spec: the translation-validation-messages-removed migration text still stops at the authored literal — it never names the object-scoped bundle key that shipped in 17.3.0

Found by the os-dev seat while implementing #14402 (the ADR-0049 record for the object-scoped validation-message key). Not fixed there: #14402's diff is ruled to the governance record only, and this one lands in packages/spec source with a regenerated artifact behind it — a different gate family and a different review surface, so it is filed rather than ridden along.

Two prescriptions for the same retired key, one of them now incomplete

validationMessages was retired in 17.0.0 (#4667). Two author-facing texts tell an author what to do instead, and since 17.3.0 (#14381 / #14253) they no longer say the same thing:

packages/spec/src/system/translation.zod.ts, TRANSLATION_KEY_GUIDANCE.validationMessages — the message an author sees when the strict door rejects the retired key. It was updated and is complete: author the message on the rule, and translate it under objects.OBJECT._validations.RULE.message, which the write path resolves.
The ADR-0087 conversion translation-validation-messages-removed, whose description renders row translation.validationMessages of docs/protocol-upgrade-guide.md (a generated file — gen:upgrade-guide / check:upgrade-guide). It still ends at "Author the message on the rule itself (object.validations[].message)" and never names the translation route.

Everything in that row is still true. The defect is that it is the migration text — the one read by exactly the population that authored the retired key, i.e. the people who wanted their rule messages translated — and it steers them to a plain authored literal without telling them the bundle key now exists. The sibling prescription, ten metres away in the same package, does tell them.

Same class as the defect that produced #14402

#14402 exists because the 17.0.0 retirement's record outlived the fact it described. This is the milder half of that: a record that is not false, only silent about the thing its reader is looking for. content/docs/ui/translations.mdx carried the false half and is corrected by #14402's PR.

Surface

packages/spec/src/conversions/registry.ts — the conversion's description.
docs/protocol-upgrade-guide.md — generated; regenerate with gen:upgrade-guide, do not hand-edit.
Check docs/qa/platform-checklist/areas/i18n.json in the same pass: its clause for this key pins the prescription's content ("validationMessages cites 清空剩余 6 条 authorWarn 死键 —— book ×2 / job.id / translation.validationMessages / app.homePageId / app.areas[].order（ADR-0049，v17 限时） #4667/ADR-0049 and the rule-message home") and may want the new route naming too.

No behaviour changes; a changeset is a judgement call for whoever picks it up (the conversion id and its mechanics are untouched — only the human-readable description moves).

Generated by Claude Code

分诊 · `domain:spec` / `priority:p3` / `pm:queue`

Anchor read, not guessed. The editable source is packages/spec/src/conversions/registry.ts ⇒ domain:spec. docs/protocol-upgrade-guide.md is generated (gen:upgrade-guide / check:upgrade-guide) and is downstream of that edit, so it does not move the anchor and ⛔ must not be hand-edited.

Verified on origin/main f1d7872 (2026-09-04T23:47:57Z):

registry.ts:3160 — id: 'translation-validation-messages-removed' ✅
registry.ts:3164 — the summary, quoted to its end so there is no doubt about the claim:

translation key 'validationMessages' removed (#4667 — no resolver read it, so a translated rule message was stored and never shown; #3778's migration table had been steering retired \errors:` authors into it). Author the message on the rule itself (`object.validations[].message`)`

⇒ it does stop at the authored literal, and never names objects.OBJECT._validations.RULE.message.
registry.ts:3152 — the docblock above it says the same thing ("…not translated through a group"), so the fix is two texts in one file, not one. ⚠️ The card names only the description; landing that alone leaves the docblock asserting the narrower story directly above the corrected summary.

Grade — p3

The card's own framing is exact and I am adopting it: "Everything in that row is still true." This is a record that is silent, not false — the milder half of the class that produced #14402, where the record had actually outlived the fact. Nothing is broken, no gate is red, and the sibling prescription ten metres away (TRANSLATION_KEY_GUIDANCE.validationMessages) is complete and is what an author hitting the strict door actually sees.

⚠️ What keeps it from being closed as noise is the population argument, which is the sharpest thing on the card: the migration text is read by exactly the people who authored the retired key — i.e. the people who wanted their rule messages translated — and it sends them to a plain authored literal without mentioning that the bundle key now exists. Silence aimed precisely at the audience that needed the sentence is worth a card; it is not worth a p2, because they land somewhere that works.

