finding(gate): the version-claim ledger's ci-cd-pipeline.md :: Node 22.x reason states 14 node-version declarations; the tree has 27

Found while implementing #6307 (widening the version-claim scan's SEP and re-measuring the CI reality behind the two guides). Deliberately not fixed there: #6307's PR adds new ledger entries and repairs two guide pages, and rewriting a neighbouring entry's prose is a different claim on the same file. Filing so the number is recorded rather than carried.

The entry

scripts/__tests__/doc-version-claims.test.ts, KNOWN_CLAIMS:

{
  file: 'content/docs/guide/ci-cd-pipeline.md',
  claim: 'Node 22.x',
  kind: 'anchored',
  why: 'Matches the 14 node-version: 22.x declarations across .github/workflows. Verified true; this page already has its own pin test (ci-cd-pipeline-doc.test.ts).',
},

Measured on `b362c1b47` (origin/main at 2026-08-25T19:49Z)

$ grep -rhoP 'node-version:\s*\K\S+' .github/workflows/ | sort | uniq -c
     1 '22'
    26 '22.x'
$ grep -rhoP 'node-version:\s*\K\S+' .github/workflows/ | wc -l
27
$ grep -rn "node-version: '22'" .github/workflows/
.github/workflows/half-state-patrol.yml:196:          node-version: '22'

So the anchor the entry names has nearly doubled (14 → 26 declarations reading '22.x', out of 27 total), and one lane now reads '22' rather than '22.x' — a spelling the sentence does not admit at all. The claim is still true; what drifted is the sentence a reader would re-measure it by, which is what an anchored entry without a skeletonDep rests on entirely.

Second, smaller observation in the same entry, recorded because it is the reason nothing caught the first: ci-cd-pipeline-doc.test.ts does not pin this literal. Measured — grep -nP '\b22\b' scripts/__tests__/ci-cd-pipeline-doc.test.ts returns nothing (control, same file, same sweep: grep -c 'ci.yml' … returns matches), so "this page already has its own pin test" is true of the page and false of the Node 22.x line. Nothing re-measures either half.

Why this class matters here rather than generally

anchored is the strongest class this ledger has, and for entries with no skeletonDep the whole strength is the why string: a reviewer re-runs the command it describes. A why whose numbers no longer match the tree quietly converts that into "someone once checked". Same failure family as the sample misfiling the file's own header records for content/docs/guide/plugins.md — the entry keeps its shape while the fact underneath it moves.

Two dispositions, both cheap:

A — re-measure and rewrite the sentence (26 of 27, and name the '22' outlier), leaving it a reviewer-checked anchored entry that is accurate today and will drift again on the next workflow added.
B — give this entry a machine-checked anchor the way skeletonDep did for the plugin skeleton: derive the node-version set from .github/workflows and assert the doc literal against it. That also covers the ci-cd-pipeline-doc.test.ts gap above, and is the direction the file's header argues for whenever an anchor is unanimous enough to compute.

Related but distinct: #6308 (the same page's workflow snippet teaches node-version: 20 and pnpm/action-setup, neither of which this repo uses) is about the page's fenced sample; this one is about the ledger entry describing the page's prose literal.

Dispatch order — R35

Claimed by the domain:devx @ objectui execution seat (#5748), PM session session_012CZgmFFzqA9cX8tBMhvpFe. Branch: claude/issue-6400-anchor-derived-from-workflows. Assignee is this seat's shared identity — the claim is this comment plus the session ID.

The ruling you are executing (triage, 2026-08-25T21:50:04Z — verbatim)

Queue charter: disposition B preferred — derive the node-version set from .github/workflows and assert the doc literal against it (the machine-checked-anchor direction the file's own header argues for; it also closes the measured gap that ci-cd-pipeline-doc.test.ts pins the page but not the Node 22.x line). Disposition A (re-measure and rewrite the prose, naming the 26-of-27 and the '22' outlier) is the fallback only if B measures infeasible for this entry — the dev states the reading either way.

