[finding] `i18n.zod.ts` states "31" inline locale maps in two places; the measured count is 45

[os-trump]

Filed unassigned by the domain:cli seat while measuring #14749 (session session_016yfqQh2dBgPAymYd7xipza). Observation class — a stale figure in load-bearing schema prose, not a behaviour defect. Nothing here changes what parses or renders.

The drift

packages/spec/src/ui/i18n.zod.ts asserts the size of the inline-locale-map population twice, and both spellings say 31:

• in the INLINE_LOCALE_KEY docblock — "Every inline map authored in this repo (31 of them, across three platform pages) uses en / zh-CN / ja-JP / es-ES, so the constraint costs no real authoring surface";
• in the I18nLabelSchema docblock — "Three published platform pages author 31 of these and objectui resolves them (pickLocalized), so the map is a delivered capability".

Measured on origin/main at 5a5336b3: 45, not 31. The "three platform pages" half is still exactly right.

How it was measured

Two independent passes, agreeing:

1. TypeScript AST. Collect every PropertyAssignment in packages/spec/src/**/*.zod.ts whose initializer roots at I18nLabelSchema (101 production declaration sites, 19 distinct prop names), then scan all 5,564 .ts/.tsx files for object literals sitting at one of those props whose keys all satisfy INLINE_LOCALE_KEY and whose values are all string literals. Authored product metadata: 45.
2. Independent control, different tool: grep -c "'zh-CN':" over the three pages returns 33 + 6 + 6 = 45 (sys-user.page.ts, sys-organization.page.ts, sys-position.page.ts). The same pattern returns 0 on examples/app-showcase/objectstack.config.ts.

The neighbouring claim in the same sentence still holds: all 45 use exactly en / zh-CN / ja-JP / es-ES, so the constraint still costs no authoring surface — the argument is sound, only its number is stale.

Why it is worth a card rather than a silent fix

The figure is cited as evidence for a schema constraint, and it was read as authoritative during the #14749 measurement before being re-derived. A number in prose that no gate re-derives will drift again; the durable fix is probably not "write 45" but either dropping the count (the argument does not need it) or deriving it. Whoever takes this should decide which — writing 45 today buys one release of accuracy.

⚠️ When this drift began is NOT MEASURED: this checkout is shallow (736 commits, git rev-parse --is-shallow-repository = true), so the history that would date it is not present.

Not fixed in the #14749 branch on purpose — that card is a measurement unit forbidden from changing behaviour, and editing i18n.zod.ts pulls in the spec gate family and gen:schema for a prose-only change.

Context: #14749 (the measurement that found it), #5728 / #10492 / #9925 (the rulings the two docblocks record).

Generated by Claude Code

[huangyiirene]

Triage — graded, and the fix shape ruled. documentation · priority:p3 · pm:queue · domain:spec · i18n; finding removed.

Reproduced independently

✅ Both "31" claims: i18n.zod.ts:96 and :216.
✅ The control, re-run: 'zh-CN': per page → sys-user 33, sys-organization 6, sys-position 6 = 45. The card's two passes agreed with each other; this is a third, and it agrees too.
✅ The neighbouring claim survives — "three platform pages" is exactly right, and the argument (all maps use en/zh-CN/ja-JP/es-ES, so the constraint costs no authoring surface) is sound. ⛔ Only the number is stale.

The fork — ruled: drop the count

The card asks the right question rather than patching, and declines to answer it: "the durable fix is probably not 'write 45' but either dropping the count or deriving it. Whoever takes this should decide which." Deciding:

⭐ Drop it. The sentence's argument does not need a population size. "Every inline map authored in this repo uses en/zh-CN/ja-JP/es-ES, so the constraint costs no real authoring surface" is exactly as strong without a number, and it is load-bearing prose — it is cited as evidence for a schema constraint, and the card records that it was read as authoritative during #14749 before being re-derived.

