spec: a single-word language name ("Chinese", "English", "Japanese") still passes as a translation locale — the #25822 fix refuses multi-word names only, though its changeset and refusal text say "a language name" is refused

Filing gate: ① a product defect with a live reproduction. It is the residual of closed #25822, found by the retest of that fix (PR #25983).
Business reason: an author who writes locale: 'Chinese', a plausible spelling for a human or an AI author, gets a green save. Their strings are stored under a locale the console's language picker never selects (zh-CN), so they never render. This is the same silent "not translated" failure #25822 closed for zh_CN, and the changeset says this case is closed too.
Reader: the packages/spec lane, at src/system/translation.zod.ts (LOCALE_TAG_SHAPE, LocaleSchema, LOCALE_TAG_REFUSAL) and SYS_USER_LOCALE_TAG_PATTERN in @objectstack/platform-objects/identity, which shares the shape.
Dedupe (open AND closed): searched "LocaleSchema language name accepted English Chinese locale tag shape". The results are #25822 (the closed parent) and #26128 (open: a translation item named zh-CN is refused by the item-name grammar, which is a different defect).
reach: public HTTP entry, and the defineTranslation / defineTranslationBundle authoring functions. Measured on a fresh stock showcase boot at 756220c4 as admin, and against the built packages/spec/dist by both the runner and the coordinator.
Checklist item priority: P1
type: Bug
QA-source: #26456 · i18n.strict-translation-key-rejection · observed beside the clauses

Reproduction

As admin, PUT /api/v1/meta/translation/qa_bad_locale with body {"name":"qa_bad_locale","locale":"Chinese","objects":{"showcase_task":{"label":"x"}}} → 200. The read-back serves locale:"Chinese" with _diagnostics.valid:true.
With the built spec, defineTranslation({name:'qa_x', locale:'Chinese', objects:{t:{label:'x'}}}) (from /system) is accepted, and so is defineTranslationBundle({Chinese:{…}}). English and Japanese behave the same.
Controls refused as designed: not a locale!!, zh_CN, 'zh-CN ', ' zh-CN' and Simplified Chinese → 422 invalid_format. zh-CN is accepted.

Expected, per .changeset/25822-locale-tag-shape.md ("a language name is refused"; "replace a language name with its tag") and LOCALE_TAG_REFUSAL ("… a language name …"): 422 invalid_format at locale.
Actual: accepted.

Mechanism

LOCALE_TAG_SHAPE = /^[A-Za-z]{2,8}(?:-[A-Za-z0-9]{1,8})*$/ (translation.zod.ts:31). A 2–8-letter primary subtag admits any single English word of that length. The regression test (translation.test.ts, around line 56) pins only the multi-word 'Chinese (Simplified)'.

The source comment records "shape only, by decision", because there is no registry to check membership against. So the fix is a choice for the spec lane:

tighten the primary subtag to the BCP-47 forms that occur in practice (2–3 letters), or
correct the changeset, refusal text and docs so they stop promising that a language name is refused.

Either way, SYS_USER_LOCALE_TAG_PATTERN shares the shape and should move with it.

Generated by Claude Code

Triage: first grade — `bug` · `priority:p2` · `domain:spec` · `area:i18n` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-27T04:57Z

⛔ Not a claim, ⛔ not a dispatch.

p2, a protocol narrowing handled directly in 18.x: LocaleSchema still accepts a single-word language name (Chinese), which the spec: a translation item accepts any locale string ("not a locale!!" saves 200, valid) — LocaleSchema is a bare z.string() #25822 changeset says is refused. Refuse it with the same refusal text, in translation.zod.ts and in the shared SYS_USER_LOCALE_TAG_PATTERN. Clause-②: yes (narrowing), with an ADR-0087 entry. Skills-docs: is owed ([Decision] pm-dispatch: which PRs owe a Skills-docs: line (the #26184 audit Part B rule draft) — protocol-kind only, protocol-kind plus narrowing, or not adopted #26369 B). Found by the 18.2.0 wave-2 sweep ([sweep] v18.2.0 release test sweep #25807).

Generated by Claude Code

Premise check and wait · domain:spec seat 3 · os-support-ai · session session_01WGufzgbJhFoXYsLSe5WSPP · 2026-10-27T05:19Z · ⛔ not a claim, no label change

Premise: holds (origin/main c91544a4). LOCALE_TAG_SHAPE at packages/spec/src/system/translation.zod.ts:31 is still /^[A-Za-z]{2,8}(?:-[A-Za-z0-9]{1,8})*$/, so a 2–8 letter word such as Chinese passes. The docblock above it (lines 16–30) and LOCALE_TAG_REFUSAL both say a language name is refused. SYS_USER_LOCALE_TAG_PATTERN (packages/platform-objects/src/identity/sys-user.object.ts:33) carries the same string. The refusal row in translation.test.ts:56 pins only 'Chinese (Simplified)'. The triage route (refuse it, Clause-②: yes (narrowing), an ADR-0087 entry, Skills-docs: owed) stands.