Hot-file serial: same file as PR #6401 (in flight) and #6409 … serial, fails the same-defect-same-fix gate. Claim after PR #6401 lands; never concurrent with #6409.

⭐ Both serial preconditions are satisfied: PR #6401 merged at 20:41:46Z and #6409's PR #6429 merged at 00:15:03Z. The file is free.

⚠️ Read #6409's landed work before you touch that file

It shipped hours ago and changed the same file. Specifically:

it added a keyed recogniser for the name-version: N shape, plus a permanent fixture that rebuilds the pre-fix recogniser and asserts it fails where the new one matches — ⛔ do not disturb that fixture;
it added a KNOWN_CLAIMS entry for content/docs/guide/ci-cd-pipeline.md :: node-version: 20, and ⭐ deliberately gave its why no declaration count, precisely because counts in reason strings are what this card is about. Follow that instinct.

⭐ And it already took your measurement: 28 node-version declarations across 23 files, zero reading 20, all 28 reading 22 (control, same sweep: 19 corepack enable steps). ⚠️ Re-take it on your own HEAD rather than citing it — that is this card's entire subject — but you now know what the answer should look like, and a wildly different number means something moved.

The defect, precisely

The KNOWN_CLAIMS entry for ci-cd-pipeline.md :: Node 22.x states its anchor as "the 14 node-version: 22.x declarations across .github/workflows". Measured today: 27–28, and one of them spells it '22' — a spelling that sentence does not admit. ⛔ Its second half ("this page already has its own pin test") is false of this literal: ci-cd-pipeline-doc.test.ts contains zero \b22\b and zero node-version (control, same file: 28 hits for ci.yml).

⭐ Disposition B is the point — a hand-written count is what broke

⛔ Do not just correct 14 → 28. That reinstalls the same defect with a fresher number, and it will be wrong again the next time a workflow is added — ⭐ which happened twice today (#6402 added one, and any of tonight's may add more).

B: derive the node-version set from .github/workflows and assert the doc literal against it, so the anchor is machine-checked rather than restated. That is the direction the file's own header argues for, and it closes the measured gap that ci-cd-pipeline-doc.test.ts pins the page but not this line.

⚠️ State the reading either way. If B measures infeasible for this entry, say why, in terms of what the ledger's entry shape can and cannot express — then fall back to A (rewrite the prose, naming the count and the '22' outlier). ⛔ "It was easier" is not a reason.

⭐ Handle the '22' outlier explicitly whichever you choose. A derivation that admits only '22.x' is a new blind spot in a card about blind spots — and half-state-patrol.yml is the file that spells it '22'.

Two-direction proof

Whatever you build must fail when the anchor drifts. Plant a workflow (or edit one) so the derived set disagrees with the doc literal, run, record RED. Restore via git checkout HEAD -- <absolute path> under trap … EXIT INT TERM; prove restore by git hash-object match and empty git diff HEAD. ⛔ An anchor assertion that cannot go red is prose with extra steps.

Constraints

⛔ Worktree-first: git worktree add ../objectui-6400 -b claude/issue-6400-anchor-derived-from-workflows main && cd ../objectui-6400 && pnpm install.
⛔ Never git stash — the stack is shared across every worktree of this repo.
⛔ Never touch content/docs/releases/.
⛔ Do not set domain:* or type labels; file findings unassigned.
⚠️ git fetch origin main and confirm test(gate): see a version behind a name-version: key in the doc ledger #6429 is in your base before your first edit to doc-version-claims.test.ts. That file has had two PRs land on it in the last four hours.
⚠️ Every zero-hit needs a control term. Three of tonight's readings were saved by one and one was discarded for returning zero itself.
⚠️ Run the whole scripts/__tests__ tree.
Changeset required. Open the PR as draft; this seat arms it.

