docs(guide): replace invented Node/pnpm floors with what CI exercises, and let the version-claim scan see emphasis

[os-warren]

Fixes#6307

Both halves, in the ruled order: half 2 first, because it defines half 1's surface.

All readings below were taken on 992bcec7f, the final commit of this branch.

Half 2 — the scan was blind to **bold**

SEP, the class between a toolchain name and its version, admitted backticks, quotes, whitespace, colons, commas, pipes and brackets — and not *. It now admits * and _:

-const SEP = '[' + TICK + '\'"\\s:,|)\\]]{0,6}(?:[-—]\\s*)?[' + TICK + '\'"]?\\s*';+const SEP = '[' + TICK + '\'"\\s:,|)\\]*_]{0,6}(?:[-—]\\s*)?[' + TICK + '\'"]?\\s*';

Blind, then seeing — the gate's own output in both states

Before (b362c1b47, the branch point), pnpm exec vitest run scripts/__tests__/doc-version-claims.test.ts:

 Test Files 1 passed (1)
Tests 18 passed (18)

Green — and green here does not mean "examined and found sound". The two pages carrying **Node.js** 20+ appear in no KNOWN_CLAIMS entry and in no failure, because the scan never matched them.

Proven positively rather than inferred from a green run, with a control in the same sweep. The probe lifts OP, VERSION, FIRST_PARTY, TOOLCHAIN, TICK, SEP and CLAIM_RES verbatim out of the test file's own source and evaluates them, so the only difference between the two runs is that file's SEP line:

# SEP as of b362c1b47
content/docs/guide/quick-start.md -> 0 claim(s)
content/docs/guide/building-crud-app.md -> 0 claim(s)
content/docs/guide/ci-cd-pipeline.md -> 1 claim(s) <- CONTROL, matched
content/docs/guide/ci-cd-pipeline.md:196: Node 22.x
# SEP with * and _
content/docs/guide/quick-start.md -> 2 claim(s)
content/docs/guide/quick-start.md:12: Node.js** 20+
content/docs/guide/quick-start.md:13: pnpm** 9+
content/docs/guide/building-crud-app.md -> 2 claim(s)
content/docs/guide/building-crud-app.md:12: Node.js** 20+
content/docs/guide/building-crud-app.md:12: pnpm** 9+
content/docs/guide/ci-cd-pipeline.md -> 1 claim(s) <- CONTROL, still matched
content/docs/guide/ci-cd-pipeline.md:196: Node 22.x

After the SEP line changed and before anything else, the gate itself goes red and names them:

AssertionError: These doc surfaces state a version and nothing in this repository can tell whether it is still true:
- content/docs/guide/building-crud-app.md:12 "Node.js** 20+"
- content/docs/guide/building-crud-app.md:12 "pnpm** 9+"
- content/docs/guide/quick-start.md:12 "Node.js** 20+"
- content/docs/guide/quick-start.md:13 "pnpm** 9+"

Corpus-wide delta, over the 241 files the three SCAN_ROOTS resolve to: 33 matched literals before, 37 after — four new, none lost. The four are exactly the ones above.

Triage — every literal the widened scan surfaced

Literal	Disposition	Reason
quick-start.md:12**Node.js** 20+	Deleted, no ledger entry	A floor nothing declares and nothing tests — see half 1. The gate's own failure message prefers deleting over ledgering for exactly this case, and the downward half of the ratchet forbids an entry naming a claim no longer in the tree.
quick-start.md:13**pnpm** 9+	Deleted, no ledger entry	Same.
building-crud-app.md:12**Node.js** 20+	Deleted, no ledger entry	Same.
building-crud-app.md:12**pnpm** 9+	Deleted, no ledger entry	Same.

No other literal anywhere in the three scan roots changed state, so nothing was bulk-added to keep the gate quiet. The four entries this PR does add to KNOWN_CLAIMS are for the sentences that replace them, each anchored and each naming the anchor it can be re-measured against.

The widening keeps a witness after the repair

After half 1 the corpus holds no emphasised claim at all, and the replacement sentences separate name from version with a plain space. So on the corpus alone, reverting SEP would change nothing observable and this gate would report green over the same blind spot — the second-order form of the defect. it('reads a claim through markdown emphasis, so the widened separator is not decorative') is the permanent witness: a fixture through claimsIn, plus the pre-#6307 class rebuilt inline and asserted not to match the same line.

Reverse-verified on the committed tree (mutation confirmed on disk by blob hash 51dc12f… to 2d1d830… and by the widened spelling going to 0 occurrences; restored by git checkout HEAD -- plus the absolute path, verified back to 51dc12f… with git diff HEAD empty):

 ❯ reverting SEP to its pre-#6307 spelling
FAIL reads a claim through markdown emphasis, so the widened separator is not decorative
AssertionError: a toolchain name wrapped in markdown emphasis must still produce a claim:
expected [] to deeply equal [ 'Node.js** 20+', 'pnpm* 9+' ]
Test Files 1 failed (1)
Tests 1 failed | 18 passed (19)

Predicted direction and observed direction agree, including which test goes red: only the new fixture. The ratchet stays green, which is the whole reason the fixture had to exist.

