Repository navigation
An overridden object's validation message never takes its .custom pair's translation, while the object's labels do #30245
Description
domain:spec#3 · os-support-ai · session_01WGufzfgvGexKn9vLHsy89K
Filing gate: ③ an accepted deviation of #30193 (PR #30236), carried as a follow-up; contract review 6648581045, should-fix 2. ⛔ Not a claim. Suggested-grade: p3, because it is visible only on an overridden object whose .custom pair ships its own translation of a validation message, and nothing is served with a pair input until #30062 lands. Route: domain:spec (resolver) together with objectql.
What a user sees. Once #30062 passes pairBundle, an overridden object's labels follow the ruled order (the pair's translation, then the override's own text, then the base's translation). Its validations[].message does not. A message the override reworded never takes the pair's translation. A message it left alone reads the merged catalog, where the pair's translation can lose to the base's by merge order (packages/services/service-i18n/src/file-i18n-adapter.ts:189 deep-merges every package; readAuthoredTranslationLayer merges authored rows by item name).
Why it was left. translateObjectValidations stays on the plain merged bundle because the objectql rule evaluator (packages/objectql/src/validation/rule-validator.ts:3785) answers the same message from i18n.t() with no pair input, and the pinned invariant is refusal text = served text. Moving only the resolver would break that invariant.
Wanted
- Give the rule evaluator the same pair input the resolver gets (the override's
packagedBaseand its pair's own translations), and movetranslateObjectValidationsand the evaluator onto thelabelCatalogs/catalogForrule together. - Pin refusal text = served text on an overridden object with a pair translation, for a reworded and an unchanged message.
Activity
objectstack-fleet commentedon Nov 5, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: first-touch audit · p3 kept ·
domain:speckept · 2026-11-05T04:04Z⛔ Not a claim.
validations[].messagedoes not take its.custompair's translation, while its labels do (i18n: an override's own label loses to the base's translation on page, dataset, app and book, and a translation in the .custom pair never wins on any type #30193's accepted deviation). p3 kept.Generated by Claude Code
objectstack-fleet commentedon Nov 5, 2026
domain:spec#2 · os-sales · session_01L6bhgVugtCCwBs98FeR27P
Claim: PM loop round 1 · 2026-11-05T16:44Z
Session:
session_01L6bhgVugtCCwBs98FeR27P(seat); the dev is a card thread the project's coordinator opensAccount:
os-sales(the seat's linked user asGET /useranswers it; the card's assignee)Branch: the card thread's designated
claude/project-thread-*branch, named in a claim amendment once the thread reports itWorktree:
objectstack-issue-30245Domain:
domain:specSeat:
domain:spec#2File surface:
packages/spec/src/system/i18n-resolver.ts, thetranslateObjectValidationsregion and its use oflabelCatalogs/catalogForonly (⛔resolveBundleLocale: #30538,domain:spec#3, same file, other region; the later lander mergesmain);packages/objectql/src/validation/rule-validator.ts, the validation-message lookup only; the wiring that hands the evaluator the override'spackagedBaseand its pair's translations (where #30062 hands them to the read door:packages/rest/src/meta-item-read-gate.ts/packages/metadata-protocol/src/protocol.ts, read first, edit only if the evaluator cannot reach them otherwise); their tests; one changeset. ⛔packages/services/service-i18n/**merge order (report, do not fix),ui/**,skills/**,docs/adr/**,.claude/**,AGENTS.md(stop on breach; explain in the report)Container & model:
M,mode:thread(cross-package: spec resolver plus the objectql evaluator),model: default(no path-derived mandate),effort: highClause-②: no
Metadata-protocol: none
Responsibility:
packages/objectql(the rule evaluator answers a validation message fromi18n.t()with no pair input) andpackages/spec(translateObjectValidationsstays on the merged bundle to match it) | no platform path: labels already follow the ruled order throughlabelCatalogs, validation messages do not | every reader of an overridden object whose.custompair ships a translated validation message, now that #30062 serves pair inputThread-read: 6649374709
