fix(i18n): restore en/de key parity, translate usePhotos errors, add parity guard test

.claude/agent-memory/qa-integration-tester/MEMORY.md

@@ -15,6 +15,7 @@
 
 ## Recent bug/story notes (2026-07)
 
+- [Issue #1814 — i18n parity guard + usePhotos](issue-1814-i18n-parity-guard.md) (2026-07-07) — new `client/src/i18n/i18n.parity.test.ts`: generalized 14-namespace en/de key parity + 28-file duplicate-key raw-text scanner (canonical pattern going forward); `usePhotos.test.ts` reworked to assert real translateApiError/t() output; parallel translator/frontend edits landed before my first test run — all-green on first try was real, verified via git diff
 - [Issue #1813 — formatter consolidation](issue-1813-formatter-consolidation.md) (2026-07-07) — "shadow render" LocaleProvider-wrap pattern; CODE_BUG found (BudgetSourcesPage.tsx `formatPercent` ReferenceError, real TS2552, latent/untested by pre-existing tests); client-project ts-jest does NOT catch this TS error class at test-run time (inline tsconfig ≠ full program) — always cross-check with `npx tsc -p client/tsconfig.json --noEmit`; SignatureCapture.tsx lines 113-127 are genuinely unreachable dead code (pre-existing, out of scope); canvas/jsdom mocking recipe for SignatureCapture.test.tsx (new file, closed test-parity gap)
 - [issue-1812-i18n-sweep-json-dup-keys.md](issue-1812-i18n-sweep-json-dup-keys.md) (2026-07-07) — 3 CODE_BUGs found: diary.json duplicate top-level keys (filterBar/page/detailPage) silently wiped ~25 pre-existing translations via JSON.parse last-key-wins; GanttChart.tsx crashed (ReferenceError: t is not defined — useTranslation never added despite spec); 2 wrong-key-path typos. Detection recipe + patterns documented.
 - [issue-1811-fastify-error-code-mapping.md](issue-1811-fastify-error-code-mapping.md) (2026-07-07) — FST\_\* → ErrorCode mapping test pattern (realistic body-limit/malformed-JSON routes + synthetic fallback-path errors); TOCTOU branch in davTokens.ts /profile route not independently testable, coverage-check-only per spec

.claude/agent-memory/qa-integration-tester/issue-1813-formatter-consolidation.md

@@ -24,6 +24,7 @@ function render(ui: ReactElement, options?: Parameters<typeof rtlRender>[1]) {
   return rtlRender(<LocaleProvider>{ui}</LocaleProvider>, options);
 }
 ```
+
 Works even when `ui` already contains `<MemoryRouter>` — LocaleProvider just wraps whatever is passed.
 For files using the shared `renderWithRouter` from `client/src/test/testUtils.tsx`, do NOT edit that shared util
 (16 consumers) — instead define a local `renderWithRouter` override combining `<LocaleProvider><MemoryRouter>`.
@@ -44,15 +45,15 @@ to the formatter assertion.
 `formatPercent: _formatPercent` (line 95) but the diff introduced two call sites (lines 183–184) that reference
 the bare identifier `formatPercent`, which doesn't exist in that scope. Confirmed via
 `npx tsc -p client/tsconfig.json --noEmit` (TS2552 "Cannot find name 'formatPercent'. Did you mean
-'_formatPercent'?") **and** at runtime: a test that hovers a `.segment` div (`fireEvent.mouseEnter`) to trigger
+'\_formatPercent'?") **and** at runtime: a test that hovers a `.segment` div (`fireEvent.mouseEnter`) to trigger
 the tooltip render throws `ReferenceError: formatPercent is not defined`. The bug was latent/undetected by the
 108 pre-existing tests because none of them ever hover a segment. Test written and left in place (documents
 correct expected behavior); it will pass once the fix lands (rename destructure to `formatPercent`, or use
 `_formatPercent` at the two call sites). Fix belongs to `frontend-developer`, not QA.
 
 ### ts-jest gap (client project only): does NOT catch this class of TS error at test-run time
 
-`npx jest BudgetSourcesPage.test.tsx` passed 108/108 *before* my hover test was added, despite the file having a
+`npx jest BudgetSourcesPage.test.tsx` passed 108/108 _before_ my hover test was added, despite the file having a
 real TS2552 compile error the whole time. The client project's `jest.config.ts` ts-jest transform uses an
 **inline** `tsconfig` object (`module: 'ESNext', moduleResolution: 'bundler', jsx: 'react-jsx', ...`), not a path
 to `client/tsconfig.json` — this produces isolatedModules-like behavior that does NOT do full-program
@@ -61,7 +62,7 @@ type-checking, so scope-resolution errors like "cannot find name" silently pass
 actually executes). **Always run `npx tsc -p client/tsconfig.json --noEmit`** (or the equivalent server/shared
 tsconfig) as a supplementary check when reviewing frontend-developer changes — `npx jest --coverage` alone will
 not surface this bug class unless a test happens to exercise the exact broken line. CI's separate `npm run
-typecheck` step *does* catch it (uses the real tsconfig.json), so this is CI-safe but Jest-blind locally.