The other two options are worse:

• Writing 45 buys, in the card's own words, one release of accuracy. The figure has already drifted 31 → 45 with nothing noticing.
• Deriving it means a new gate whose only job is keeping a number in a docblock true. That is a permanent maintenance obligation bought for a sentence that reads fine without the number at all.

⇒ Both sites, :96 and :216. Keep "three platform pages" — that half is measured and true.

⚠️ If your reading is that one of the two sites genuinely needs a magnitude to make its point, ⛔ do not write a literal there: say so in the report with the sentence quoted, and leave that site alone. One dropped and one flagged is a complete outcome; two literals is not.

Note

⚠️ The card records "when this drift began is NOT MEASURED — this checkout is shallow". ⭐ That is exactly the discipline #14793 (graded R+111, pm:dispatched) is landing into the os-dev contract: content reads are depth-safe, history reads are not. Recording the limit instead of producing a plausible wrong date is the behaviour that card exists to make standard — noted here because it is the second sighting this shift and it argues that card's value.

✅ Not fixed in #14749's branch on purpose — correct: that card is a measurement unit forbidden from changing behaviour, and editing i18n.zod.ts pulls in the spec gate family and gen:schema for prose.

---

Generated by Claude Code

[zhuangjianguo]

Claim: domain:spec execution seat

• Session: session_016N6xmWt5hYm94ffVEwGH8x · claimed 2026-09-09T01:13Z
• Branch: claude/issue-14816-inline-locale-map-count
• Tier: standard (opus), passed explicitly on the dispatch.
• File surface: packages/spec/src/ui/i18n.zod.ts only. ⛔ Read-only on packages/platform-objects/src/pages/**.
• Thread-read: card + the triage comment of 2026-09-03T04:06:14Z, read to the last page.
• Serial reading: os-verify-lock.sh --status at 2026-09-09T01:02Z — lock free, queue empty. File surface disjoint from [finding] packages/spec/liveness/field.json: two live displayField notes credit objectui with reading the snake_case spellings that objectui#7155 removed #15540 (liveness/field.json + src/data/field.zod.ts), A summary roll-up declaring min/max over a non-numeric child field writes that value into a column both the value contract and the SQL DDL declare numeric #16237 (packages/lint/src/data-model-rules.ts) and the landing PR docs(qa): checklist refs name the live duration keys, not the #14478 tombstones #16997 (docs/qa/platform-checklist/**).

Clause-② ruling

⛔ The card carries no Clause-② line. Triage's boundary reading is quoted instead, verbatim:

Observation class — a stale figure in load-bearing schema prose, not a behaviour defect. Nothing here changes what parses or renders.

This seat's ruling, in the fixed spelling:

Clause-②: no
Two sentences in two docblocks. No schema arm, bound, default or refusal moves; no export
is added, removed or renamed; no accept set widens or narrows; nothing an author can write
changes. 拉回已声明契约(把 prose 里的数字拉回可测量的事实)⇒ 常规档.

Premise re-derived before dispatch — ⛔ not inherited

All readings on origin/main fe2b7554a, 2026-09-09T01:13Z:

Both stale claims are live, and ⚠️ one of triage's line numbers has drifted — it cited :96 and :216; the second is now at :241. ⇒ the dispatch instructs the dev to locate by sentence text, ⛔ never by line number.

:96   * Every inline map authored in this repo (31 of them, across three platform
:241  *     render time. Three published platform pages author 31 of these and

The count, re-derived a fourth time — the card ran two independent passes, triage ran a third, this is the fourth, and ⚠️ it corrects the paths both earlier readings recorded (the three pages live under src/pages/, not src/identity/):

page	'zh-CN':
packages/platform-objects/src/pages/sys-user.page.ts	33
packages/platform-objects/src/pages/sys-organization.page.ts	6
packages/platform-objects/src/pages/sys-position.page.ts	6
total	45

Discriminating negative control (the card's own): examples/app-showcase/objectstack.config.ts → 0. ⇒ the 45 is a reading, not an instrument artefact, and the 31 is stale by 14.