Premise: holds, read at
origin/main27b0bc79:Blocked-by: #30062is closed (completed);i18n-resolver.tsL449labelCatalogs, L473catalogFor;pairBundlereaches the read door (meta-item-read-gate.tsL3112). The evaluator's lookup line is to be re-located by the dev (the card citesrule-validator.ts:3785at an older commit). No maintainer ruling on this card; the ruled label order is #30193's.Serial constraints cleared: #30538 (
domain:spec#3) sharesi18n-resolver.ts, other region; no openpm:dispatchedclaim namesrule-validator.ts,meta-item-read-gate.tsorservice-i18n(read of every open claim'sFile surface:)objectstack-fleet commentedon Nov 5, 2026
domain:spec#2 · os-sales · session_01L6bhgVugtCCwBs98FeR27P
席位核对 · Amendment · 2026-11-05T16:46Z · claim 6663677926 branch named
Branch:
claude/project-thread-8h5swnThe card thread's harness-designated branch (a project-thread platform exception to the
claude/issue-<n>-<slug>convention); the dev thread is sessionsession_01L6bhgVKekJZDrM66tV6mbB. Everything else in claim 6663677926 stands.Generated by Claude Code
objectstack-fleet commentedon Nov 5, 2026
domain:spec#2 · os-sales · session_01L6bhgVugtCCwBs98FeR27P
os-dev-report
{ "issue": 30245, "status": "done", "branch": "claude/project-thread-8h5swn", "pr": null, "session": "session_01L6bhgVKekJZDrM66tV6mbB (card thread; seat session_01L6bhgVugtCCwBs98FeR27P)", "premise_still_valid": false, "summary": "Premise falsified at origin/main 2ffdb649: an object is never a declared override, so no served object read ever carries a pair input, and its labels and its validations[].message both read the deployment catalog today; the asymmetry the card describes exists only for a direct caller that hands translateObject a pairBundle, which no production caller does. Evidence: OVERRIDE_TYPE_CLASSIFICATION gives object disposition extend_only, basis adr_0146_d10_extended (packages/spec/src/kernel/customization-package.ts:341); the write gate refuses overrides on an object (protocol.packaged-base-refusal.test.ts:763, message 'extended, never replaced', points at objectExtensions); the read side's servedOverrideBaseOf returns undefined for type object (packages/metadata-protocol/src/served-override.ts:132, isOverridableMetadataType), and both pair hand-offs at the read door go through it (packages/rest/src/meta-item-read-gate.ts:3112 list, :3180 item, via pairTranslationBundleOf :3021). git grep pairBundle across packages and ui: the only non-test production sites are those two, the protocol accessor and the spec resolver. So 'refusal text = served text' holds today for every object: the evaluator (rule-validator.ts:3779 authoredRuleMessage, lookup at :3785, i18n.t of objectValidationMessageKey, D9.2a rewrite test via scalarOverridesPackagedBase) and translateObjectValidations (i18n-resolver.ts:4668, the merged bundle) read one catalog by one rule. The merge-order concern does not bite either reader for a pair translation: a pair ships only authored translation rows (protocol getPairTranslationBundle docblock), and FileI18nAdapter lays the authored layer over every code bundle in both getTranslations (file-i18n-adapter.ts:189, the bundle the resolver reads) and t() (resolveFromLocale :367, what the evaluator asks), so an unchanged message takes the pair's authored row over the base's code-bundle translation in both. Moving the two readers onto labelCatalogs/catalogFor would wire a pair input no object can have; writing it is speculative code against an unreachable state, so no PR.", "tests": "Reading, no code change. Closure build turbo --filter=@objectstack/metadata-protocol^... under os-verify-lock VERDICT command-exit 0 (12/12 tasks). Existing pins run at 2ffdb649: vitest metadata-protocol served-override.test.ts 'names none for a type that takes no declared override' (object row) + protocol.packaged-base-refusal.test.ts 'overrides on an object is refused' under lock VERDICT command-exit 0, Test Files 2 passed (2), Tests 2 passed, 99 skipped. No ablation: nothing tested changed.", "gates": "n/a — no diff, no PR; dispatch-gates --commands derives nothing from an empty change set.", "dogfood_battery": "n/a — no PR", "mcp_calls": "2 — reads only: issue_read get x1, issue_read get_comments x1 (issue 30245). No MCP write tool.", "api_writes": "1 — post-stamped comment on issue 30245 (this report). Two REST reads via gh api GET of comments 6648581045 and 6506001053.", "open_questions": [ { "question": "Should a .custom pair's translations rank first on an EXTENDED object (objectExtensions, ADR-0146 D10), as ruling 6631250894 ranks them on a declared override? Today neither labels nor validation messages of any object take pair-first order, consistently, because no object read carries a pair input.", "options": ["A close #30245 as premise-false: the ruled order covers declared overrides, an object cannot be one, both readers agree today", "B file a product question card for pair-first order on extended objects; if ruled yes, one card moves the object read door, translateObject and the evaluator together"], "recommendation": "A, because the ruling text scopes the order to an overridden item and D10 says an object is never one; B only on a maintainer ask" } ], "security_findings": "none", "out_of_scope_findings": [ "observation, not filed, carried from the card body and not re-measured: two AUTHORED translation rows from different packages for one key and locale merge in readAuthoredTranslationLayer by item name, so a pair's authored row can lose to another authored package's row; both readers read the same layer, so refusal = served still holds. Code reading only, no door measured; no class a/b/c reach. carrier: none", "observation: the card body's premise sentence 'Once #30062 passes pairBundle, an overridden object's labels follow the ruled order' does not hold for type object; i18n-resolver.ts translateObject's labelCatalogs use is reachable only from a direct spec-level caller. carrier: seat (card disposition)" ] }objectstack-fleet commentedon Nov 5, 2026
domain:spec#2 · os-sales · session_01L6bhgVugtCCwBs98FeR27P
Release: claim 6663677926 (domain:spec#2), at 2026-11-05T16:52Z. Reason: the dev's premise reading is false; the card goes back to
pm:queuewithpm:retriagefor its disposition.premise_still_valid: false,pr: null), read atorigin/main2ffdb649.origin/main2ffdb649:packages/spec/src/kernel/customization-package.tsOVERRIDE_TYPE_CLASSIFICATIONrowobject: { disposition: 'extend_only', basis: 'adr_0146_d10_extended' };packages/metadata-protocol/src/served-override.tsservedOverrideBaseOfanswersundefinedfor a typeisOverridableMetadataTyperefuses, and its table namesobjectas "extended, never replaced (D10)". So no served object read carries a pair input, and an object's labels and itsvalidations[].messageread one catalog by one rule today: refusal text = served text holds.pm:retriage): the dev recommends (A) close as not planned, premise false: the ruled pair-first order covers declared overrides and an object cannot be one. (B), a product question whether an EXTENDED object should rank its.custompair first, only on a maintainer ask. The seat concurs with A; the disposition is triage's.Resume-from: none (nothing to resume: no diff was written; branch
claude/project-thread-8h5swncarries no commit for this card)objectstack-fleet commentedon Nov 6, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: retriage answered · to the maintainer's decision box · 2026-11-06T12:17Z
⛔ Not a claim.
维护者速读
事由： spec#2 的开发实测，本卡前提不成立。对象不会是声明式覆盖（
OVERRIDE_TYPE_CLASSIFICATION对object是extend_only），所以对象的标签和校验消息都读部署目录，两者一致，不存在卡上说的不对称（os-dev-report 6663836042）。选项：
objectExtensions扩展的对象，是否也应让.custom配对包的翻译优先？推荐：A。 现状自洽。若要 B，等于扩展 ruling 6631250894 的范围，应另立决策卡，不在本卡上改。
按是非关卡属维护者决裁，分诊不代关。已去掉
pm:retriage和pm:queue，加上needs-user-decision。Generated by Claude Code
objectstack-fleet commentedon Nov 6, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Closed as not planned · 2026-11-06T12:18Z
The maintainer ruled A in the triage session (2026-11-06, 「同意关」). The premise is false: an object is never a declared override, so its labels and its validation messages both read the deployment catalog, consistently (os-dev-report 6663836042). If an extended object should rank its
.custompair's translations first, that is a new decision card that widens ruling 6631250894.Generated by Claude Code