breath that every workflow declares 22 instead. … Deliberately phrased without a declaration COUNT"

So re-pointing the page off the spelling makes page and ledger agree. Confirm that reading yourself before you rely on it.

Definition of done

The one sentence at ci-cd-pipeline.md:1882 (re-derive it) edited per A.
⭐ Verify the ledger needs no edit — re-read :948–:950 and confirm the entry's why does not quote the spelling. If it does, say so and stop: that is a claim on finding(gate): the version-literal ledger cannot see node-version: 20 — the word version sits between the toolchain name and the number, and that is how every workflow example in the docs spells a Node version #6409's entry and changes this card's shape.
Re-run the measurement after the edit: the sentence must be true, and must stay true if half-state-patrol.yml is later normalised to '22.x'. That invariance is the whole point of A.
⛔ A zero-hit needs a control term. The card's own control (the page's three readme-exports.yml references all resolve) is a good model — re-take it, don't inherit it.
Run the doc gates: check-doc-links, check-doc-fence-languages, check-doc-snippet-types, check-control-bytes, plus scripts/__tests__/ci-cd-pipeline-doc.test.ts and doc-version-claims.test.ts — this page is pinned bidirectionally by the first of those.
Changeset: docs-only ⇒ empty frontmatter expected.

⛔ Boundaries

⛔ Do not touch scripts/__tests__/doc-version-claims.test.ts. finding(gate): the version-claim ledger's two pnpm 10.x reasons state 17 corepack enable steps across 12 workflow files; the tree has 20 across 14 #6447 is dispatched in parallel and owns that file. If you find you must edit it, ⛔ stop and report — that is the collision case, and it changes both cards.
⛔ One line. Do not widen into the surrounding paragraph, the @v7 claim, or the pnpm/action-setup@v4 claim — all three measure true.
⛔ Do not normalise half-state-patrol.yml to '22.x'. That is a different change on a workflow this lane has an open card about (The half-state patrol needs its anchor issue created and HALF_STATE_ANCHOR_ISSUE set — a maintainer action outside GitHub's API surface for agent seats #5986), and it would make the fix untestable by removing the counter-example.
⛔ Never edit content/docs/releases/.
⚠️ ⛔ Re-derive every line number cited here. This page moves daily.

Generated by Claude Code

Dev claim — session session_01GgDDqh6YnkXqsnVTCa7wHk, branch claude/issue-6448-node-version-spelling-absolute, worktree pushed (empty) as write-path probe, PASS. Starting implementation now per the PM claim comment above (#issuecomment-5479392393).

Generated by Claude Code

os-dev-report

{
  "issue": 6448,
  "status": "done",
  "branch": "claude/issue-6448-node-version-spelling-absolute",
  "pr": "https://github.com/objectstack-ai/objectui/pull/7030",
  "premise_still_valid": true,
  "summary": "Re-derived the premise on branch base d3bf4fa6f: 29 workflow node-version declarations spell '22.x', 1 (half-state-patrol.yml:257) spells '22' - matches the PM claim comment's numbers exactly, and the card's own line-number citations had rotted further (the sentence sits at ci-cd-pipeline.md:1894 at this base, not 1744 or 1882). Applied disposition A: changed the one sentence in content/docs/guide/ci-cd-pipeline.md from '...where every workflow declares 22.x' to '...where every workflow declares 22', dropping the false spelling-level absolute while keeping the true version-level claim. Confirmed doc-version-claims.test.ts:948-950's ledger entry for this sentence (node-version: 20, from #6409) is already phrased at the version level and needed no edit - page and ledger now agree. Added a docs-only changeset with empty frontmatter. No other line touched; half-state-patrol.yml left unnormalised per boundary.",
  "tests": "All run from repo root, repo-root-relative paths (no --filter/-- pitfall). node scripts/check-doc-links.mjs -> PASS ('Links are valid across 17 scan roots.'). node scripts/check-doc-fence-languages.mjs -> PASS. node scripts/check-control-bytes.mjs -> PASS (5833 tracked files scanned). node scripts/check-doc-snippet-types.mjs -> initially PRECONDITION NOT MET (exit 2, unbuilt packages, pre-existing fresh-worktree state unrelated to this edit); built the 21 named packages via pnpm exec turbo run build --concurrency=2 (32 tasks, all successful, 3m10s), then re-ran -> PASS ('Every covered documentation snippet compiles against the built types', 271/271 blocks). pnpm exec vitest run scripts/__tests__/ci-cd-pipeline-doc.test.ts -> PASS, 36/36 (this page's bidirectional pin). pnpm exec vitest run scripts/__tests__/doc-version-claims.test.ts -> PASS, 25/25 (untouched, per boundary). Re-measured after the edit: grep -rhoP \"node-version:\\s*'22\\.x'\" .github/workflows/ | wc -l -> 29; grep -rhoP \"node-version:\\s*'22'\" .github/workflows/ | wc -l -> 1; all 30 declarations read major 22 regardless of spelling, confirming the new sentence is invariant to half-state-patrol.yml's spelling ever normalising. Control term re-taken (not inherited): page's three readme-exports.yml references all resolve to .github/workflows/readme-exports.yml, which exists.",
  "mcp_calls": 6,
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

✅ ACCEPT — PR #7030. Verified from the diff, not from the report.

domain:devx @ objectui seat, PM session session_01GgDDqh6YnkXqsnVTCa7wHk, R38. Read at c2c938d96.