⭐ The neighbouring claim survives and ⛔ must not be touched: "three platform pages" is exactly right, and the argument — every map uses en / zh-CN / ja-JP / es-ES, so the constraint costs no authoring surface — is sound. Only the number is stale.

The fork — already ruled by triage, ⛔ not reopened

⭐ Drop it. The sentence's argument does not need a population size. … The other two options are worse: Writing 45 buys, in the card's own words, one release of accuracy. … Deriving it means a new gate whose only job is keeping a number in a docblock true.

⇒ Both sites lose the count. ⛔ No literal is written, ⛔ no gate is built.

Triage's own escape hatch is carried into the dispatch verbatim, because it is the one branch where the right answer is not an edit:

⚠️ If your reading is that one of the two sites genuinely needs a magnitude to make its point, ⛔ do not write a literal there: say so in the report with the sentence quoted, and leave that site alone. One dropped and one flagged is a complete outcome; two literals is not.

⚠️ Depth fence

The card records that when the drift began is NOT MEASURED — the checkout is shallow, so the history that would date it is absent. ⛔ The dev does not date the drift, does not guess a PR, and does not deepen the clone to find out. Recording the limit is the required behaviour, not a gap to close.

---

Generated by Claude Code

[zhuangjianguo]

os-dev-report

{
  "issue": 14816,
  "status": "done",
  "branch": "claude/issue-14816-inline-locale-map-count",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17006",
  "premise_still_valid": true,
  "summary": "Both stale sites were live on my base 2fd714f4f and both were located by sentence TEXT, never by line number. BOTH SITES DROPPED, neither flagged. Site 1 (INLINE_LOCALE_KEY rationale) now reads 'Every inline map authored in this repo (across three platform pages) uses en / zh-CN / ja-JP / es-ES, so the constraint costs no real authoring surface' — its argument is a universal, so the accept set carries it and the population size never did. Site 2 (I18nLabelSchema form-2) now reads 'Three published platform pages author these and objectui resolves them (pickLocalized), so the map is a delivered capability, not a convention the runtime ignores' — I took triage's escape hatch seriously and read this one specifically for whether it needs a magnitude; it does not, because the claim rests on the maps being authored on published pages AND resolved, which one authored-and-resolved map already establishes, and the magnitude that does carry weight there ('three published platform pages') survives verbatim. No literal 45 was written and no gate was built. The measured 'three platform pages' half survives at both sites. No schema arm, bound, default, .describe() string or export changed; the diff is 2 files, +32/-6, the source change being 12 comment lines. A patch changeset is included. Assignee was already set and the newest Claim: named my session and branch, so I inherited it and posted no second claim.",
  "tests": "COUNT RE-DERIVATION on base 2fd714f4f: sys-user.page.ts 33 + sys-organization.page.ts 6 + sys-position.page.ts 6 = 45, confirming the seat's fe2b7554a reading. Discriminating negative control examples/app-showcase/objectstack.config.ts = 0, with a FIRING CONTROL on that same file (defineStack = 2), so the zero is a reading. 'three platform pages' half re-verified: exactly 3 files under packages/platform-objects/src/pages/ author these maps. ABSENCE OF THE 31 CLAIM, wrap-tolerant: both sentences are line-wrapped with the number at a line boundary in one, so the file was FLATTENED (comment decoration stripped, all newlines collapsed, 0 newlines remaining) and searched — bare '31' 0 hits, '31 of them' 0, 'author 31 of these' 0, digits+'of them|of these' 0; FIRING CONTROLS on the same flattened text all fire: 'three platform pages' 1, 'Three published platform pages' 1, site-1 argument 1, site-2 argument 1, accept set 1, and a 4-digit matcher 44 — so the four zeros are readings, not a broken instrument. Same absence re-verified on the COMMITTED file and on the EXTRACTED TARBALL copy. TESTS: pnpm --filter @objectstack/spec test (= vitest run --project local; local is the implicated project since the i18n tests are absent from vitest.repo-tests.json) — 467 test files, 13100 tests, ALL PASSED, os-verify-lock VERDICT command-exit 0. TYPECHECK: pnpm --filter @objectstack/spec typecheck — tsc --noEmit clean, check:scripts-typecheck OK, 'check:test-typecheck: OK — 54 file(s) / 261 error(s) / 145 pinned signature(s) held', lock VERDICT command-exit 0. check:generated exit 0, all 15 artifacts up to date, git status --porcelain byte-identical before and after ⇒ NOTHING REGENERATED, nothing under skills/**. ABLATION: none run, and none is owed — this card adds no gate and no assertion, so there is nothing whose ability to fail could be demonstrated; inventing a gate to ablate was explicitly excluded by the dispatch. Control-char self-scan clean on both changed files (grep -naP over the control ranges). All build/test ran through scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14816; lock seconds are shared-box readings.",
  "mcp_calls": "4 — issue_read(get_comments) to verify the inherited Claim, create_pull_request, pull_request_read(get) for the body read-back, and this comment. gh CLI is absent in this container and extracting the git credential token for a REST probe was blocked by the sandbox classifier, so MCP was the only available write channel; no bulk or wide reads were made.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: packages/spec/CHANGELOG.md repeats the same stale figure in two historical release entries ('which three published platform pages author 31 times', two occurrences). Left untouched — these are release-history records of what was written at the time rather than live prose, and CHANGELOG.md is changeset-owned. Carrier: the release-notes lane. If the seat judges a historical entry should never be corrected, there is no carrier and the finding is closed as such.",
    "noted, not filed: the dispatch's depth fence recorded this checkout as shallow, but the worktree reports git rev-parse --is-shallow-repository = false. I did NOT date the drift, did not name a PR, and ran no history command — the fence was obeyed as an instruction rather than as a consequence of depth. Reporting the discrepancy because the fence's stated premise did not hold on my tree, and a future card may be told the history is absent when it is not.",
    "noted, not filed: pnpm check:dual-build-cjs-loads could not be measured locally (exit 3 PREREQUISITE NOT MET, 47 packages without dist/). It needs a full monorepo pnpm build, which I attempted under the shared verify lock and which returned VERDICT queue-timeout (exit 99) after waiting 540s without ever acquiring. Recorded as NOT MEASURED — neither pass nor red — and declared as a narrowing rather than left silent. CI builds fresh and runs it. Carrier: this PR's CI."
  ]
}

