Focus moves to Cancel when confirm appears (no focus-drops-to-body),
confirm area has role=group with aria-label, and Escape dismisses.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Status/error/empty messages rendered outside <ul role="listbox"> so the
element only contains role="option" children (fixes aria-required-children
axe violation). Options no longer carry tabindex or onkeydown — keyboard
navigation stays on the input per ARIA combobox pattern.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
CI caught three spots the targeted local runs missed: the relevance-path
unit tests still stubbed findAllById (the path now calls findByIdIn — the
batchMetadata stubs legitimately keep findAllById), the second
GeschichtenCard test file still expected the removed email fallback, and
the AND/OR-toggle describe lacked the wait-for-slide-transition guard its
sibling describe documents — the flake that failed run 2208.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Addresses the remaining #792 review blockers and concerns in the journey
editor cluster:
- Interlude rows show 'Zwischentext' (dedicated key), not the add-button text
- All four mutation handlers route the backend ErrorCode through
getErrorMessage (a 409 duplicate no longer says 'bitte Seite neu laden')
and console.error their failures so client-side errors leave a trace
- Remove implements the spec'd pending state: row stays dimmed with an
aria-live 'wird entfernt…' until the DELETE resolves; failure keeps the row
- Move announcements fire after the reorder resolves (no false 'verschoben')
- Touch targets ≥44px (remove ×, note links, create submit); focus moves to
the new row after add, to a sensible neighbor after remove, back to × on
confirm-cancel; drag handle is pointer-only; title/intro get aria-labels;
publish-disabled reason is a visible hint, not a title tooltip
- Amber warning styles use new --color-warning-* tokens with dark remaps
- Blocked interlude-clear restores the draft instead of showing phantom text
- useBlockDragDrop moves to $lib/shared/hooks — geschichte no longer imports
another domain's internals
- Test hardening: reorder-failure rollback (non-ok + reject), publish/
unpublish/empty-warning surface, destructive confirm path, maxlength
assertions, JourneyCreate failure path, edit-page STORY/JOURNEY branch,
fixture factory, m.* assertions, all fixed sleeps replaced with polling
67 component tests green across 6 spec files; transcription consumer of the
moved hook re-verified (30 green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The input declared role=combobox + aria-autocomplete=list while arrow keys
did nothing (WCAG 4.1.2). Wired the useTypeahead activeIndex the same way
PersonTypeahead does: ArrowUp/Down cycle, Enter/Space select, Escape closes,
aria-activedescendant tracks the active option; the list is a real listbox
with option roles again (the interim role downgrade is reverted). Zero hits
now render a 'Keine Treffer' row instead of silently vanishing, aria-expanded
matches the visible state, and the hook sets loading at setQuery so the
debounce window can't read as 'no results'. DocumentMultiSelect renders the
shared error state too. _uid counter replaced with $props.id().
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
DocumentMetadataDrawer links to /geschichten?documentId={id}, but the list
loader silently dropped the param — the user got the unfiltered list. The
loader now validates the UUID and forwards it to GET /api/geschichten,
returning it as documentIdFilter in page data.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
JourneyEditor fed GeschichteView.PersonView (no displayName) straight into
the displayName-rendering PersonMultiSelect, so every chip on a journey was
empty. The name mapping both editors need now lives once in
$lib/person/personOption.ts (with the [Unbekannt] fallback matching
GeschichteService.toView), and PersonMultiSelect/GeschichteSidebar import
the narrow PersonOption contract from there.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Captures the read-model invariant that was only living in a code comment:
views are assembled in-transaction, entities never cross the controller
boundary in the geschichte domain. CLAUDE.md §DTOs now names the exception
so the convention text stops contradicting the code.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The OWASP sanitizer entity-encodes ('&' → '&') while JourneyReader
renders the intro via Svelte text interpolation — a curator typing
'Müller & Söhne' saw 'Müller & Söhne', re-encoded cumulatively on every
editor round-trip. JOURNEY bodies now bypass the sanitizer (safe: the reader
never uses {@html}); STORY bodies keep the full allow-list sanitization.
This makes the code match the PR's documented design note.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GET /api/geschichten shipped every author's AppUser email to all readers via
GeschichteSummary.AuthorSummary — contradicting the documented rule that
author projections never expose email or group memberships. The frontend
only used it as a display-name fallback; it now falls back to [Unbekannt],
matching the server-side rule in GeschichteService.toView.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The append dedup guard was check-then-insert: two concurrent appends of the
same document both pass exists() and both insert. The partial unique index
on (geschichte_id, document_id) WHERE document_id IS NOT NULL closes the
race; append saveAndFlush-es and maps the DataIntegrityViolationException to
the same 409 JOURNEY_DOCUMENT_ALREADY_ADDED as the friendly pre-check. The
CHECK on note length pins the 2000-char contract in the schema, mirroring
chk_text_length on transcription_blocks.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GESCHICHTE_TYPE_IMMUTABLE and JOURNEY_NOTE_TOO_LONG were declared in
errors.ts, translated, and documented — but never existed in the backend.