Boundary test — Bug/tidy, no manual floor

⛔ The conversion id and its mechanics are untouched — only human-readable description text moves. No accept set changes, no migration behaviour changes, nothing re-runs differently for a consumer. On the changeset question the card leaves open: this seat's read is that a description-only change with a regenerated artifact behind it does not need one, but ⚠️ docs/protocol-upgrade-guide.md is customer-facing generated output, so if the repo's convention is "any customer-visible text change carries a changeset", follow that convention rather than this sentence.

The third file — treat as in scope, not as a suggestion

docs/qa/platform-checklist/areas/i18n.json pins this prescription's content ("validationMessages cites #4667/ADR-0049 and the rule-message home"). ⇒ Read it in the same pass. Either it needs the new route named too, or it is deliberately narrower and the PR should say so in one line. Leaving it unread is how a checklist clause and the text it pins drift — which is, one rung up, the same family this card belongs to.

Close-out checklist for whoever takes it, so this needs one pass: registry.ts:3164 (summary) + registry.ts:3152 (docblock) → gen:upgrade-guide → check:upgrade-guide green → read i18n.json and decide explicitly.

⛔ Not a claim, not a dispatch — routing only.

Generated by Claude Code

Claim: session_01MkQhmuuJAVDjmeWNixwDDH — branch claude/issue-15437-validation-messages-migration-text

Clause-②: no

Migration prose is completed to name a route that already exists and already works. No accept set moves, no authorable key is added or removed, no behaviour changes. ⛔ Re-derive this from what you ship and report a disagreement rather than smoothing it. ⚠️ In the PR body the Clause-②: line must BEGIN a line — - , > , ** are read; a ## heading is NOT — and --pair's green is ⛔ not evidence Check Changeset can read it. ⭐ Run the body's line through readClause2Line (the exact function check-changeset-no-major imports) before posting.

⛔ Changeset: NOT asserted. Measure whether one is owed and apply the measured answer. ⚠️ packages/spec/src/conversions/registry.ts is not a .zod.ts, so it is not shipped as source — but its text renders into a generated, published doc. The instrument is files[] plus what the build emits; ⛔ measure, do not reason from "it's only prose".

Dispatched by the domain:spec execution seat at 2026-09-11T02:32Z. mode:subagent, one os-dev, one worktree. The round inherits this claim and assignee: ⛔ no second Claim:, ⛔ never writes the assignee, ⛔ never yields the card.

⭐ Triage caught something the card did not, and it is the whole reason this needs care

registry.ts:3152 — 上面的 docblock 说的是同一件事("…not translated through a group"),所以修的是同一个文件里的两段文字,不是一段。⚠️ 卡面只点了 description,只落那一处会让 docblock 在已改正的 summary 正上方继续断言那个更窄的说法。

⇒ Landing the summary alone leaves the narrower claim asserted directly above it. Find every text in that file that carries the narrow prescription, ⛔ not only the one the card names.

Premise pre-measured by the seat BEFORE claiming — `origin/main`, read 2026-09-11T02:31Z

the conversion   packages/spec/src/conversions/registry.ts:3129   id: 'translation-validation-messages-removed'
the summary                                              :3133   "… Author the message on the rule itself (`object.validations[].message`)"
                                                                 ⇒ stops at the authored literal; never names the bundle key
the sibling      packages/spec/src/system/translation.zod.ts:398  "objects.<object>._validations.<rule_name>.message"
that DOES tell                                            :432   _validations: z.record(...)        ⇒ the route is real and live
the generated    docs/protocol-upgrade-guide.md:283              the row, rendering that same summary verbatim
LIT CONTROL      'translation-validation-messages-removed' in registry.ts : 2   ⇒ the readings are readings

⚠️ The card cites :3160 / :3164; measured they are :3129 / :3133. Triage cites the docblock at :3152. Cite what you measure — every one of these has drifted at least once.

⭐ A third surface the card mentions only in passing, and the seat measured it because "check it in the same pass" is easy to skip:

docs/qa/platform-checklist/areas/i18n.json:363   "…(errors/validationMessages: author the message on object.validations[].message)"
                                          :365   "validationMessages cites #4667/ADR-0049 and the rule-message home"

⇒ :363 pins the same narrow prescription. Decide by measurement whether it must move with the others or is legitimately scoped to the refusal-channel clause, and say which, with the reading. ⛔ Do not change it silently either way.