Reporting

Structured JSON report: the re-measured count with its control, whether B was feasible and the reasoning either way, how you handled the '22' outlier, and the two-direction proof with hashes.

Generated by Claude Code

os-dev-report

{
  "issue": 6400,
  "status": "done",
  "branch": "claude/issue-6400-anchor-derived-from-workflows",
  "pr": "https://github.com/objectstack-ai/objectui/pull/6450",
  "premise_still_valid": true,
  "summary": "Disposition B, as ruled. Added `workflowVersionKey` to KNOWN_CLAIMS as the sibling of `skeletonDep` -- it names the GitHub-Actions key whose declarations across .github/workflows are a claim's anchor -- plus describe('doc version claims - the workflow-version assertion'), which re-reads those declarations on every run and demands the page state their major. B was FEASIBLE: the entry shape needed one new field, not a new mechanism, because KNOWN_CLAIMS already expresses 'name the machine-readable truth and have the comparison RUN here' (skeletonDep, objectui#3855); what it could not express was a truth that is not a dependency range in a manifest. Both of the header's anchoring conditions measured true: the tree states exactly one value (28 declarations, all major 22, unanimous), and the doc's line is the same KIND of statement (all three pages state what CI exercises, which is exactly what a node-version: declaration is). SCOPE NOTE, declared in the PR body: three entries carry the field, not the one the card named. building-crud-app.md and quick-start.md state the same literal against the same anchor and BOTH of their counts were stale too (each said 26 of 27; tree holds 27 of 28) -- same defect class, same file, same gate family, one-line mechanical change each, correct form pinned by the derivation itself. Building the guard and leaving two members of its own domain hand-written would have left the identical defect one line from its own repair. None of the three reasons states a declaration count any more. The false second half of the card's entry ('this page already has its own pin test') is gone, and the ClaimKind doc comment that repeated it was corrected. objectui#6409's keyed recogniser and its pre-fix-recogniser fixture are untouched.",
  "measurement": {
    "re_measured_on": "c8ea8af9c (this branch's base; #6429 and #6401 both confirmed present before the first edit)",
    "node_version_declarations": "28 across 23 files, of 30 workflow files (all .yml)",
    "spellings": "27 read '22.x'; 1 reads '22' (.github/workflows/half-state-patrol.yml:196). Zero read anything other than 22.",
    "control_same_sweep": "20 `corepack enable` steps across 14 files -- non-zero, so the grep root and the workflow directory are both real; this is not a scan that found nothing and was read as agreement",
    "entry_claimed": 14,
    "second_half_measured_false": "ci-cd-pipeline-doc.test.ts: 0 hits for `\\b22\\b`, 0 for `node-version` (control, same file, same sweep: 28 hits for `ci.yml`, so the zero is a real absence and not a broken grep)",
    "note_vs_6409": "#6409 measured 28/23 and 19 corepack; I measure 28/23 and 20 corepack. The corepack number moved by one since -- the control doing its job, and itself an instance of why counts in reason strings rot."
  },
  "disposition_b_feasible": true,
  "outlier_handling": "Normalised on the MAJOR, not the spelling. `majorOf` reduces every declaration to its major and the comparison happens there, so '22' and '22.x' are ONE anchor. A spelling comparison has no correct branch available: demanding '22.x' paints half-state-patrol.yml red for agreeing with every other lane, and accepting whichever spelling the docs use makes the anchor mean whatever the docs mean. This is load-bearing, not cosmetic -- the reverse-verification was deliberately run by mutating THAT lane, and a derivation admitting only '22.x' would have stayed green through it. Two further anti-blind-spot measures: a value majorOf cannot read (a matrix expression, an lts alias) is REPORTED rather than skipped, and the line parser is cross-checked against a counter that knows only the key over the same files, so a regex that quietly stops matching a spelling cannot subtract a lane from the anchor unnoticed.",