Waiting on: #25785 (PR #26159, in flight on domain:spec seat 2). Both cards are area:i18n, and their file surfaces intersect on packages/spec/src/system/translation.zod.ts. So the one-in-flight-per-axis rule holds this card until PR #26159 lands, the same as #26128. Spec seat 1 left this card to spec seat 3's i18n lane, and seat 3 dispatches it when #26159 merges, after judging it by file surface against #26128 (whose edits sit in the TranslationItemSchema region of the same file).

Generated by Claude Code

Claim: PM loop round 2 · 2026-10-27T13:18Z
Session: session_01WGufzgbJhFoXYsLSe5WSPP
Account: os-support-ai
Branch: claude/project-thread-zxzeeg (card thread; the platform assigns the thread's branch and it can push only that one, written as measured instead of claude/issue-26463-<slug>)
Worktree: objectstack-issue-26463
Domain: domain:spec
Seat: domain:spec#3
File surface: packages/spec/src/system/translation.zod.ts (LOCALE_TAG_SHAPE / LOCALE_TAG_REFUSAL / LocaleSchema region and its docblock only), packages/platform-objects/src/identity/sys-user.object.ts (SYS_USER_LOCALE_TAG_PATTERN line and docblock only), packages/services/service-messaging/src/recipient-locale.ts (its compiled copy of the shape) and recipient-locale-shape-parity.test.ts, packages/plugins/plugin-auth/src/sys-user-locale-write-contract.test.ts (the shape-not-membership pin), translation.test.ts refusal rows, generated output, one changeset with the ADR-0087 entry (stop on breach; explain in the report)
Container & model: S, mode:thread, model: TIER_DEFAULT — --tier: no path-derived mandate (schema face, narrowing; one CONTRACT_REVIEW_TIER review owed before the queue, run by the seat), effort: medium
Clause-②: no (narrowing)
Metadata-protocol: narrowed
Responsibility: n/a — the published refusal text and the #25822 changeset already promise this refusal; the shape fails to deliver it
Thread-read: 6434121727
Premise: holds — re-read on origin/main 43731031. Seat reading on the two pins the thread measured as asserting the opposite: sys-user-locale-write-contract.test.ts:205-215 pins Japanese as accepted, and recipient-locale-shape-parity.test.ts:72-82 pins null as shape-legal and asks readers not to tighten the shared regex. The governing text is the maintainer ruling of 2026-09-03 (option B) the plugin-auth file quotes: "a malformed value is refused loudly by the column's BCP-47 shape check". It names a SHAPE check and fixes no subtag length. A 2–3 letter primary language subtag is still shape, not membership (xx stays accepted, unknown tags still fall to the delivery ladder), so the ruling is unchanged and the two pins are a test author's reading of the old 2–8 shape. They move with this card: Japanese and null become shape-refused on both sides, the parity pin keeps the two spellings byte-identical, and NOTHING_LITERALS stays as a read-side floor. Clause-② is no (narrowing) per the seat rule for narrowing a published accept set (the dispatch order's yes (narrowing) is corrected here; precedent #25983).
Serial constraints cleared: #25785 (PR #26159) landed 11:56Z; siblings #26112+#26142, #26170, #26128 edit other regions of translation.zod.ts (region-disjoint; merge main after each lands, generated files via os-regen-merge.sh); no open PR touches sys-user.object.ts, recipient-locale.ts or the plugin-auth pin

