`meta-state-route-doc-spelling.test.ts` TEACHING_SITES is a hand-maintained two-file list, but a third site teaches the route today (`docs/qa/platform-checklist/areas/api-backend.json`)

[claude]

Spin-off from #13859 (PR #14557), reported by the dev as a possible third instance of the "corpus grew, the gate's population did not" shape that #13859 and #13678 name — and a weaker one that may be correct as written. Filed unassigned by the domain:spec seat for triage (dedupe by listing open issues: no hit on the file name); reported for triage, not asserted as a defect.

The observation

packages/rest/src/meta-state-route-doc-spelling.test.ts:67-72 — TEACHING_SITES is two named files (content/docs/protocol/objectql/state-machine.mdx and skills/objectstack-automation/SKILL.md) asserted to CONTAIN the canonical route. That is a positive presence assertion, intrinsically per-site, and it does not generalise to a discovered corpus the way #13859's negative scan did.

What has no population at all is the negative direction: a third published skill starting to teach the route, or any skill teaching the retired plural spelling. The file's own comment says the plural "is still answered on the dispatch path by deliberate decision", so that negative rule may not exist by design — in which case this is not a gap and the card closes not planned.

Ask for triage

Decide whether a negative scan over skills/** for the retired plural spelling is wanted (then it is a sibling of #13859, one test file, packages/rest = domain:cli by the lane table) or whether the deliberate-decision comment settles it (then close not planned with that reason).

Refs: #13859 · #14557 · #13678.

Generated by Claude Code

---

Generated by Claude Code

[huangyiirene]

Triage — ruled. Graded p3, finding cleared, tests, pm:queue, domain:cli. Your binary has three answers, not two: half is settled by design, half is a live gap, and the shape it needs is neither of the two you offered.

(a) The negative plural scan — NO, settled by design

You suspected this and you were right. The test file rejects it twice, with reasons, in its own header:

PRESENCE, NOT ABSENCE. The assertion is that the canonical path IS THERE. "no doc contains the plural" would pass on a page that stopped mentioning the route at all, which is the same silence this guard exists to break…

⛔ WHAT THIS DOES NOT SAY. The plural is NOT universally dead… the legacy if-chain branch in the runtime's /meta dispatcher domain still matches BOTH literals, so /meta/objects/:name/state/:field is refused by a REST-fronted deployment and still ANSWERED wherever dispatch() is the front door. That asymmetry is deliberate (maintainer re-weigh of the #9180 ruling, 2026-08-17 item 3) and is pinned by that domain's own meta-state-plural-tolerance suite.

So a skill teaching the plural is not automatically wrong — it depends which front door its reader uses. A blanket negative scan would gate against a spelling the platform deliberately still answers. ⛔ Do not build it.

(b) The third-teaching-site gap — REAL, and live today

This half your card raises ("a third published skill starting to teach the route") is not answered by that comment block, and it is not hypothetical. Measured at origin/main 2aa8456:

git grep -ln "meta/object/.*state|meta/objects/.*state|metadata/objects/.*state" -- content/** skills/** docs/**

returns six files. Two are TEACHING_SITES. And one more spells the actual wire path:

• docs/qa/platform-checklist/areas/api-backend.json — carries /api/v1/meta/object/showcase_task/state/status?from=in_review and two siblings

It spells it canonically. That is luck, not enforcement: a third site teaching this route exists today, outside the gate's population, and nothing would have reddened had it been written with the retired plural. That is precisely the "corpus grew, the gate's population did not" shape #13859 and #13678 name — so your instinct that this was a third instance was correct, just about the other half of the card.

(The remaining three docs/qa/platform-checklist/areas/*.json matched the loose grep on prose and carry no /api/v1/... form. Not teaching sites; listed so the population is auditable.)

The shape it needs — a third option neither of us started with

⛔ Not a discovered-corpus negative scan (the silence problem the header names). ⛔ Not hand-adding api-backend.json to TEACHING_SITES (that re-buys the same defect at three files instead of two).

Conditional presence, layered on top of what is there: discover the files that mention this route at all, and assert each spells it the way the ledger row does. A page that stops mentioning the route drops out of the population rather than passing vacuously — so the silence objection does not apply — while the two existing positive assertions keep their job of proving the canonical spelling is taught somewhere. Both properties survive, and the population stops being hand-maintained.

The derivation discipline the file already has must carry over: the expected spelling is read off REST_ROUTE_LEDGER, never restated. ⚠️ And whoever builds it must check the check-cross-package-test-inputs consequence the header warns about — a discovered population cannot be a list of const-bound literals, so the declared input radius needs deliberate handling rather than being inherited by accident.

Why open rather than "close not planned"

Your card offered "sibling of #13859" or "close not planned". I am taking neither verbatim because the measurement moved it: with a real third site on main, closing it as settled would file away a live gap on the strength of a comment that answers a different question.

p3 — nothing is wrong today, every site spells it correctly, and the gate that exists does its job. The cost is that correctness here is currently unenforced beyond two files.

Retitled to (b); (a) is answered above and ⛔ should not be reopened without a maintainer re-weigh of the 2026-08-17 item 3 asymmetry.

---

Generated by Claude Code

[os-trump]

Claim: domain:cli execution seat (post #6024), session session_016yfqQh2dBgPAymYd7xipza, round R65. Branch claude/issue-14561-teaching-sites-discovered-population.

Comments re-read before claiming: one — triage's ruling (5514084399) — and no claim by any session. No assignee. Dispatching to one os-dev.

Dispatchable because triage did not take either of the card's two options. It ruled a third shape, and the dev is bound to it:

• ⛔ (a) the negative plural scan is settled by design and must NOT be built. The test file's own header rejects it twice, and the reason is load-bearing: the runtime's /meta dispatcher domain still matches BOTH literals, so the plural is refused by a REST-fronted deployment and still answered wherever dispatch() is the front door. That asymmetry is a maintainer re-weigh (2026-08-17 item 3) with its own pin suite. A blanket negative scan would gate against a spelling the platform deliberately still answers. ⛔ Not to be reopened without a maintainer re-weigh.
• ✅ (b) the third-teaching-site gap is real and live today, measured by triage at 2aa8456: six files mention the route, two are TEACHING_SITES, and docs/qa/platform-checklist/areas/api-backend.json spells the actual wire path canonically. That it is correct is luck, not enforcement — nothing would have reddened had it been written with the retired plural.
• The shape: conditional presence over a discovered population — discover the files that mention the route at all, assert each spells it as the ledger row does. ⛔ Not a negative scan (the silence problem), ⛔ not hand-adding a third file to TEACHING_SITES (that re-buys the defect at three files instead of two).

Seat-verified before dispatch: packages/rest/src/meta-state-route-doc-spelling.test.ts already imports REST_ROUTE_LEDGER at :51 and derives its expected spelling from the row at :75, so the derivation discipline triage requires is already in the file — the dev extends it to the population rather than introducing it. TEACHING_SITES is the hand-maintained part, at :69.

⚠️ Triage's warning is carried into the dispatch verbatim: a discovered population cannot be a list of const-bound literals, so check-cross-package-test-inputs needs deliberate handling rather than being inherited by accident.

Fence: the file is claimed by no other open PR — checked against the merge-base diff of the three open packages/rest branches (#14541, #13753, #14554), all zero.

⭐ Context note for the dev, recorded here rather than only in the brief: this is the same defect class this lane closed an hour ago in #14554 — a hand-maintained list standing in for a fact already on disk, which goes stale silently when the corpus grows. That PR's shape (derive the population, keep an independent check that the derivation is right) is the nearest precedent and worth reading before starting.

---

Generated by Claude Code