+typecheck` step _does_ catch it (uses the real tsconfig.json), so this is CI-safe but Jest-blind locally.
 
 ### Discovered dead code (out of scope, not touched): SignatureCapture.tsx lines 113–127
 
@@ -75,7 +76,7 @@ Issue #1813's diff — flagged as a discovered follow-up, not fixed (QA does not
 
 ### File-size mock pattern for locale-aware Intl mocks (BackupsPage.test.tsx)
 
-When a component's `formatters.js` module is *fully* `jest.unstable_mockModule`'d (not a partial mock), and you
+When a component's `formatters.js` module is _fully_ `jest.unstable_mockModule`'d (not a partial mock), and you
 need a locale-switchable mock function (e.g. `formatFileSize`), declare a mutable `let mockFormattersLocale =
 'en-US';` **above** the `jest.unstable_mockModule(...)` call and reference it inside the factory closure — reset
 it in `beforeEach`. The factory function isn't invoked until the dynamic `import()` inside `beforeEach` runs, by
@@ -86,20 +87,21 @@ looking suspicious.
 
 The initial sweep missed 3 suites that mock `formatters.js`/`LocaleContext.js` per-file via
 `jest.unstable_mockModule` rather than through the shared `render` wrapper or `testUtils.tsx` — CI failed on:
+
 - `GanttChart.test.tsx`: `GanttHeader` (rendered by `GanttChart`) started statically importing the new
   `formatWeekdayMonthDay` as a bare top-level named export (NOT via `useFormatters()`), so the file's existing
   `jest.unstable_mockModule('../../lib/formatters.js', ...)` block needed a new top-level key added alongside
   `formatCurrency`/`formatDate`/`formatPercent` — outside the `useFormatters` mock's return object.
 - `DocumentBrowser.test.tsx` and `DiaryEntryEditPage.test.tsx`: a descendant started calling `useLocale()`
-  directly (not via `formatters.js`), so these files needed a *new* `jest.unstable_mockModule('.../LocaleContext.js', ...)`
+  directly (not via `formatters.js`), so these files needed a _new_ `jest.unstable_mockModule('.../LocaleContext.js', ...)`
   block added (mirroring the standard shape: `locale`/`resolvedLocale: 'en'`, `currency: 'EUR'`,
   `setLocale`/`syncWithServer: jest.fn()`, passthrough `LocaleProvider`) — `formatters.js` itself stayed unmocked
   in both.
 
 **Lesson**: when a component gains `useFormatters()` or a new static formatter/locale import, the
 consumer-blast-radius sweep must include every ancestor→descendant render chain, not just direct importers —
 search for `jest.unstable_mockModule('.../formatters.js'` AND `'.../LocaleContext.js'` across the whole client
-tree, and check whether the *specific new symbol* is exported at the mock's top level vs. nested inside the
+tree, and check whether the _specific new symbol_ is exported at the mock's top level vs. nested inside the
 `useFormatters` return, since a component can consume the same module both ways in different files.
 
 ### Round 2 CI fix (same PR #1845): one more grandparent consumer, closure walked to fixpoint
@@ -112,9 +114,10 @@ standard `LocaleContext.js` mock block (mirrors `DocumentBrowser.test.tsx`'s blo
 `../../contexts/LocaleContext.js` since both live two levels under `client/src/`).
 
 **Corrected process — blast-radius checks must be a transitive closure, not one grep pass:**
+
 1. `grep -rl <ComponentName> client/src --include='*.tsx' --include='*.ts' | grep -v '\.test\.' | grep -v
-   '/<ComponentName>\.tsx?$'` to get direct importers.
-2. Re-run step 1 on *each new importer found*, accumulating a visited set, until a pass finds zero new files
+'/<ComponentName>\.tsx?$'` to get direct importers.
+2. Re-run step 1 on _each new importer found_, accumulating a visited set, until a pass finds zero new files
    (fixpoint). Two hops was not enough this time — went 3 deep (`DocumentCard` → `DocumentBrowser` →
    `InvoicePaperlessPickerModal`; separately `LinkedDocumentsSection` → `HouseholdItemDetailPage`/etc. →
    nothing further).
@@ -124,7 +127,7 @@ standard `LocaleContext.js` mock block (mirrors `DocumentBrowser.test.tsx`'s blo
 4. For every file in the closure that has a `.test.tsx`, either confirm existing LocaleContext/formatters
    handling (mock block, or a real unmocked `LocaleProvider` ancestor like `App.test.tsx` uses) or actually run
    it — do not assume "probably fine" from reading the mock list only. Batching ~5 suites per `npx jest <paths...>
-   --maxWorkers=1` invocation from repo root works fine and stays within the 2-minute default timeout.