os-dev-report

{
  "issue": 26463,
  "status": "done",
  "branch": "claude/project-thread-zxzeeg",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/26770",
  "session": "session_01WGufziYkxRFm6oPR7CmuDg",
  "premise_still_valid": true,
  "summary": "The locale shape's primary language subtag goes from 2-8 to 2-3 letters, so Chinese, English, Japanese (and null) are refused, while en, zh-CN, zh-Hans-CN, es-419, yue-HK and case folding stay accepted. The one string moves in LocaleSchema (spec translation.zod.ts, docblock and refusal text re-read and made true), SYS_USER_LOCALE_TAG_PATTERN (platform-objects) and service-messaging's read-side LOCALE_TAG_SHAPE, still byte-identical; the two opposite pins (plugin-auth Japanese-accepted, service-messaging null-shape-legal) are rewritten per the seat's ruling, and the 18.x D3 entry translation-locale-non-tag-refused now says 2-3 letters. Not moved, with reasons in the PR body: the plugin-hono-server fixture rule (tests the endpoint evaluates any declared rule; outside the claim surface), ui i18n provider no-Intl fallback (its Intl branch accepts Chinese, so narrowing the fallback alone splits the two branches) and ui doc-draft add-locale hygiene (a doc translations map keyed by any string; outside the claim surface). MEASURE FIRST: zero examples, fixtures, seeds or bundles carry a locale with a 4-8 letter primary subtag (the one hit was the plugin-auth pin itself).",
  "tests": "Pre-merge head ee3058c2: pnpm --filter @objectstack/spec test 731 files, 21928 passed, 1 todo; service-messaging test 61 files 732 passed; plugin-auth test 198 files 3780 passed 10 skipped; platform-objects test 96 files 1361 passed; plugin-hono-server current-user-endpoints-localization.test.ts 13 passed. typecheck exit 0 for spec, platform-objects, service-messaging, plugin-auth (plugin-auth after full build). Merged head 84506a0c: translation.test.ts 189 passed, platform-objects 1361 passed, recipient-locale-shape-parity 19 passed, sys-user-locale-write-contract 24 passed; pnpm build 116/116; spec check:generated all 14 up to date. Ablation via scripts/ablation-replace.mjs (spec pattern back to 2-8, anchor 1->0, blob fa4a2333->1108375e): translation.test.ts 6 failed / 183 passed (Chinese, English, Japanese, chinese-CN, null, the door test); restored blob == HEAD, git diff HEAD empty; test imports src, no build leg. test:repo not run locally (no edited test in it, not derived), declared to CI.",
  "gates": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack: 87 families on 84506a0c, each run and recorded CMD :: exit N, all 87 exit 0; --ran: 87 derived, 87 run, 0 NOT-MEASURED, 0 UNRUN. First pass on ee3058c2/32f73697: check-adr-0087-registration exit 1 (marker spelled bare already-registered; respelled not-required (already-registered translation-locale-non-tag-refused), exit 0) and dual-build-cjs-loads, check:i18n, check:i18n-walk-parity exit 3 before a full build (exit 0 after). CI check-runs on #26770: in_progress at report time.",
  "dogfood_battery": "84506a0c · os-verify-lock: VERDICT command-exit 0 · Test Files 302 passed | 3 skipped (305) · Tests 2505 passed | 20 skipped (2525) · no failures (same counts on pre-merge head 32f73697)",
  "mcp_calls": "4 — mcp__github__issue_read x4 (get #26463, get_comments #26463 x2, get #26369); no write tool",
  "api_writes": "4 — via fleet-write relay: pr_create (POST /repos/objectstack-ai/objectstack/pulls draft + POST /issues/26770/assignees os-support-ai), pr_request_reviewers (POST /pulls/26770/requested_reviewers os-support-ai, the harness's PR-attribution rule), comment (POST /issues/26463/comments, this report); plus git push x3 (not REST)",
  "open_questions": [],
  "out_of_scope_findings": [
    "carrier: none · noted, not filed — ui/packages/i18n/src/provider.tsx:224 isWellFormedLanguageTag accepts a language name through Intl.getCanonicalLocales (and its 2-8 fallback); reads a UI language preference, not stored metadata, no measured wrong answer at a public door",
    "carrier: none · noted, not filed — ui/packages/app-shell/src/views/metadata-admin/previews/doc-draft.ts:28 LOCALE_TAG keeps the 2-8 shape for the doc add-locale box; the spec keys a doc's translations by any string, so no contract is violated"
  ],
  "deviations": [
    "Order said Clause-②: yes (narrowing); re-judged and the seat agreed: no (narrowing), as the claim carries.",
    "main gained 6 unrelated commits after the pre-PR merge (none touch the 8 files); not re-merged to keep the dogfood reading on the final head; CI judges the merge ref.",
    "Test comments cite #26463 at the seat's request."
  ]
}

Generated by Claude Code

ACCEPT → PR #26770 comment 6448177860 · domain:spec seat 3 · 2026-10-27T16:59Z · contract review PASS 6446067343 (head 84506a0c; current head 25fe0e3f adds a changeset-only correction and a main merge); ready with auto-merge armed.

Landed · domain:spec seat 3 · os-support-ai · 2026-10-27T18:29Z

reading	value
PR #26770	merged through the queue at 2026-10-27T17:25Z, merge commit `63e38732aa`
`origin/main`	`d8b943b6` contains the merge commit (`git merge-base --is-ancestor`)

State: Fixes #26463 closed this card as completed. pm:dispatched comes off in this stroke; domain:spec, area:i18n and the grade stay.

Generated by Claude Code

Navigation Menu

Repository navigation

spec: a single-word language name ("Chinese", "English", "Japanese") still passes as a translation `locale` — the #25822 fix refuses multi-word names only, though its changeset and refusal text say "a language name" is refused #26463

Reproduction

Mechanism

Triage: first grade — `bug` · `priority:p2` · `domain:spec` · `area:i18n` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-27T04:57Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

spec: a single-word language name ("Chinese", "English", "Japanese") still passes as a translation locale — the #25822 fix refuses multi-word names only, though its changeset and refusal text say "a language name" is refused #26463

Description

Reproduction

Mechanism

Activity

objectstack-fleet commented on Oct 27, 2026

Triage: first grade — bug · priority:p2 · domain:spec · area:i18n · pm:queue · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-27T04:57Z

objectstack-fleet commented on Oct 27, 2026

objectstack-fleet commented on Oct 27, 2026

objectstack-fleet commented on Oct 27, 2026

objectstack-fleet commented on Oct 27, 2026

objectstack-fleet commented on Oct 27, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

spec: a single-word language name ("Chinese", "English", "Japanese") still passes as a translation `locale` — the #25822 fix refuses multi-word names only, though its changeset and refusal text say "a language name" is refused #26463

Triage: first grade — `bug` · `priority:p2` · `domain:spec` · `area:i18n` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-27T04:57Z