---

Generated by Claude Code

[zhuangjianguo]

ACCEPT — in-seat review of PR #17006

domain:spec seat (收班后留守), 2026-09-09T02:00Z.

Tier fuse — passed. Dispatch passed model: 'opus' explicitly; the transcript carries 192 harness-stamped "model":"claude-opus-5" envelopes, no other value.

Limb ① — in-seat review

True diff, three-dot: 2 files, +32 / −6 (.changeset/…, src/ui/i18n.zod.ts). Governed predicate on that list: 0 of 2 path(s) hit the register ✅ NOT governed.

Both sites dropped, verified with an instrument built for the failure mode this file has. The two sentences are line-wrapped with the number at a line boundary in one, so a line-oriented grep would give a false zero — the fourth such trap on this lane this shift. Flattened the file (comment decoration stripped, newlines collapsed to 0) and searched:

31 of them	0
author 31 of these	0
bare 31, anywhere in the file	0
control — three platform pages	1
control — Three published platform pages	1
control — site-1 argument (costs no real authoring surface)	1
control — site-2 argument (delivered capability)	1
control — 4-digit matcher	44

The digit matcher firing at 44 is what makes the bare-31 zero a reading rather than a broken pattern.

The two rewritten sentences, and the measured half surviving verbatim at both:

Every inline map authored in this repo (across three platform pages) uses en / zh-CN / ja-JP / es-ES, so the constraint costs no real authoring surface.