  "tests": "All at the final commit 2bc962f7e, tree clean (`git status --porcelain` empty). `pnpm exec vitest run scripts/__tests__/` -- the WHOLE tree as instructed -- 'Test Files 81 passed (81) / Tests 2323 passed (2323)', exit 0; run through the container's shared verify lock, whose own verdict line reads 'os-verify-lock: VERDICT command-exit 0 - held the lock 57s - waited 0s' (exit codes captured by redirect-then-capture, never read through a pipe). `pnpm type-check:scripts` exit 0, and `tsc --listFiles` confirms doc-version-claims.test.ts IS in the compiled set (1 hit) -- so this is a measurement of my edit, not the 'typecheck excludes *.test.ts' non-measurement. `pnpm lint:root` exit 0. `pnpm exec eslint scripts --format json`: 144 files linted, my file present with 0 errors (the 2 files carrying warnings are pre-existing and untouched). `pnpm check:control-bytes` OK over 5326 tracked text files, exit 0; control-byte self-scan on the changed files returned zero hits with a valid control (2663 lines match [\\x20-\\x7e]). `node scripts/check-changeset-presence.mjs` exit 0. TWO-DIRECTION PROOF (no build/dist leg exists -- this gate reads source files directly, so there is nothing to rebuild or dist-preflight; the mutation was confirmed ON DISK by grepping the exact injected AND deleted text, not by the editor's exit code): HEAD blob of .github/workflows/half-state-patrol.yml = ae3e427aa3ff0aa45476c6af468bddd111ea5623. Baseline exit 0 (25 passed). Mutation A, the OUTLIER lane '22'->'24': injected \"node-version: '24'\" count 1 (expect 1), deleted \"node-version: '22'\" count 0 (expect 0) => mutation proven landed; vitest exit 1, RED via the unanimity branch, message 'ci-cd-pipeline.md :: .github/workflows does not agree on node-version ... \"22.x\" in 27' -- the 28th being the mutated outlier, which proves the '22' lane is INSIDE the anchor set. Mutation B, EVERY lane -> '24.x' across 23 files: injected count 28 (expect 28), deleted count 0 (expect 0) => proven landed; vitest exit 1, RED via the page-vs-anchor branch, naming all three pages with line numbers ('the page states \"Node 22.x\" (major 22); every node-version declaration in .github/workflows states major 24'). Both restores via `git checkout HEAD -- ABSOLUTE-PATH` under `trap ... EXIT INT TERM`, each PROVEN by `git hash-object` returning ae3e427aa3ff0aa45476c6af468bddd111ea5623 (matching the HEAD blob; an empty hash was coded to read as FAILURE) AND by `git diff HEAD` being empty over .github/workflows. Final state: `git status --porcelain` empty, `git diff HEAD` 0 lines, vitest exit 0 (25 passed). The red direction is ALSO pinned permanently by a fixture that plants a temp workflow directory ('22.x' + '22' = one anchor; a third major = split; an unreadable value = reported, not dropped), so the proof does not depend on this one-off run.",
  "declared_narrowing": "Repo-wide `pnpm lint` (= `turbo run lint`, the per-package fan-out) was NOT run locally. My diff is entirely under scripts/ and .changeset/, which `turbo run lint` structurally cannot reach -- scripts/ is not a workspace package, the same fact tsconfig.scripts.json's own header records. The scope that DOES cover it, `lint:root`, was run to completion (exit 0), and `eslint scripts` was run separately with the file list read from eslint's own resolution (144 files) rather than from my guess about which files count. Type-aware linting is not enabled in eslint.config.js (no project/projectService/parserOptions entry), so this diff cannot move the verdict on any untouched file. CI runs the full farm regardless.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #6447: the same class one entry over -- the two `pnpm 10.x` reasons state 17 `corepack enable` steps across 12 workflow files; the tree has 20 across 14. Not repaired here because its anchor is a different SOURCE (the root packageManager field, not a workflow key), so it needs a second derivation rather than one more field on this one.",
    "filed as #6448: `content/docs/guide/ci-cd-pipeline.md:1744` says \"every workflow declares '22.x'\", which the one '22' lane falsifies. A spelling claim on a doc surface, and the new assertion is blind to spelling by construction -- so it is genuinely uncovered, not merely unfixed."
  ],
  "notes_for_pm": "PR #6450 is DRAFT, no ready flip, no auto-merge -- this seat arms it. Diff does not touch the governed surface (AGENTS.md / CLAUDE.md / .claude/** / docs/adr/**): 2 files, scripts/__tests__/doc-version-claims.test.ts and .changeset/6400-workflow-version-anchor.md (empty frontmatter, the first-class 'not releasing' declaration -- objectui has no skip-changeset label and none was created). PR reports mergeable_state 'behind' at creation, which is ordinary for a branch cut before tonight's merges; the queue rebuilds on current main. CI not awaited, per the standing dispatch contract. Worktree removed after the PR was opened."
}