+--maxWorkers=1` invocation from repo root works fine and stays within the 2-minute default timeout.
 5. Root-level aggregators (`App.tsx`/`App.test.tsx`) are worth including explicitly since they're the
    terminal node of nearly every closure in this codebase — check them once at the end rather than per-branch.
 

.claude/agent-memory/qa-integration-tester/issue-1814-i18n-parity-guard.md

@@ -0,0 +1,49 @@
+---
+name: issue-1814-i18n-parity-guard
+description: New generalized i18n.parity.test.ts (14-namespace en/de key parity + 28-file duplicate-key guard) and usePhotos.test.ts translateApiError rework — patterns and gotchas
+metadata:
+  type: project
+---
+
+## What shipped (2026-07-07, issue #1814)
+
+- New file `client/src/i18n/i18n.parity.test.ts`: generalizes the single-namespace pattern from
+  `errorTranslation.test.ts` (`'German error JSON has the same keys as English'`) to all 14
+  registered namespaces via a hand-maintained `NAMESPACES` array, cross-checked against
+  `fs.readdirSync('client/src/i18n/en').filter(f => f.endsWith('.json')).length` so an
+  added/removed namespace file fails the test until `NAMESPACES` is updated. Also added a
+  raw-text recursive-descent duplicate-JSON-key scanner (`findDuplicateKeys`) run over all 28
+  locale files (`en/*.json` + `de/*.json`) via `fs.readFileSync` — deliberately NOT `JSON.parse`
+  since parsing silently resolves duplicates via last-key-wins (see also
+  [issue-1812-i18n-sweep-json-dup-keys.md](issue-1812-i18n-sweep-json-dup-keys.md), same failure
+  class). The parser tracks a real scope stack (not a shallow brace counter) specifically to avoid
+  false positives on `{{interpolation}}` placeholders adjacent to nested-object boundaries — ship
+  the 3 inline self-tests (genuine dupe / different-scope same-name / interpolation-adjacent) to
+  prove the parser itself before trusting it against real files.
+- `import.meta.url` + `fileURLToPath` works fine in this project's client Jest ESM setup for
+  locating `en`/`de` dirs relative to the test file (no prior precedent in the codebase for this
+  pattern in a test file, but it worked cleanly with
+  `NODE_OPTIONS=--experimental-vm-modules npx jest ...`).
+- `usePhotos.test.ts`: reworked the 4 hardcoded-string error assertions to import
+  `enErrors`/`enPhotoViewer` and assert against the real translated values (matches the
+  `translateApiError()`/`t()` pattern `usePhotos.ts` already implements), added a 5th test for the
+  unknown-code humanized fallback (`SOME_UNKNOWN_CODE` → `'Some Unknown Code'`). 100% statement
+  coverage on `usePhotos.ts` maintained (38/38 tests pass); no changes needed to hoisted mock setup.
+
+## Coordination note — parallel frontend/translator edits landed mid-session
+
+Per the spec, `usePhotos.ts` + `en/*.json` (frontend-developer) and `de/*.json`
+(translator) were being edited in parallel while I wrote tests. By the time I ran the new
+`i18n.parity.test.ts`, **both had already landed** — all 46 tests passed on the first real run
+(no known-pending-translator failures to report). Do not assume a first-try all-green result on a
+brand new parity/dupe-guard test means the test is too weak — verify by checking
+`git status --short client/src/i18n/de/` for what actually changed before concluding the guard is
+ineffective. In this case `git diff --stat` confirmed real deletions (duplicate `title`/`delete`/
+`edit`+`view` keys) and additions (`networkError`/`unexpectedError` translations) had landed.
+
+## Reusable pattern
+
+The `NAMESPACES` array + `flatten()` parity check + `findDuplicateKeys()` raw-text scanner in
+`client/src/i18n/i18n.parity.test.ts` is now the canonical generalized i18n guard for this repo —
+future new namespaces just need one line added to `NAMESPACES` (test fails loudly otherwise via
+the file-count cross-check) and don't need a new bespoke parity test.

.claude/agent-memory/ux-designer/MEMORY.md

@@ -20,4 +20,6 @@
 - `role="status"` already implies `aria-live="polite"` — never add both attributes.
 - `BadgeVariantMap` entries need both `label` and `className` — a missing `className` makes the variant's style dead on arrival.
 - New component reuse audit: check `Badge`, `SearchPicker`, `Modal`, `Skeleton`, `EmptyState`, `FormError` before proposing anything new; see [component-patterns.md](component-patterns.md) for established section-card/badge/picker conventions.
