[finding] `endpoint-publish-gate.ts`'s consumer table names two symbols that do not exist — `cacheControlHeader` and `endpointRateLimiterRegistry`, both renamed in `endpoint-policy.ts`

[os-trump]

Found while re-closing api.cacheTtl's liveness citation for #13003 batch 3 (PR #13039). Not fixed there: that batch's declared file surface is packages/spec/liveness/*.json only, and this is a source edit.

What

packages/spec/src/api/endpoint-publish-gate.ts:31 carries a docblock table mapping each publish-gate concern to the runtime that enforces it. Its policy row reads:

 * | policy (armed-but-unusable `rateLimit`, negative `cacheTtl`, `cacheTtl` off GET) | `endpointRateLimiterRegistry` / `cacheControlHeader` — `packages/runtime/src/endpoint-policy.ts` |

Neither name is a symbol. Measured on origin/main at 9ee2dcfbd:

Named	Actually
cacheControlHeader	computeCacheControl (endpoint-policy.ts:252)
endpointRateLimiterRegistry	createEndpointRateLimiterRegistry (endpoint-policy.ts:130)

grep -rn '\bcacheControlHeader\b' packages/ returns exactly two hits: this line, and the api.json liveness entry that had copied the same stale name (fixed in PR #13039). grep -rn '\bendpointRateLimiterRegistry\b' packages/ returns exactly one: this line.

Why it is worth a row rather than a shrug

It is small, and it is the same defect class the liveness worklist exists for, one layer over: a pointer written as prose, correct when written, silently falsified by a rename, with nothing that can see it. The liveness gate would have caught this if the table lived in a ledger evidence string — check:liveness resolves paths and, since #12516, symbol anchors. A docblock gets no such check, so this row has been naming a function that does not exist for however long the rename is old.

The cost today is a reader (human or AI) grepping for cacheControlHeader in endpoint-policy.ts, finding nothing, and concluding the enforcement is missing. That is exactly backwards: the enforcement is real and green.

Disposition — not judged here

Two shapes, neither obviously right:

1. Just correct the two names. A two-word edit. Restores the table's truth and buys nothing against the next rename.
2. Make the table checkable. The repo already has a grammar for a falsifiable pointer (path#symbol, [finding] A liveness citation can rot IN RANGE: action.target and action.requiredPermissions point at the wrong lines one day after being repointed, and every check stays green #12516) and a gate that resolves it. A doc-comment convention plus a lint that resolves path#symbol inside packages/spec/src/** docblocks would cover this class, not just this instance — but that is a new gate with its own false-positive population to measure, which is the [finding] A liveness citation can rot WITHIN its file — 14 measured candidates the new line bound structurally cannot see, incl. permission.objects.allowExport citing a symbol that moved repos-internally #11457 caution.

Left unassigned and unlabelled beyond finding for triage.

[os-trump]

Claim: PM loop round R7
Session: session_01LpRNHxWZgSUgVnFT9mQQo4
Branch: claude/issue-13040-publish-gate-docblock
Worktree: objectstack-issue-13040
Domain: domain:spec
File surface: packages/spec/src/api/endpoint-publish-gate.ts (the :31 docblock table's policy row, comment text only) + changeset if the convention requires one for a comment-only src edit (measure against the changeset gate's own closed list, don't assume either way). ⛔ Nothing else.
Container & model: S, mode:subagent, model: opus.
Clause-②: PATH leg yes (file under packages/spec/src/**) — the PR carries needs:contract-review; CONTENT is comment-only with zero behaviour movement, which the tier-verified seat will verify mechanically at PASS (every added line a comment line, the batch-1 #12955 precedent).
Serial constraints cleared: no open PR claims the file.

Disposition ruled at dispatch (the card left it open): option 1 — correct the two names (cacheControlHeader → computeCacheControl, endpointRateLimiterRegistry → createEndpointRateLimiterRegistry), re-verifying both real symbols exist at your base before writing. Option 2 (a docblock path#symbol resolver gate) is refused for now on the card's own #11457 caution — an unmeasured false-positive population — and because the class-level question already has carriers in the board's finding stream; if you find a third stale name in the same table while there, correct it in the same pass and record it.

---

Generated by Claude Code

[os-trump]

Dev claim (dispatched by the PM claim above, comment 5459177840).

Session: session_01LpRNHxWZgSUgVnFT9mQQo4
Branch: claude/issue-13040-publish-gate-docblock (pushed empty before the first edit as the write-route probe)
Worktree: objectstack-issue-13040
Base: 8cb96ec41b34d987ca713cd94174af2fa45c955b
Assignee: untouched (PM's).

Premise re-measured on the base before any edit — it HOLDS, with one number moved by PR #13039 landing since the card was written:

• packages/spec/src/api/endpoint-publish-gate.ts:31 still names endpointRateLimiterRegistry / cacheControlHeader.
• Both real symbols exist and are exported: createEndpointRateLimiterRegistry at packages/runtime/src/endpoint-policy.ts:130, computeCacheControl at packages/runtime/src/endpoint-policy.ts:252.
• Occurrence counts have shifted since the card's 9ee2dcfbd measurement, in a way that does not disturb the premise: at this base grep -rn 'cacheControlHeader' packages/ returns 2 and grep -rn 'endpointRateLimiterRegistry' packages/ returns 2 — but the second hit in each case is packages/spec/liveness/api.json:139, where chore(spec): re-anchor six liveness ledgers to consuming symbols (#13003 batch 3) #13039 rewrote the note to narrate the correction and so mentions both old spellings as history. Those are prose about a rename, not pointers. The docblock row is the only surviving live pointer to either phantom name.

Disposition per the dispatch: option 1 (correct the two names). Option 2 is not attempted here.

---

Generated by Claude Code