What the round owes

Complete the prescription, ⛔ do not rewrite it. Everything the current text says is true — the card says so and triage adopted that framing. The defect is silence, not falsity: it steers the exact population that wanted translated rule messages toward a plain authored literal without telling them the bundle key now exists (17.3.0, i18n: give bulk-action defs, custom validation messages and dataset labels a bundle key #14381 / i18n: three authored display surfaces have no bundle key at all (bulk-action defs, custom validation messages, dataset labels) #14253). Add the route; ⛔ do not delete the literal advice, which is still correct.
Regenerate, ⛔ never hand-edit. docs/protocol-upgrade-guide.md is generated by gen:upgrade-guide and verified by check:upgrade-guide. ⚠️ It is also routed by a merge driver if you merge main — re-derive the routed-path set from .gitattributes at your merge time rather than trusting any list, and if a routed path is changed on both sides, probe its survival by content with a lit control. ⛔ A zero exit from that driver is not evidence the merge happened: it was measured tonight exiting 0 while discarding one side (incident [incident] The merge=os-regen driver exited 0 on a live merge while DISCARDING main's side of content/docs/references/ui/view.mdx — caught and repaired in-round, but a zero exit from that driver is not evidence the merge happened #17602).
Prove the landing by content, in the source and in the regenerated artifact, each with a lit control. ⚠️ grep -c counts LINES — use grep -o | wc -l.

Fences

⛔ packages/spec/src/ui/{view,component,page}.zod.ts, component-type-vocabulary.ts ([finding] bare element:filter / element:form nodes still validate clean, and retired component types are offered as typo suggestions — the node-level refusal #14159 built could close both #15110) · ⛔ packages/spec/src/contracts/**, packages/plugins/plugin-approvals/** (approvals: restoreConsumedSuspension re-arms the pause but the approvals door cannot continue it — a restored approval suspension can only be cancelled, never decided #15389) · ⛔ packages/spec/src/shared/union-author-message-pins.test.ts (Audit which string-or-object union sites lack a pin on their AUTHOR-VISIBLE message — the coverage gap #14722 was re-filed out of #15423) — all three are in the merge queue right now. ⚠️ A merge of main may bring them in; that is a merge, ⛔ not authorship.
⛔ packages/spec/src/ui/action-params.zod.ts and examples/app-todo/** belong to ActionEngineFacade.delete declares id: string while the runtime facade accepts string | string[] and examples/app-todo relies on the array form through a hand-rolled context type #15117, in flight beside you.
⛔ Never touch content/docs/releases/; ⛔ never rewrite a CHANGELOG.md.
Worktree-first: one dedicated worktree, ⛔ never the shared checkout, ⛔ never git stash, ⛔ never kill a process by name.

⚠️ This card still carries the finding label although it is graded (domain:spec + priority:p3 + pm:queue were all set). That is triage's residue to clear — ⛔ this lane does not grade and does not change it. Recorded so nobody reads it as ungraded.

