feat(i18n)!: retire 23 measured-dead pack keys, two families whole (objectui#8754) - #9487
Merged
os-try-charles merged 2 commits intoSep 14, 2026
Merged
Conversation
…bjectui#8754) Re-ran `check:i18n-dead-keys` on this change's own base after the objectui#9222 instrument repair and deleted what it confirms today, rather than the list frozen in the card's body. The card said 25; the measured population is 23. The difference is arithmetic in the card, not tier movement: it counted eleven non-family leaves plus two families of six, double-counting the two leaves that belong to both sets. Every one of the 13 keys PR objectui#8753 uncovered is still CONFIRMED today. Both families retire whole, as roots rather than emptied leaf by leaf. Their two NEEDS-REVIEW members are admissible under the ruling's family rule — one hit each, both non-liveness (a changelog line; a pin-test allowlist entry, deleted with the key on that file's own `workflow.*` precedent). Census, same instrument, before -> after: pack keys 3010 -> 2987 (-23) candidates 358 -> 335 (-23, exactly the deleted set) CONFIRMED 119 -> 98 (-21) NEEDS-REVIEW 239 -> 237 (-2) No key joined the candidate set and no surviving key changed tier. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013VGeMu3p6qEFWR6K6GGLaW
Contributor
✅ Console Performance Budget
The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it. 📦 Bundle Size Report
Size Limits
|
Contributor
✅ Console Performance Budget
The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it. 📦 Bundle Size Report
Size Limits
|
Collaborator
CI green on the current head — and the landing probe, written BEFORE the merge
The CI reading, with a control that fires
32 + 3 + 0 + 1 = 36 ✓ (the last before shard 1 landed). ⭐ CONTROL, same instrument, same call shape, same minute — the identical File surface, verified independently of the PR body14 files: 10 locale packs at exactly
|
| # | probe | expectation | control in the same reading |
|---|---|---|---|
| P1 | diff M^ → M over the ten packages/i18n/src/locales/*.ts |
each +0 / −27 |
dead-pack-keys-retired-8754.test.ts shows +253 / −0 ⇒ the reader can see additions, so +0 is a reading |
| P2 | GET contents packages/i18n/src/__tests__/dead-pack-keys-retired-8754.test.ts at merged main |
200 | a sibling path known absent ⇒ 404 |
| P3 | counts in packages/i18n/src/locales/en.ts, measured at the actual M^ and M |
^\s*help: −1, ^\s*approvals: −1, ^\s*pricing: −1 |
survivors unchanged across the same pair: helpTooltip, approvalsInbox, ^\s*category: |
| P4 | resolve M and M^ |
⛔ after the merge, from the merge commit itself | — |
⛔ P3's absolute numbers are NOT pinned here on purpose. On this branch's own base they read 5 / 2 / 1, but M^ is whatever main is when the queue builds it, and other PRs touch these packs. The delta is the assertion; the endpoints get measured, never predicted.
⛔ P4 is a red line for this seat: the parent gets resolved from the actual merge commit after the fact. Predicting M^ has bitten this lane four times, and the parent has repeatedly turned out to be this seat's own previous merge.
⇒ Flipping ready, re-reading Governed Surface Queue Guard after the flip, then enqueueing. ⛔ No approval from this seat, ⛔ no direct merge, ⛔ no queue bypass.
Generated by Claude Code
os-try-charles
marked this pull request as ready for review
September 14, 2026 10:59
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Sep 14, 2026
Collaborator
⛔ Dequeued at 11:22:13Z — ⭐ the tests PASSED. Root-caused, ⛔ not called a flake, and re-queued once.
What actually happenedMerge group
Job wall clock 20m14s against ⇒ the job was killed on the step AFTER the tests, by its own ceiling. This workflow's own comment at ⛔ Why this is not this PR's failure, established rather than asserted
⇒ filed as objectui#9499 (
What I am doing about itRe-queueing once, and only once. That is the single re-run this seat allows itself after establishing a failure is not the PR's — ⛔ it is not a "let's see if it's flaky" retry: the tests are green on this head, the cause is named, and 11 of the last 14 runs of this job finish inside the ceiling. ⛔ If it is dequeued a second time, objectui#9499's trigger (1) fires, it becomes p1, and this PR waits for the CI repair rather than being re-queued a third time. ⛔ No test skipped, ⛔ no ceiling raised, ⛔ no queue bypassed, ⛔ no approval from this seat. The landing probe registered at Generated by Claude Code |
os-try-charles
deleted the
claude/issue-8754-retire-dead-pack-keys-round2
branch
September 14, 2026 11:55
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #8754
Round 2 of the director seat's summon #22 ruling (2026-09-12): instrument first, deletion second. Round 1 — the instrument repair — landed
c4a1d39fe9(PR #9222). This is the deletion.The admission criterion was a measurement, not this card's body
⭐ The 25 keys in the card's body were not the input.
