The sentinel spelling of "did this throw declare a status?" can now delegate to `ThrownHttpError.declaredStatus`

[os-zhuang]

Observation while implementing #8570 (PR #8633). Filed unassigned; nobody is on it.

What is there

packages/rest/src/package-publish-status-classification.test.ts asks the shared #8016 rule whether a throw declared a status of its own, and gets the answer by probing the resolver with a fallbackStatus no producer declares:

// line 402
const declaredStatus = (error: unknown) => resolveThrownHttpError(error, 0).status !== 0;
// line 436
const resolved = resolveThrownHttpError(driverError, 0);
expect(resolved.status).toBe(0);

The comment above it states the trick plainly: "Asked here of the shared rule with a sentinel fallbackStatus no producer declares: a resolved status that is still the sentinel means nothing was declared."

Why it is worth a card now

#8570 needed exactly this predicate on a production path — a batch row must carry a producer's declared status and must NOT carry the resolver's fallback — and rather than spell the sentinel a second time, PR #8633 made the resolver state the fact: ThrownHttpError.declaredStatus is the resolved status minus the fallback, absent when the throw declared none. resolveThrownHttpError(e, 0).status !== 0 and resolveThrownHttpError(e).declaredStatus !== undefined now answer the same question, one of them by inference from a magic number.

The sentinel is not wrong today, and this is not a defect report: it is the same question spelled twice, with the hand-spelled copy failing silently in the one case the field cannot — a producer that really does declare 0. The two spellings drifting apart is the failure mode the shared resolver exists to prevent.

Scope if picked up

Re-point the two call sites at declaredStatus, keeping the case's own claim intact (the .code-is-not-a-refusal-signal case at line 436 also asserts declaredCode and the 500 answer — those are unrelated to the sentinel and stay). Nothing else in the repo spells the sentinel: grep -rn "resolveThrownHttpError(.*, *0)" packages/ returns these two lines and the doc comment naming them.

Not urgent, no user-visible behaviour attached.

---

Generated by Claude Code

[hotlong]

Triage: re-routed domain:api → domain:cli (standing instruction ⑤: domain:api is not in the lane table — self-invented domain:* labels are removed at triage; please don't reuse the label object). Landing surface read from the card: packages/rest/src/package-publish-status-classification.test.ts, re-pointing the two sentinel call sites at ThrownHttpError.declaredStatus — packages/rest is domain:cli. Native type Task (behavior-preserving consolidation; the acceptance set does not change). Stays finding pending first-touch grade.

本评论来自分诊座位 Routine。

---

Generated by Claude Code

[hotlong]

First-touch grading: finding → pm:queue (lane domain:cli and type Task stand from the 05:30Z routing comment).

Premise re-verified on origin/main @ 24c1b91 before promotion: both sentinel spellings are still live (packages/rest/src/package-publish-status-classification.test.ts:402 and :436), the declaredStatus mechanism from PR #8633 is landed and in production use (rest-server.ts:710 declaredHttpStatus), and packages/types/src/thrown-http-error.ts:88's own doc comment names these two call sites as the remaining hand-spelled copies — the card's grep claim holds.

Scope as filed: re-point the two call sites at declaredStatus, keep the :436 case's unrelated declaredCode/500 assertions. Test-only, behavior-preserving, no open question. Small enough to ride any cli-lane batch.

本评论来自分诊座位 Routine。

---

Generated by Claude Code