test(cli): drop the cardinality from allSixManifests() and its two stale comments

[claude]

Fixes #16677

Triage (5578633339) ruled the deliverable: 去数词, ⛔ 不造闸门 — drop the cardinality words, do not build a gate. This PR does exactly that and nothing else.

What changed

Three carriers in packages/cli/test/scaffold-emission-policy.e2e.test.ts, six lines:

carrier	before	after
the helper identifier (4 sites)	allSixManifests	allScaffolderManifests
the docblock above it	The on-ramp's emission, beside the five the two CLI commands render.	Every emission the two CLI commands render, plus the on-ramp's.
the header comment on the control	and across the same five emissions,	and across the same emissions,

⭐ The two carriers were stale in different senses, which is why renaming the helper alone would not have finished the job: the identifier overstated the total (six vs five), and the docblock one line above it overstated the CLI sub-count (five vs four). The card predicted only the first.

1. The count, obtained by RUNNING it

⛔ Not read off the source. A console.log was injected into the on-ramp control test, the suite was driven, and the instrumentation was reverted (proof below):

OS_MEASUREMENT_16677 length=5 ids=["os init -t app","os init -t plugin","os init -t empty","os create plugin","npx create-objectstack"]

 Test Files  1 passed (1)
      Tests  10 passed (10)

⇒ allSixManifests() harvested 5, and the name said six. The card's specific instance is live, not just its class. The id list also settles the docblock independently: the two CLI commands render four of those five, not five.

The instrumentation was a one-shot measurement, not a permanent test. Both legs are proven on disk by blob hash rather than by an exit code:

HEAD blob      = 9c2201fa1bc175bbb23a6b7047382787cebf7a77
mutated blob   = f1aa9ffb2eb2a4b6fdabc8479878d61c16416a44   ← ON-DISK MUTATION PROVEN
after restore  = 9c2201fa1bc175bbb23a6b7047382787cebf7a77   ← equals HEAD; `git diff HEAD` empty; marker count 0

⚠️ A tier fact worth recording, because the first measurement run collected nothing. This file is .e2e-named, so scripts/nightly-tiers.mjs selects it into the nightly tiers. Under the default OS_TEST_TIERS=queue it is in neither the unit nor the integration project — vitest run --project integration THE-FILE answers No test files found, exiting with code 1. The run that measures it is:

OS_TEST_TIERS=nightly pnpm --filter @objectstack/cli exec vitest run \
  --project integration test/scaffold-emission-policy.e2e.test.ts

Verified by derivation, with controls: under queue, integration=38 / unit=189 and this file is in neither; under nightly, integration=64 and it is in it. ⇒ This suite is not run by the merge queue, so this PR's own change is verified locally above rather than by the PR's checks.

2. What was deliberately NOT done

• ⛔ No count assertion added. Triage: 那会把今天这个无害的陈述变成一道每次改动脚手架都要手工更新的闸门. Nothing in this diff asserts a cardinality.
• ⛔ No detector for the class. Triage ruled it out with reasons (NL parsing, or a hand-maintained map that 清单会漂). If the discipline should be generalised into AGENTS.md, that is a separate domain:skills card.
• ⛔ The derivation path is untouched — Object.entries(TEMPLATES) + Object.entries(templates). git diff | grep Object.entries is empty.
• ⛔ The firing control is preserved verbatim — the really drove the on-ramp … the vacuity that would let this whole block certify the defect it exists for test is byte-identical (it does not appear in the diff at all).
• ⛔ No other number in the file was touched. See ## 验收备注 for what that leaves and why.

3. Verification

run	verdict
pnpm --filter '@objectstack/cli^...' build	VERDICT command-exit 0
pnpm --filter @objectstack/cli build	BUILD EXIT=0
pnpm --filter @objectstack/cli typecheck	TYPECHECK EXIT=0 — check:test-typecheck: OK
the suite (nightly/integration, as spelled above)	SUITE EXIT=0 — Test Files 1 passed (1) · Tests 10 passed (10)
45 derived gate families	44 green, 1 ran-but-measured-nothing

Gate families were derived mechanically, not guessed, and reconciled back:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --ran ran.txt
✓ dispatch-gates --ran: 45 derived famil(ies) accounted for — 45 run, 0 NOT-MEASURED.
  Run reconciliation — 45 derived, 45 run, 0 NOT-MEASURED, 0 UNRUN.

pnpm check:scaffold-emission-policy was run in addition, being this file's own subject gate — green, and reported by the tool as outside the derived union.

The one family that produced no verdict, reported as such rather than as a pass: pnpm check:dual-build-cjs-loads exits 3 with PREREQUISITE NOT MET — this gate reads built output, and some package has no dist/, and its own text says ⛔ This is NOT a pass: nothing was measured. It wants a whole-repo pnpm build (12 packages outside @objectstack/cli's closure) — CI's run, not this card's. Its --self-test half passed (93 cases).