pnpm check:i18n-dead-keyswas re-run on this branch's own base,ff1d5ea8d1, and the deletion is what it confirms today.Its verdict, quoted from its own output:
The population named in words: 3010 keys defined by the
enpack (the ten packs are parity-identical, whichall-locales-key-parity.test.tsseparately enforces); of those, 358 have no call site the AST pass can see; of those, 119 also have no bounded textual occurrence anywhere else in the repository — that is the CONFIRMED tier. This run agrees with round 1's recorded 119 CONFIRMED, and NEEDS-REVIEW has moved 238 → 239 since (one key joined, from ordinary tree movement).Does it agree with the card's 25? No — the measured population is 23, and the difference is the card's arithmetic
Every one of the 13 keys PR #8753 uncovered is still CONFIRMED today — none moved tier. Both families are intact, each 5 CONFIRMED + 1 NEEDS-REVIEW, exactly as the ruling anticipated.
The card reached 25 as
13 + 6 + 6, double-counting the two leaves that belong to both sets:approvals.rejectandmarketplace.pricing.freeare each one of the 13 and a member of one of the families. The true union is 23. ⇒ no key was dropped for being live, and no key was added; the card simply counted two keys twice.Family completeness, each zero carrying a control that fires
The ruling's family rule: a family is whole when every member is CONFIRMED, or is NEEDS-REVIEW whose only hits are non-liveness. Both families qualify, and both are removed as roots, not emptied leaf by leaf.
Membership was read from the pack source in all ten packs, not inferred:
approvalsholds exactly 6 members andmarketplace.pricingexactly 6, in every pack. All 12 are candidates ⇒ no member has a call site.approvals.*— probe and control in the same command, overpackages/ apps/ examples/ e2e/ content/ scripts/minus the ten locale packs (which define every key by construction):approvals.(approve|approveSuccess|comment|reject|rejectSuccess|rejectConfirm)at key boundarypackages/app-shell/CHANGELOG.md:6161— changelog prose about a past change. Zero source hits.approvalsInbox., same command shapeDeclaredActionsBar.tsx:230/236/241,approverIdentity.ts:365/373, …approvalsInboxhitting whileapprovals.does not is the reading the card predicted.marketplace.pricing.*:marketplace.pricing.at key boundarypackages/i18n/src/__tests__/untranslated-identity-4376.test.ts:105— a pin-test allowlist entryt()/tt()call whose key mentionspricingat allpricinganywhere inpackages/app-shell/src/console/marketplace/marketplace.category., same command shapeMarketplacePackagePage.tsx:654,MarketplacePage.tsx:185t()call whose key mentionscategorymarketplace.category.is a collected dynamic template head, which is exactly why none of that family was ever a candidate — and whymarketplace.pricing., which has no head, was.The two NEEDS-REVIEW members are the two the ruling named as admissible, and both hits are non-liveness: a changelog line and a pin test. ⇒ both families retire whole.
The eleven individual leaves
Each was held out of an earlier sweep by a longer live sibling that merely contains it, until PR #8753 put a key-boundary requirement on both sides of the text probe. A repo-wide bounded grep over all 23 keys returns exactly the two hits above and nothing else — and that same command returning non-zero for those two is what proves the other 21 zeros are readings rather than a broken pattern.
Firing control on the deletion itself
⭐ Without this, "the keys are gone" is indistinguishable from "the pack reader broke".
dead-pack-keys-retired-8754.test.tsasserts, in the same run: every retired key is absent from all ten packs; both family roots are gone; and 20 named surviving siblings still resolve to a non-empty string in all ten packs — including the longer sibling that actually renders in each of the eleven leaf cases (sidebar.helpTooltipforsidebar.help,home.starredApps.titleforhome.starred, …). A sample across writing systems pins that those survivors are real translations, not the English value ten times.Reverse-verified. One retired key was restored into one pack (
zh) on the committed tree, proven to have landed on disk by blob hash (3aab235a78→de6dff3fa1) and by the leaf spelling going0 → 1:Both pins red and name the pack and the key, while 35 tests in the same run stay green — that green half is the survivor assertion, so the red is about the retired key and not about the reader. The mutation was restored under a shell
trapand the restore verified by state, not exit code: on-disk hash equals theHEADblob,git diff HEADempty,git statusclean.All-locales key parity — before and after
Measured on two real trees, not inferred.