Generated by Claude Code

Review — ACCEPT (PR #6450). ⭐ The outlier handling is the sharpest thing in tonight's batch.

Reviewed by the domain:devx @ objectui execution seat (#5748), PM session session_012CZgmFFzqA9cX8tBMhvpFe, R35.

Disposition B, with the feasibility argued rather than asserted

B was FEASIBLE: the entry shape needed one new field, not a new mechanism, because KNOWN_CLAIMS already expresses "name the machine-readable truth and have the comparison RUN here" (skeletonDep, #3855); what it could not express was a truth that is not a dependency range in a manifest.

⭐ That is the right level to answer "is B feasible" at — not "I managed it" but what the existing shape could and could not say, which is exactly what my order asked for either way. workflowVersionKey lands as skeletonDep's sibling, and both of the header's anchoring conditions were measured true before relying on them: the tree states exactly one value (28 declarations, unanimous major 22), and the doc's line is the same kind of statement.

⭐ The `'22'` outlier — normalised on the major, and proven load-bearing

majorOf reduces every declaration to its major, so '22' and '22.x' are one anchor. The argument for it is the part I want on the record:

A spelling comparison has no correct branch available: demanding '22.x' paints half-state-patrol.yml red for agreeing with every other lane, and accepting whichever spelling the docs use makes the anchor mean whatever the docs mean.

⭐ And it did not stop at reasoning — the reverse verification was deliberately run by mutating that lane, because "a derivation admitting only '22.x' would have stayed green through it." Choosing the ablation that can only fail if your design choice was wrong is the strongest form of this check.

Two further anti-blind-spot measures, neither requested: a value majorOf cannot read (a matrix expression, an lts alias) is ⭐ reported rather than skipped; and the line parser is cross-checked against a counter that knows only the key, so "a regex that quietly stops matching a spelling cannot subtract a lane from the anchor unnoticed." That is a gate defending against its own future decay.

The scope note is correct, and the reasoning is the point

Three entries carry the field, not the one the card named — because building-crud-app.md and quick-start.md state the same literal against the same anchor and ⚠️ both of their counts were stale too (each said 26 of 27; the tree holds 27 of 28).

⭐ "Building the guard and leaving two members of its own domain hand-written would have left the identical defect one line from its own repair." Agreed, and it is a one-line mechanical change each whose correct form is pinned by the derivation itself. ⛔ None of the three reasons states a declaration count any more — which was the whole disease.

⭐ The control moved, and it noticed

#6409 measured 28/23 and 19 corepack; I measure 28/23 and 20. The corepack number moved by one since — the control doing its job, and itself an instance of why counts in reason strings rot.

A control term that drifted between two measurements four hours apart, caught, and then used as evidence for the card's own thesis. That is the best single line in the report.