-- Cannot `gh pr review --request-changes` on your own PRs — use `--comment` and note it in the body; if `--body-file` posting fails silently, fall back to `gh api repos/.../issues/{N}/comments`.
+- Cannot `gh pr review --request-changes` **or** `--approve` on your own PRs (GraphQL rejects both, not just request-changes) — post via `gh api repos/.../issues/{N}/comments` with an explicit "Verdict: APPROVED/CHANGES_REQUIRED" line in the body instead.
+- `Intl.NumberFormat` enables thousands-grouping by default; `toFixed()`/manual string-building never did. When a PR swaps a `toFixed()` call for a locale formatter, script-verify the _old vs new_ output isn't just per-locale-correct but also identical for large values (≥1000) — grouping separators are an easy-to-miss 3rd English-output regression beyond whatever the PR explicitly discloses. See PR #1845 in [pr-review-findings.md](pr-review-findings.md).
+- `formatDateForAria`-style aria-labels that interpolate localized weekday/month into a hardcoded English sentence order (`"{weekday}, {month} {day}, {year}"`) render grammatically odd in German (`"Dienstag, Februar 24, 2026"` vs correct `"Dienstag, 24. Februar 2026"`) — this is an accepted, deferred, non-blocking gap across the app, not a new bug to flag each time.

.claude/agent-memory/ux-designer/pr-review-findings.md

@@ -43,6 +43,19 @@ Reviewed against the issue #1804 spec (see [feature-spec-history.md](feature-spe
 
 - Only finding: informational nit on `<p>` hint as a direct child of `<dl>` (HTML content-model violation) — traced back to my own spec's JSX, not an implementation deviation; non-blocking.
 
+## PR #1845 — Locale-aware display formatters (formatDate/formatPercent/formatFileSize/formatHours/formatWeekday\*/formatDateTimeWithZone) (APPROVED via gh api comment, own PR)
+
+Consolidated ~10 previously hardcoded-`'en-US'` or ad-hoc `toFixed()`/manual-string call sites in `client/src/lib/formatters.ts` behind app-locale-aware formatters (Gantt tooltips/header, calendar aria-labels, document dates, diary work-duration hours, signature timestamps, file sizes, percents). Script-verified de-DE `Intl` output for every new formatter against the PR's own test assertions (all matched) rather than eyeballing — see [pr-review-findings.md](pr-review-findings.md) process notes and the #1844 lesson below.
+
+Findings, both informational/non-blocking:
+
+- `formatPercent`'s `Intl.NumberFormat` grouping means English output for values ≥1000 gained a thousands separator (`"1,500%"` vs old `"1500%"`) that the PR's disclosed "byte-identical except two cases" claim didn't call out — a 3rd, undisclosed English-output edge case. Only affects an sr-only announcement, only at ≥10x-over-budget values. General lesson recorded in MEMORY.md: always check `toFixed()`→`Intl.NumberFormat()` swaps for grouping-separator diffs at large values, not just decimal-separator correctness.
+- `formatDateForAria` (pre-existing pattern, not introduced by this PR but newly localized here) keeps English sentence order in German output ("Dienstag, Februar 24, 2026" instead of "Dienstag, 24. Februar 2026") — accepted deferred grammar gap per task brief.
+
+Verified mechanism of the disclosed "document dates fixed off-by-one-day UTC bug": old code did `new Date(isoString).toLocaleDateString('en-US')` (parses as UTC instant, converts to browser-local zone — can roll the calendar day for non-UTC users); new `formatDate` slices `YYYY-MM-DD` and builds a local-midnight `Date` directly, avoiding the UTC→local conversion. Real fix, not a regression.
+
+Confirmed `gh pr review --approve` (not just `--request-changes`) also fails with "Can not approve your own pull request" on own PRs — same workaround as request-changes: `gh api repos/.../issues/{N}/comments`.
+
 ## Process notes
 
 - Cannot `--request-changes` on your own PRs — use `--comment` instead, and note this in the review body

client/src/components/photos/PhotoAnnotator/PhotoAnnotator.module.css

@@ -67,11 +67,6 @@
   min-height: 44px;
 }
 
-.resetButton {
-  composes: btnSecondary from '../../../styles/shared.module.css';
-  min-height: 44px;
-}
-
 /* Shared iconButton sub-pattern (dark surface — used by PhotoViewer info bar buttons) */
 .iconButton {
   display: flex;
@@ -108,18 +103,6 @@
   border-width: 0;
 }
 
-/* Modal buttons (for reset confirmation) */
-.modalActions {
-  display: flex;
-  gap: var(--spacing-3);
-  justify-content: flex-end;
-  margin-top: var(--spacing-4);
-}
-
-.confirmButton {
-  composes: btnPrimary from '../../../styles/shared.module.css';
-}
-
 .liveRegion {
   position: absolute;
   width: 1px;