ff1d5ea8d1(this branch's base, in its own worktree)ebcbb22e5fParity passes on both sides; the
+1 file / +5 testsis exactly this PR's new pin. Parity is preserved rather than trivially satisfied because every deletion is a ten-pack edit — 27 source lines removed from each of the ten packs, identical in each.The card forbids pinning keys by spelling them, because a spelled key becomes a textual hit for the sweep and is pushed back into NEEDS-REVIEW. A first draft of the new pin's header comment spelled out five keys as a warning not to name them. The post-deletion census then read 93 CONFIRMED where the arithmetic said 98 — those five, demoted by the very sentence telling the reader not to do it. The spellings were replaced with a description of the shape, and the census landed on 98 exactly.
Recorded here and in the test's header because the demotion is silent, and the five keys are not exotic: they are the nearest live-looking sibling in each of the
dashboard,form,sidebar,homeandworkspacenamespaces, and every one of them is CONFIRMED-dead itself today.Census, before → after, with every number accounted for
ff1d5ea8d1)ebcbb22e5f)en)The set that left the candidate list is exactly the 23 deleted keys; no key joined, and no surviving key changed tier. 21 of the 23 were CONFIRMED and 2 were the admissible NEEDS-REVIEW members ⇒
119 − 21 = 98and239 − 2 = 237, which is what the run prints.Bytes — re-measured, ⛔ no budget constant moved
The ruling asks for the⚠️ That chunk no longer exists: objectui#7479 made nine of the ten catalogues lazy and re-keyed the ceiling to
i18n-localeschunk figure.i18n-locale-en. Reported against the successor, from two real console builds:i18n-locale-enchunkcheck:eager-closuregreen on both trees. ⛔PER_CHUNK_GZIP_CEILINGS,PER_CHUNK_BASELINEandMAX_EAGER_CLOSURE_GZIP_BYTESare untouched — this PR changes no file underscripts/.Gates run
pnpm --filter @object-ui/i18n testpnpm --filter @object-ui/i18n type-checktsc --noEmit+ test tsconfigpnpm --filter @object-ui/i18n lintcheck:i18n-keyscheck:i18n-driftenvalues changed in rangecheck:i18n-designer-parityenrow has azhrowcheck:i18n-dead-keyscheck:eager-closurecheck:eager-locale-cataloguescheck:control-bytescheck:test-path-rootscheck:new-line-citationscheck-changeset-presencecheck-changeset-no-majormajordeclaredgates_weakened:NONE — no threshold lowered, no ceiling raised, no non-vacuity floor moved (the> 2000floors read 2987), no test skipped.Repo-wide
eslint .is CI's run, not this PR's: the 13 changed TS files lint 0 errors / 0 warnings, andpnpm lintisturbo run lint(per-package), whose@object-ui/i18nleg is the row above.Changeset — measured, not assumed
Owed, and
check-changeset-presencesaid so in its own words.@object-ui/i18nis published (privateunset,v17.6.0) and in the 40-package fixed release group, and this PR changes 12 files under itssrc/..changeset/8754-dead-pack-keys-retired.mdlists every one of the 23 keys, states the breaking semantics and the out-of-repo consumer risk the ruling accepts, and names the live replacement surface for each family. Markedminorper this repo's rule that a changeset never declaresmajor.维护者速读(草稿)
改了什么:从十个语言包里删掉 23 条已确认没人用的句子 —— 两个完整的功能族(审批
approvals.*、应用市场定价marketplace.pricing.*),外加 11 条零散的旧句子。为什么改:这些句子仓库里没有任何地方再引用,但仍随每个版本发布、占着打包预算。卡片正文写的是 25 条,但那是算错了(有两条被数了两遍);本次按仪器当天实测的结果删了 23 条,不是照抄卡片上的数字。
风险与代价(含回滚):
@object-ui/i18n是对外发布的包。仓库外如果有人在用被删的句子,那里会显示原始键名而不是句子,不报错。这个风险是维护者 2026-08-27 已经拍板接受的(「项目在创业阶段,用户也很少,短期不考虑渐进」)。回滚 = 一次 revert,把键加回去即可。打包体积减少约 1.8 KB(gzip,十个语言包合计)。席位意见:(留空,待评审)
你要做的:确认「立即退役、不设过渡期」这一条仍然成立;如果您希望改成保留一段时间,请说一声,这个 PR 需要重做。
🤖 Generated with Claude Code
https://claude.ai/code/session_013VGeMu3p6qEFWR6K6GGLaW
Generated by Claude Code