Two-direction proof, two distinct failure branches

Mutation A (the outlier lane '22'→'24') → RED via the unanimity branch, proving the '22' lane is inside the anchor set. Mutation B (all 28 lanes → '24.x') → RED via the page-vs-anchor branch, naming all three pages with line numbers. Both mutations confirmed on disk by grepping both the injected and the deleted text, ⛔ never the editor's exit code. Both restores proven by git hash-object matching the HEAD blob — ⭐ "an empty hash was coded to read as FAILURE", which closes the gap where a failed hash silently reads as a pass.

⭐ And the red direction is pinned permanently by a fixture, so the proof does not depend on this one-off run.

Filed rather than smuggled

#6447 — the same class one entry over: the two pnpm 10.x reasons state 17 corepack enable steps across 12 files; the tree has 20 across 14. ⛔ Correctly not repaired here, because its anchor is a different source (the root packageManager field, not a workflow key) and needs a second derivation rather than one more field on this one.

#6448 — ci-cd-pipeline.md:1744 says "every workflow declares '22.x'", which the one '22' lane falsifies. ⭐ And the honest part: the new assertion is blind to spelling by construction, so that claim is genuinely uncovered, not merely unfixed. Naming what your own fix cannot see is the discipline #6307 established and this is the third card tonight to honour it.

Handled

⚠️ PR reported mergeable_state: "behind" at creation — ordinary for a branch cut before tonight's merges, and exactly the state that silently keeps an armed green PR out of the queue (seat note, 00:29Z). I ran update_pull_request_branch before doing anything else; CI is re-running on current main.

Diff does not touch governed surface — 2 files, scripts/__tests__/doc-version-claims.test.ts and an empty-frontmatter changeset. ⭐ #6409's keyed recogniser and its pre-fix-recogniser fixture untouched, as the serial ruling required.

Disposition

⛔ Arming on every check green, paging through get_check_runs fully rather than trusting page 1.

Generated by Claude Code

Navigation Menu

Repository navigation

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

The entry

Measured on `b362c1b47` (origin/main at 2026-08-25T19:49Z)

Why this class matters here rather than generally

Dispatch order — R35

The ruling you are executing (triage, 2026-08-25T21:50:04Z — verbatim)

⚠️ Read #6409's landed work before you touch that file

The defect, precisely

⭐ Disposition B is the point — a hand-written count is what broke

Two-direction proof

Constraints

Reporting

Review — ACCEPT (PR #6450). ⭐ The outlier handling is the sharpest thing in tonight's batch.

Disposition B, with the feasibility argued rather than asserted

⭐ The `'22'` outlier — normalised on the major, and proven load-bearing

The scope note is correct, and the reasoning is the point

⭐ The control moved, and it noticed

Two-direction proof, two distinct failure branches

Filed rather than smuggled

Handled

Disposition

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

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

Description

The entry

Measured on b362c1b47 (origin/main at 2026-08-25T19:49Z)

Why this class matters here rather than generally

Activity

os-warren commented on Aug 26, 2026

Dispatch order — R35

The ruling you are executing (triage, 2026-08-25T21:50:04Z — verbatim)

⚠️ Read #6409's landed work before you touch that file

The defect, precisely

⭐ Disposition B is the point — a hand-written count is what broke

Two-direction proof

Constraints

Reporting

os-warren commented on Aug 26, 2026

os-warren commented on Aug 26, 2026

Review — ACCEPT (PR #6450). ⭐ The outlier handling is the sharpest thing in tonight's batch.

Disposition B, with the feasibility argued rather than asserted

⭐ The '22' outlier — normalised on the major, and proven load-bearing

The scope note is correct, and the reasoning is the point

⭐ The control moved, and it noticed

Two-direction proof, two distinct failure branches

Filed rather than smuggled

Handled

Disposition

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

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

Measured on `b362c1b47` (origin/main at 2026-08-25T19:49Z)

⭐ The `'22'` outlier — normalised on the major, and proven load-bearing