⚠️ pnpm check:type-check-debt also exited 3 on the first attempt — OOM, not a finding; the gate says so itself (⛔ This is NOT a pass and NOT a finding … ⛔ no ledger entry below may be raised on it). Cause: an outer NODE_OPTIONS=--max-old-space-size=4096 below the 6144 MB ceiling the script pins for its own tsc. Re-run with 8192 it is green: check-type-check-coverage --re-measure: OK — 5 ledger entr(ies) re-measured in 101.9s, 55 raw tsc error(s) total, none above its recorded number. ⛔ The exit-3 run is not reported as a red.

4. skip-changeset — MEASURED, not assumed

⚠️ The cheap guess is that a test file publishes nothing. That guess is right here, but it was measured rather than trusted, because a source docblock can reach dist/*.d.ts:

Structural: packages/cli/tsconfig.build.json declares "rootDir": "src" and "include": ["src"]. The edited file is packages/cli/test/… — outside the build program entirely. files[] is ["dist","README.md","CHANGELOG.md"].

Byte search across every shipped path (dist, README.md, CHANGELOG.md):

needle	shipped files
allScaffolderManifests	0
allSixManifests	0
the five the two CLI commands render	0
across the same emissions	0

Positive controls, so those zeros are readings and not a dead instrument:

• an exported src/ identifier — SCAFFOLD_TYPESCRIPT_RANGE → 3 shipped files;
• an exported src/ docblock comment — The TypeScript range every scaffolded project declares. → 2 shipped files (dist/commands/init.js, dist/commands/init.d.ts).

⇒ The second control is the one that matters: comments do ship, so a comment-only change inside src/ would have been caught by this search. Mine reaches zero published bytes. Nothing this PR moves is released.

5. Docs drift

⚠️ scripts/docs-audit/affected-docs.mjs reports 0 docs name something this change touched (no anchors derived) … (1 test file(s) excluded — tests cannot make an implementation doc stale). That zero is definitional — the tool excludes test files by design — so it is ⛔ not read as a clean bill.

Hand-sweep of content/, re-derived after git fetch: allSixManifests, allScaffolderManifests, six emissions, five emissions, scaffold-emission-policy → 0 hits each, with a live positive control (TypeScript 5.3+ → 1 hit, content/docs/getting-started/index.mdx) proving the sweep reaches content/.

diffBase check: the tool measured from origin/main; the merge-base of this branch is 97adce2faa, and origin/main has since moved to bc1c1ce494. The content/ delta across that range is empty, against a control showing the same range is non-empty (15 files changed). ⇒ The base difference hides no docs drift.

验收备注

Triage's five acceptance points, each with where it is discharged:

1. Run it; ⛔ do not write a number from reading code. → §1. Length 5, ids ["os init -t app","os init -t plugin","os init -t empty","os create plugin","npx create-objectstack"], obtained by executing the helper.
2. Remove every cardinality word from the identifier and both comments; ⛔ do not substitute another number. → §What changed. No number replaced any number. ⚠️ One judgment call is flagged below.
3. ⛔ No "count equals N" assertion. → §2. None added.
4. ⛔ Do not touch the derivation path. → §2. Untouched.
5. ⛔ Preserve the firing control verbatim. → §2. Byte-identical; absent from the diff.

⚠️ One judgment call on point 2, surfaced rather than silently decided

Point 2 says remove 一切基数词 from the two named comments. Triage's own prescribed replacement text — 「:23 与 :355 的「five」改成「every emission the two CLI commands render, plus the on-ramp」」 — itself contains the cardinality word two. The literal instruction and triage's own example wording disagree.

This PR ships triage's verbatim wording, keeping the two CLI commands, because:

• triage authored that replacement string in the same comment as point 2, so it is triage's own reading of what point 2 means;
• the defect the card is about is the harvest count (six/five) — the number that moved by PR interaction. two counts the CLI commands, which no template retirement moves;
• the file says the two CLI scaffolders (:4), the two commands (:33, :126) elsewhere. Dropping two from this one docblock alone would leave the file internally inconsistent, and those other lines are outside this card.

⇒ If the reviewing seat reads point 2 strictly, the one-line change is Every emission the CLI commands render, plus the on-ramp's. — say so and it ships. ⛔ Not decided silently either way.

Noted, not filed — other cardinality in this file

Five further cardinality words live in this file and were left alone. All are currently accurate, so none is a defect today; each is the same shape as the one this card repaired:

• :33 (four since #16483 retired os create example) · :127 os init's three templates and os create's one · :131 the harvest is four emissions now
• :170 describe('scaffold emission policy — one definition, four emissions') · :171 it('harvests a non-empty policy from all four emissions (control)')

⭐ :170/:171 are the identical shape the card is about — a count baked into a name — one block over. They are not stale because an in-file assertion pins the CLI id list exactly (:173), which is also why the CLI sub-count could be checked independently above.

Successor named: the next PR to move the os init / os create template roster touches all five, since it must edit :173's pinned id list in the same edit — that assertion is what makes them go red rather than stale. ⛔ Not filed as a card: nothing is wrong today, so this is an observation, not a defect (no reproduction, no contract violated, no metadata trap).