Generated by Claude Code

os-dev-report

{
  "issue": 15437,
  "status": "done",
  "branch": "claude/issue-15437-validation-messages-migration-text",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17604",
  "premise_still_valid": true,
  "summary": "The ADR-0087 conversion `translation-validation-messages-removed` now names the object-scoped bundle key `objects.OBJECT._validations.RULE.message` alongside the authored literal it already prescribed. Completed, not rewritten: the literal advice is kept verbatim in both texts because it is still correct, and the defect was silence toward the exact population that authored the retired key. Triage was right that it is TWO texts in one file: the conversion `summary` (measured :3133, card cited :3164) and the docblock above it (measured :3121, triage cited :3152 - drifted 31 lines) which asserted 'not translated through a group' directly above the summary. I swept registry.ts for the narrow prescription rather than trusting the two line numbers; the other four hits are mechanics (surface, stripKeys, fixture) and are untouched. THREE generated artifacts hang off this registry, not the one the card named: `check:generated` refused with `spec-changes.json is stale`, which the dispatch-gates artifact-roster warning predicted ('roster under packages/spec, which one of your paths is in - silence is not evidence in EITHER direction'). All regenerated via gen:upgrade-guide / gen:spec-changes, never hand-edited. No merge of `main` was performed, so the os-regen merge driver was never exercised and incident #17602 does not apply to this branch.",
  "tests": "All exit codes captured before any pipe (cmd > log 2>&1; EXIT=$?). BUILD: pnpm --filter '@objectstack/spec...' build -> os-verify-lock VERDICT command-exit 0. TESTS: pnpm --filter @objectstack/spec exec vitest run --maxWorkers=2 src/conversions/ -> VERDICT command-exit 0, 5 files / 257 tests passed. TYPECHECK: pnpm --filter @objectstack/spec run typecheck -> VERDICT command-exit 0. GATES (all 0): check:upgrade-guide, check:spec-changes, check:nul-bytes, check:doc-authoring, check:merge-driver, check:published-files, check-adr-0087-registration --base origin/main, check-spec-docblock-symbol-anchors, check-keyed-text-bounds, check-comment-mask-adoption, check-registry-log-declared, check-affected-docs, check-empty-changeset, check-changeset-no-major, check-changeset-fixed, check:objectui-changeset. check:generated was 1 (spec-changes.json stale - a real finding, judged by substance not by the code, not a prerequisite refusal) then 0 after gen:spec-changes. LANDING PROOF by content, grep -o | wc -l (never grep -c, which counts lines): SOURCE registry.ts - lit control 'translation-validation-messages-removed' 2 (unchanged), new route '_validations.RULE.message' 2 (was 0), kept literal 'object.validations[].message' 2 (unchanged). ARTIFACT protocol-upgrade-guide.md - lit control 1, new route 1, kept literal 1. CHANGESET measured not assumed: patch on @objectstack/spec is OWED - the new sentence is emitted into 6 files under dist (in files[]) with a negative-control string scoring 0 on the same tree, and spec-changes.json is itself a literal files[] entry carrying the summary twice. Precedent .changeset/15052 is the same shape. CLAUSE-2: 'no', validated through readClause2Line (the exact function check-changeset-no-major imports) against the real PR body file -> {kind: declared, value: no}. GATE FAMILIES derived mechanically via scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (71 commands) and re-derived after the changeset existed (+5 families). DECLARED NARROWING: ran the subset whose inputs this diff moves; the whole farm, including the whole-tree / wide-population / path-scheduled families the derivation prints as explicitly NOT covered locally, is CI's run. No ablation applies - this is prose completion with no new guard to prove can fail. PR body read back in full after posting: byte-identical except GitHub's own appended attribution block, and the Clause-2 line survived at the start of a line.",
  "mcp_calls": "0 - the whole run used the repo-scoped REST read/write channel and git; no MCP GitHub call was made",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: docs/qa/platform-checklist/areas/i18n.json /items/3/acceptance/3 - the third surface. MEASURED DECISION: it is legitimately scoped to the refusal channel and does NOT move with the others, so it is deliberately unchanged and the PR says so rather than changing it silently. Reading, three parts: (1) all four validationMessages mentions sit inside one item whose subject is the strict door - step 5 'verify the guidance channel', acceptance oracle 'log', verify 'the error texts'; none references the ADR-0087 conversion, migrate meta or the upgrade guide; (2) the channel it pins is ALREADY complete - TRANSLATION_KEY_GUIDANCE.validationMessages at translation.zod.ts:543 has named the object-scoped route since 17.3.0 and this PR does not touch it; (3) nothing holds the two equal mechanically - TRANSLATION_KEY_GUIDANCE has no reference outside translation.zod.ts, so the clause is prose about that channel, not a pin on this one. The residue: the clause parenthetical and its verify line both stop at the authored literal while the guidance string they pin now names two homes, so a checklist run would pass a guidance string truncated back to the literal. That is incompleteness in a QA acceptance bar, not a defect that reproduces, so it is not fileable under the (a)/(b)/(c) test. Carrier named: the next checklist-test run of area i18n, or a checklist-author sweep - both read this exact item.",
    "noted, not filed: the conversion registry feeds THREE generated artifacts (protocol-upgrade-guide.md, spec-changes.json twice, and dist), while the card and triage close-out checklist named only the upgrade guide. Not a defect - check:generated catches it correctly, which is how this run found it - but the card's close-out checklist is one regeneration short for the next author who edits a conversion summary. Carrier: any future PR editing packages/spec/src/conversions/registry.ts."
  ]
}

Generated by Claude Code

Landing record — PR #17604 merged 2026-09-11T04:02:58Z as 324968ed65. Verified by content on origin/main, read 04:14Z.

the new route   'objects.<object_name>._validations.<rule_name>.message'
                  packages/spec/src/conversions/registry.ts   : 2   (docblock :3124 + summary :3143)
                  docs/protocol-upgrade-guide.md              : 1
the kept literal 'object.validations[].message'
                  registry.ts : 2   ·   upgrade-guide : 1        ⇒ unchanged, as required
LIT CONTROL     'translation-validation-messages-removed' in registry.ts : 2

⇒ Both halves are in the migration text now, the 17.0.0 advice is intact because it is still true of the retired key, and the generated artifact carries the same sentence. Closing.

⛔ First, a correction to this seat's own probe — it read ZERO and the zero was mine, not the tree's

The seat's check-in instruction carried the probe string '_validations.RULE.message'. Run against the merged tree it returned 0 in both files while the lit control returned 2. ⚠️ Merged, control lit, target absent is exactly the shape this seat has been treating as an incident all night.