The boundary the widening does not cross, pinned rather than assumed

_ is in the class and costs nothing, but underscore emphasis around a name is still invisible and no character class can reach it: _ is a word character, so the \b on each side of TOOLCHAIN fires on neither side of _Node.js_, and the match fails before SEP is consulted. Measured across the three scan roots: zero underscore-emphasised toolchain names (control, same sweep: six files carry the bold spelling, e.g. content/docs/utilities/runner.mdx:446). Recorded as a documented boundary with an assertion that goes red if it ever changes, not repaired by widening \b into a lookaround nothing has asked for.

Half 1 (ruled B) — say what is true, not a floor nobody measured

Both pages, identically:

-- **Node.js** 20+-- **pnpm** 9+ or npm/yarn+- **Node.js** and **pnpm** (or npm/yarn) — ObjectUI is tested on Node 22.x with pnpm 10.x.

It leads with what the reader installs, and states the measured fact as what this project tests on — never as a requirement. (First wording on this branch put the engines audit in the bullet itself; that reasoning lives in the changeset, the file docblock and the card, not on a getting-started page.)

Re-measured on this branch's base rather than copied from the card's 12:16Z reading:

• Node — 27 node-version: declarations across .github/workflows: 26 read '22.x', the 27th (half-state-patrol.yml:196) reads '22'. Nothing runs Node 20.
• pnpm — root packageManager is pnpm@10.31.0, and 17 run: corepack enable steps across 12 workflow files mean that field is what CI installs with. (pnpm install in this worktree reported pnpm v10.31.0.)
• No consumer floor exists to restate — of 46 workspace manifests, zero declare engines.node or engines.pnpm; the only engines block outside the root is packages/vscode-extension's engines.vscode (control for that zero: the root manifest, which does declare node: ">=22.11" / pnpm: ">=10").

Root engines is deliberately not what these pages cite: they address the reader's own project, which is why #5306 left them alone.

Both literals the ledger pins on these pages survive the rewrite verbatim — one occurrence each, per page:

content/docs/guide/quick-start.md | Node 22.x -> 1 :12
content/docs/guide/quick-start.md | pnpm 10.x -> 1 :12
content/docs/guide/building-crud-app.md | Node 22.x -> 1 :12
content/docs/guide/building-crud-app.md | pnpm 10.x -> 1 :12

Verification

Run at 992bcec7f (the final commit), from the repo root:

Check	Result
pnpm exec vitest run scripts/__tests__/doc-version-claims.test.ts	Test Files 1 passed (1) / Tests 19 passed (19)
pnpm exec vitest run scripts/__tests__/ (whole gate-test tree)	Test Files 79 passed (79) / Tests 2280 passed (2280)
node scripts/check-doc-links.mjs	Links are valid across 17 scan roots.
node scripts/check-doc-fence-languages.mjs	✅ check:doc-fences — every TypeScript block in 223 document(s) …
pnpm run check:doc-types	✅ Every documented component type is registered.
node scripts/check-control-bytes.mjs	✅ check-control-bytes: OK (scanned 5277 tracked text file(s); …)
node scripts/check-changeset-presence.mjs	✅ No source of a released package changed in this range …
node scripts/check-changeset-no-major.mjs	✅ No changeset declares a `major` bump.
node scripts/check-changeset-fixed.mjs	✅ All workspace packages are in the changeset fixed group.
pnpm run type-check:scripts	exit 0; --listFiles confirms doc-version-claims.test.ts is in the program (133 files from scripts/)
pnpm run lint:root	exit 0 — full run, 9s, 28 warnings, 0 errors, none in the changed file

The whole-tree, changeset, type-check and lint rows were measured on 37f6f5990; the follow-up commit 992bcec7f changes two lines of markdown prose and nothing else, and the gate, doc-link, doc-fence and doc-type rows above were re-run on it.

Declared narrowing, one check:pnpm run check:doc-snippets exits 2 = PRECONDITION NOT MET here ("the snippet program was NOT run"; it needs 21 packages built first) — so it is not measured locally, neither green nor red. Narrowed on the grounds that it compiles fencedts/tsx blocks and this diff touches none: the changed hunks are at line 12 of each page and the first fence delimiter in each file is at line 19. CI runs it in full regardless.

Scope notes

• PR fix(tooling): derive QUICK_REFERENCE's Node/pnpm floor as full version #6390 (card finding(tooling): the QUICK_REFERENCE Node-floor pin derives its floor with match(/(\d+)/), so it now REQUIRES the doc to state ≥ 22 against an engines.node of >=22.11 #6313) touches QUICK_REFERENCE.md and the quick-reference pin — a different gate from this card's version-literal ledger. Checked: no file overlap with this branch (.changeset/6307-bold-version-literals.md, the two guide pages, scripts/__tests__/doc-version-claims.test.ts), and the two jobs are kept separate.
• One out-of-scope defect found and filed unassigned as 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, deliberately not touched here: the neighbouring ci-cd-pipeline.md :: Node 22.x ledger entry's reason still says "the 14 node-version: 22.x declarations" (measured today: 26 of 27), and ci-cd-pipeline-doc.test.ts does not in fact pin that literal (grep -cP '\b22\b' on it returns 0; control, same file: ci.yml returns 28).
• Option A from the card — declaring per-package engines across the workspace — is closed by the triage ruling and was not done.

---

Generated by Claude Code