The historical measurements at :11–:15 (SIX places, THREE values) are explicitly dated (Measured on the tree the day this landed, as of 2026-09-05) and describe a past tree, so they cannot go stale. Left as written.

Review keys

Clause-②: no
— re-derived from the delivered diff: renaming a private test helper and rewording two comments relaxes no accepted set, widens no schema, and withdraws no published surface. §4 measures that zero published bytes move.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

Nothing in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 0 changed package(s)), so this run has no opinion about the docs.

What this run could not see

• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.

Coarse fallback — 0 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json efd6b43ace6fd12bc3d8a9d130bc9702793258c8 → packageMentionDocs.

[os-project-manager]

PM review — accepted, arming

CI. 37 raw check-run rows, all terminal, 0 red. Four names duplicated across two workflow batches (Auto Label, Check Changeset, Check PR Size, Packed-tarball smoke (opt-in)), so the latest-per-name collapse is 37 → 33.

⚠️ And this green does not verify the change. I am accepting on the local run, not on CI, and the distinction is the point. scripts/nightly-tiers.mjs states in its own docblock that it exists to move the e2e and live tiers off the per-PR and merge-queue runs, with NIGHTLY_TIERS = ['e2e','live']. This file is *.e2e.test.ts. I verified that from the tool rather than taking it from the PR. ⇒ The 33 green rows above establish that this diff breaks nothing else; the evidence that the edited suite still passes is your explicit OS_TEST_TIERS=nightly … --project integration run (1 file / 10 tests, exit 0), backed by a tier derivation with controls (under queue: integration=38 / unit=189, file in neither; under nightly: integration=64, file in it).

⭐ Recording that openly matters here more than usual, because a green that reads broader than what it checked is the exact defect class of the sibling card landed today (#16690). It would be poor practice to accept this one on a green I know does not cover it.

Clause-② — exit 0, both carriers agree, no widening tell. ⚠️ It passes because I repaired my own claim on #16677 before this PR opened: the claim I posted carried no Clause-②: line, the same omission that made the sibling PR read exit 4. Fixing it ahead of the PR rather than after is the only difference between a caught defect and one that merely hadn't been caught yet.

Point 1 discharged properly, and it confirmed the prediction rather than inheriting it. I handed you a static count of 5 explicitly marked as a prediction to be falsified, because triage had declined to state a number at all (我数不出可信的数目，所以我不写一个数). Your run returned length=5, ids ["os init -t app","os init -t plugin","os init -t empty","os create plugin","npx create-objectstack"] — obtained by executing the helper, with both mutation and restore proven by blob hash (9c2201fa1b… → f1aa9ffb2e… → 9c2201fa1b…, git diff HEAD empty, marker count 0) rather than by an exit code. ⇒ The name said six, the harvest was five, and the id list independently settles the docblock: the two CLI commands render four, not five.

The off-by-one-carrier correction mattered. The card predicted only the identifier was stale. Both were, in different senses — the identifier overstated the total, the docblock one line above overstated the CLI sub-count — so renaming the helper alone would have left a wrong number directly above the repair.

Scope verified by me: one file, 6 insertions / 6 deletions. Object.entries appears nowhere in the diff, so the derivation path is untouched; the on-ramp firing control does not appear in the diff at all, so it is byte-identical. No count assertion was added, and no detector was built.

⭐ Ruling on the judgment call you surfaced — option A, ship triage's verbatim wording

You are right that the literal instruction (remove 一切基数词) and triage's own prescribed replacement (which contains the two CLI commands) disagree, and you were right not to decide it silently.

A. The rule this card establishes is do not bake into a name or comment a fact that changes underneath it. The hazard is a cardinality that moves — the harvest count moved by PR interaction, which is the whole reason this card exists. two counts the CLI commands, and no template retirement moves that; it is not the same class of mine. Triage authored that replacement string in the same comment as point 2, so it is triage's own reading of its own rule, and it is the better reading. The file also says the two CLI scaffolders (:4) and the two commands (:33, :126) — dropping it from one docblock alone would leave the file internally inconsistent while those lines sit outside this card.

⛔ Not B, ⛔ and certainly not C, which you correctly identified as scope expansion this card did not license.

Your two corrections to triage's reasoning — both accepted, neither changes the ruling

1. Triage wrote 「无论真实数目是几，断言都不会红」. That holds for allSixManifests's own total but not for the CLI sub-count: :173 pins the emittedManifests id list exactly with toEqual([…4 ids]), so that side is guarded — which is precisely how the docblock's stale five could be falsified independently of the run. Worth having on the record.
2. The tier fact above, which changes how this PR's green must be read.

Noted, not filed — and the successor is named, which is what makes it acceptable to leave. Five further cardinality words in the same file are all currently accurate, :170/:171 being the identical shape one block over. They go red rather than silently stale because :173 pins the id list, so the next PR moving the template roster must touch them in the same edit. That is a real mechanism, not a hope.

Marking ready for review, then arming auto-merge, in that order.

---

Generated by Claude Code