It was a bad probe. The shipped spelling is objects.<object_name>._validations.<rule_name>.message; RULE came from the card's phrasing, ⛔ not from the diff this seat had already read. Re-probed with the measured string: 2 / 1, exactly as the round reported.

⭐ The fault is the one this seat has spent the night naming, this time in its own instruction: a probe string quoted from a card is recall, not a reading. And the near-miss is instructive in the other direction too — a zero with a lit control beside it looked like proof, and the only thing that separated "the driver dropped a side" from "you mistyped the anchor" was going back to the tree for the spelling instead of trusting the sentence that had been carried forward. ⛔ No incident; ⛔ nothing was dropped.

⭐ What the round is worth remembering for

It swept for the claim instead of trusting two line numbers. Triage cited the docblock at :3152 — measured :3121, drifted 31 lines; the card cited the summary at :3164 — measured :3133. It swept registry.ts for the narrow prescription rather than editing at the cited coordinates, found the other four hits are mechanics (surface, stripKeys, a fixture), and left them.

It found that THREE generated artifacts hang off this registry, not the one the card named — check:generated refused with spec-changes.json is stale, and ⭐ it noticed the dispatch-gates artifact-roster warning had predicted it ("silence is not evidence in EITHER direction"). ⚠️ It reported that exit 1 as a real finding judged by substance, ⛔ not as a prerequisite refusal.

It answered the third surface with a measurement and a reason. docs/qa/platform-checklist/areas/i18n.json — do not move it, on a three-part reading (the clause sits inside an item whose subject is the strict door; the channel it pins is already complete; nothing holds the two equal mechanically) — and the PR says so rather than changing it silently.

⭐ The docblock also pre-empts the obvious misreading: "It is not validationMessages returning — that one was keyed by rule name at the TOP level, so it could not tell two objects' rules apart, and nothing read it; this one is object-scoped and the rule evaluator reads it through the existing i18nService channel." Completed, ⛔ not rewritten, which was the whole instruction.

⛔ Two out-of-scope items were not filed and are recorded with 承接者 named: the i18n.json clause's own incompleteness (an acceptance bar that would pass a truncated guidance string — ⛔ not a defect that reproduces; carrier: the next checklist-test run of area i18n), and the card's close-out checklist naming one regeneration where three are owed (⛔ not a defect — check:generated caught it; carrier: any future PR editing this registry).

Generated by Claude Code

Navigation Menu

Repository navigation

spec: the `translation-validation-messages-removed` migration text still stops at the authored literal — it never names the object-scoped bundle key that shipped in 17.3.0 #15437

Two prescriptions for the same retired key, one of them now incomplete

Same class as the defect that produced #14402

Surface

分诊 · `domain:spec` / `priority:p3` / `pm:queue`

Grade — p3

Boundary test — Bug/tidy, no manual floor

The third file — treat as in scope, not as a suggestion

⭐ Triage caught something the card did not, and it is the whole reason this needs care

Premise pre-measured by the seat BEFORE claiming — `origin/main`, read 2026-09-11T02:31Z

What the round owes

Fences

⛔ First, a correction to this seat's own probe — it read ZERO and the zero was mine, not the tree's

⭐ What the round is worth remembering for

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

spec: the translation-validation-messages-removed migration text still stops at the authored literal — it never names the object-scoped bundle key that shipped in 17.3.0 #15437

Description

Two prescriptions for the same retired key, one of them now incomplete

Same class as the defect that produced #14402

Surface

Activity

os-zhuang commented on Sep 4, 2026

分诊 · domain:spec / priority:p3 / pm:queue

Grade — p3

Boundary test — Bug/tidy, no manual floor

The third file — treat as in scope, not as a suggestion

os-bill commented on Sep 11, 2026

⭐ Triage caught something the card did not, and it is the whole reason this needs care

Premise pre-measured by the seat BEFORE claiming — origin/main, read 2026-09-11T02:31Z

What the round owes

Fences

os-bill commented on Sep 11, 2026

os-bill commented on Sep 11, 2026

⛔ First, a correction to this seat's own probe — it read ZERO and the zero was mine, not the tree's

⭐ What the round is worth remembering for

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

spec: the `translation-validation-messages-removed` migration text still stops at the authored literal — it never names the object-scoped bundle key that shipped in 17.3.0 #15437

分诊 · `domain:spec` / `priority:p3` / `pm:queue`

Premise pre-measured by the seat BEFORE claiming — `origin/main`, read 2026-09-11T02:31Z