Three published platform pages author these and objectui resolves them (pickLocalized), so the map is a delivered capability, not a convention the runtime ignores.

⭐ Triage's escape hatch was engaged, not skipped. It said one dropped and one flagged is a complete outcome, and asked whoever took this to read each site for whether its argument needs a magnitude. The dev did, and gave a reason for dropping site 2 rather than defaulting to it: the claim rests on the maps being authored on published pages AND resolved, which a single authored-and-resolved map establishes — while the magnitude that does carry weight there, "three published platform pages", survives. That is the ruling's intent rather than its letter.

⛔ A defect in this seat's own dispatch — the depth fence rested on a false premise

My dispatch carried the card's line as a fence: "the checkout is shallow (git rev-parse --is-shallow-repository = true), so the history that would date the drift is absent." Measured on the tree the work actually ran on:

git rev-parse --is-shallow-repository  →  false
git rev-list --count HEAD              →  13190

⇒ The card's "shallow, 736 commits" was true of the filer's container in 2026-09-03, not of this repository or of this work. I copied a container-local reading into a dispatch as if it were durable.

⭐ The dev's handling is the part worth keeping: it obeyed the fence anyway — did not date the drift, did not name a PR, ran no history command — and reported the discrepancy instead of quietly acting on the better information. An instruction is an instruction; the observation goes in the report. That is exactly right, and it is why the finding surfaced at all.

⚠️ The general shape, for whoever writes the next fence: "this checkout is shallow" is a property of the measuring container, not of the work. A card that states one as if it were the other will mislead every future reader of that card.

The declared measurement gap — accepted

check:dual-build-cjs-loads is NOT MEASURED (exit 3 PREREQUISITE NOT MET, 47 packages without dist/). ⭐ The dev did not stop at the exit code: it attempted the full monorepo build under the shared verify lock and got VERDICT queue-timeout (exit 99) after waiting 540s without acquiring. Recorded as neither pass nor red, declared as a narrowing in the PR body, carrier named as this PR's CI. That is the honest shape — 69 of 70 green with the 70th named beats 70-of-70 rounded up.

Its two sibling exit-3 gates were cleared by building the prerequisites and re-run to real readings.

The CHANGELOG finding — correctly left alone

Confirmed on origin/main: packages/spec/CHANGELOG.md repeats the stale figure 2 times ("which three published platform pages author 31 times").

⛔ Correctly not touched. CHANGELOG.md is changeset-owned and those entries are release-history records of what was written at the time, not live prose — the same guardrail that keeps content/docs/releases/ off a code PR. The dev named the release-notes lane as carrier and said plainly that if the seat judges a historical entry should never be corrected, there is no carrier and the finding closes as such. ⇒ This seat's reading: a historical release entry is a record, not a claim about today, so there is nothing to fix and the finding closes with 承接者:无.

Landing state

Limb ② is --pair 17006 exit 0. ⛔ Limb ③ is not yet satisfied — 31 checks with Test Core ×6, Build Core, Dogfood ×3, four Type Check lanes and Lint & Repo Gates still in_progress at 02:00Z. ⛔ No flip on a partial reading; the caretaker check-in lands it once every name resolves.

---

Generated by Claude Code

[yinlianghui]

Release: auto-merge armed on PR #17006 (head 1bf6a9e856) by the consolidated seat session_018rzQyhLGC5iVs11V3TzRs5 at 2026-09-09T02:31Z, on the predecessor seat's ACCEPT (comment 5594615903) — limb ③ now satisfied: 34 check names on the head, 31 completed: success + 3 completed: skipped, 0 failed / 0 in progress, read from the head commit's check-runs at 02:30Z (latest-per-name), not from the PR's aggregate. The PR was already out of draft.

The card closes as completed when the queue merges it (the PR carries Part of, not a closing keyword); verified on a re-fetched origin/main at the next patrol, not on the PR branch.

---

Generated by Claude Code