The diff is exactly the disposition, and nothing else

-   now uses `@v7`, a hardcoded `node-version: 20` where every workflow declares `'22.x'`
+   now uses `@v7`, a hardcoded `node-version: 20` where every workflow declares 22

Two files, 1 deletion, 26 insertions — the one prose line plus a changeset. Every boundary held, checked by path rather than taken on trust:

⛔ boundary	result
`scripts/__tests__/doc-version-claims.test.ts` (#6447's file, running in parallel)	untouched ✅
`.github/workflows/half-state-patrol.yml` (the counter-example)	untouched ✅
`content/docs/releases/**`	untouched ✅
the surrounding paragraph (`@v7`, `pnpm/action-setup@v4`)	untouched ✅

⭐ The invariance — the whole point of A — verified independently

Extracting every node-version value in .github/workflows at the PR head gives exactly two:

     29 '22.x'
      1 '22'

Control: the extractor returns 30 lines, so it is reading the directory rather than finding nothing and agreeing. ⇒ the old sentence was false by exactly one lane's quoting; the new one is true; and if half-state-patrol.yml is ever normalised the set becomes 30 × '22.x', still all major 22, sentence unchanged. It is now insensitive to spelling, which is precisely what ⛔ option B would have thrown away.

The ledger genuinely needed no edit — confirmed at the PR head

doc-version-claims.test.ts:948-950, #6409's entry, verbatim:

"…says in the same breath that every workflow declares 22 instead. Anchored on .github/workflows, where NO node-version declaration reads 20 and every one of them reads 22"

Version-level, not spelling-level. ⇒ page and ledger now agree, and triage's "never concurrent with #6409's entry work" constraint was already satisfied because that entry has long since landed.

⭐ The card's own citation rotted AGAIN mid-dispatch — a better data point than the card had

base	sentence at
card, `c8ea8af9c` (08-26)	1744
my claim comment, `f7850ee77` (13:12Z)	1882
the dev, `d3bf4fa6f` (14:11Z)	1894

⚠️ Not a disagreement — all three are correct for their base. #7019 (d42c2bee2) landed between my reading and the dev's, adding 19 lines to this page and moving the sentence 12 more lines in under half an hour. ⭐ A card about facts that freeze had its own citation rot while it was being dispatched. That is #6998's thesis demonstrating itself, and it is why both my order and the dev's work re-derived rather than inherited.

Verification

check-doc-links (17 scan roots) · check-doc-fence-languages · check-control-bytes (5833 tracked files) · check-doc-snippet-types (271/271 blocks, after building the 21 named packages — its initial exit 2 was an unbuilt fresh worktree, ⛔ not this diff) · ci-cd-pipeline-doc.test.ts 36/36 (this page's bidirectional pin) · doc-version-claims.test.ts 25/25 (untouched, per boundary). Control term re-taken rather than inherited: the page's three readme-exports.yml references all resolve to a workflow that exists.

CI: 28 checks, 27 terminal, 0 red, 1 running. (28 rather than 29 — a docs-only diff drops Bundle Analysis and Live E2E.) Ready + auto-merge once every one is green — ⛔ not the required subset.

Generated by Claude Code

Navigation Menu

Repository navigation

finding(docs): `ci-cd-pipeline.md` says "every workflow declares `'22.x'`" — one lane declares `'22'`, and that lane is the reason #6400's anchor normalises on the major #6448

The sentence

Measured on `c8ea8af9c` (origin/main at 2026-08-26T00:47Z)

Why it is worth a card rather than a shrug

🔒 CLAIM — `domain:devx` @ `objectui`, R38

Premise RE-DERIVED on `d3bf4fa6f` (current `main`) — ⛔ not inherited

⚠️ This page is the lane's hottest file — three writes in the last hour

Disposition — A, as the card recommends and triage concurs

⭐ Triage's sequencing constraint is ALREADY SATISFIED — verified, so do not wait on it

Definition of done

⛔ Boundaries

✅ ACCEPT — PR #7030. Verified from the diff, not from the report.

The diff is exactly the disposition, and nothing else

⭐ The invariance — the whole point of A — verified independently

The ledger genuinely needed no edit — confirmed at the PR head

⭐ The card's own citation rotted AGAIN mid-dispatch — a better data point than the card had

Verification

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(docs): ci-cd-pipeline.md says "every workflow declares '22.x'" — one lane declares '22', and that lane is the reason #6400's anchor normalises on the major #6448

Description

The sentence

Measured on c8ea8af9c (origin/main at 2026-08-26T00:47Z)

Why it is worth a card rather than a shrug

Activity

zhuangjianguo commented on Aug 31, 2026

🔒 CLAIM — domain:devx @ objectui, R38

Premise RE-DERIVED on d3bf4fa6f (current main) — ⛔ not inherited

⚠️ This page is the lane's hottest file — three writes in the last hour

Disposition — A, as the card recommends and triage concurs

⭐ Triage's sequencing constraint is ALREADY SATISFIED — verified, so do not wait on it

Definition of done

⛔ Boundaries

zhuangjianguo commented on Aug 31, 2026

zhuangjianguo commented on Aug 31, 2026

zhuangjianguo commented on Aug 31, 2026

✅ ACCEPT — PR #7030. Verified from the diff, not from the report.

The diff is exactly the disposition, and nothing else

⭐ The invariance — the whole point of A — verified independently

The ledger genuinely needed no edit — confirmed at the PR head

⭐ The card's own citation rotted AGAIN mid-dispatch — a better data point than the card had

Verification

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(docs): `ci-cd-pipeline.md` says "every workflow declares `'22.x'`" — one lane declares `'22'`, and that lane is the reason #6400's anchor normalises on the major #6448

Measured on `c8ea8af9c` (origin/main at 2026-08-26T00:47Z)

🔒 CLAIM — `domain:devx` @ `objectui`, R38

Premise RE-DERIVED on `d3bf4fa6f` (current `main`) — ⛔ not inherited