update() now rejects a type change with 409 (omitted/same type still pass);
note length is enforced at 2000 with its own code, matching the frontend
maxlength and the i18n message (resolves the #793 discrepancy in favour of
the spec). JOURNEY_ITEM_NOT_IN_JOURNEY is deleted everywhere instead — the
deliberate 404 posture for cross-journey item ids must not leak existence
via a distinct code.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
POST /api/geschichten/{id}/items with a documentId failed 500: Spring Data
resolved the derived existsByGeschichteIdAndDocumentId path as a direct
documentId attribute (shadowed by the transient getDocumentId() getter)
instead of document.id, producing JPQL Hibernate cannot map. Existing tests
only appended note items, so the document branch was never exercised.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
With create/update returning GeschichteView, no endpoint serves the raw
Geschichte entity and springdoc drops its schema. Dashboard modules and the
home loader now use GeschichteSummary; GeschichteEditor takes GeschichteView
and maps persons into the displayName shape PersonMultiSelect renders —
fixing blank person chips on story edit. PersonMultiSelect/Sidebar narrow to
Pick<Person, 'id' | 'displayName'>, mirroring the DocumentOption precedent.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The reader dashboard renders 'bearbeitet vor X' from updatedAt, but the
summary projection dropped the field when the list endpoint moved off the
entity — drafts have no publishedAt to fall back on.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The document picker parsed the response without checking r.ok, so a 401/500
rendered identically to 'no matches' and the dropdown silently vanished —
which is how the broken relevance path shipped invisibly. The fetch now
throws on non-OK, the useTypeahead hook exposes an error flag, and the
picker renders a visible failure message (de/en/es).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
PATCH /api/geschichten/{id} (save draft, publish) returned the raw entity;
with open-in-view false, Jackson serialized the lazy items collection after
the transaction closed and every save failed with LazyInitializationException.
Write methods now assemble GeschichteView in-transaction, completing the
read-model boundary already used by GET — entities no longer cross the
controller.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The pure-text RELEVANCE fast path loaded documents via plain findAllById,
which carries no entity graph. With Document.tags LAZY (ADR-022) and no
surrounding transaction, resolveDocumentTagColors hit the dead proxy and
every q-only search (document picker typeaheads) failed with 500
LazyInitializationException. Dedicated findByIdIn declares the same fetch
shape as the other search loaders.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GeschichteUpdateDTO lacked a `type` field, so the `type: 'JOURNEY'` sent by
JourneyCreate was discarded by Jackson and every new Geschichte was saved as
STORY. The edit page branched on type, so journeys always showed the STORY
editor with no document-adding capability.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
px-2 py-1 gave ~28px height — half the WCAG 2.2 / project 44px minimum.
Changed to min-h-[44px] inline-flex items-center for both buttons.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
#6b7280 (gray-500) on #f5f4f0 yielded ~4.0:1 — below the 4.5:1 AA minimum
for normal text. Changed to #4b5563 (gray-600) which provides ~6.9:1.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
role=listbox + role=option without arrow-key navigation is misleading — the
WAI-ARIA combobox pattern requires aria-activedescendant handling that isn't
implemented. Downgraded to plain <ul>/<li>; input keeps role=combobox +
aria-controls pointing to the list id.
listboxId was a module-level constant so two simultaneous instances would share
the same DOM id. Fixed with a <script module> counter.
Updated spec queries from getByRole('option') to getByText() — tests behaviour,
not the ARIA implementation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
DocumentPickerDropdown and DocumentMultiSelect had identical createTypeahead
configs, fetch logic, and formatDocLabel helpers. Extracted to
documentTypeahead.ts; all four consumers import from the shared module.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Verifies the EXISTS subquery in findSummaries actually filters by document
membership — a journey containing the document appears, one without does not.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
track_reactivity_loss in Svelte 5 async.js fires when a $bindable write
happens while the slide transition is still being tracked asynchronously.
Waiting for the toggle to be visible ensures the transition has settled.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
aria-disabled alone leaves the button keyboard-activatable, violating
WCAG 4.1.2. Native disabled removes it from the tab order and prevents
activation via Enter/Space.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
handleNoteRemove mutated UI state optimistically without try/catch.
A failed PATCH left the note visually deleted while it survived on the
server. Now uses snapshot/rollback identical to handleNoteBlur.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
null ?? undefined evaluated to undefined, causing JSON.stringify to omit
the key entirely — the backend treated an absent note field as a no-op,
so clearing a note never persisted.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The PR removed the documentId filter from list() along with the old
Geschichte.documents ManyToMany, but the document-detail page and its
frontend server still query GET /api/geschichten?documentId=<id> to show
related stories. Without the filter the endpoint silently returned every
published story. Restores the filter through a JPQL EXISTS check on
journey_items so only journeys that include the given document are returned.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds JOURNEY_DOCUMENT_ALREADY_ADDED to ErrorCode, an
existsByGeschichteIdAndDocumentId() repo method, and a 409 guard in
JourneyItemService.append() — the error code was registered on the
frontend but never thrown on the backend, allowing concurrent tabs to
add the same document twice.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
NVDA+Chrome and VoiceOver+Safari can re-announce a persistent non-empty
aria-live region when adjacent DOM mutations occur. Clearing with a
500ms delay gives the announcement time to fire once before going quiet.
The two svelte-ignore a11y_no_static_element_interactions suppressions are
given preceding comments explaining the keyboard accessibility contract so
they are not mistaken for unaddressed tech debt.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
A persistent non-empty aria-live region can cause stale re-announcements
on adjacent DOM mutations. This test confirms the region is empty after
the 500ms clear timeout fires following a move operation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The noteError alert path (role=alert paragraph) was untested.
The catch block in handleNoteBlur was already implemented; this
test verifies it renders the alert when onNotePatch rejects.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The previous check used document.body.textContent which includes ARIA labels
and hidden elements — a match in those could misfire the indexOf comparison.
compareDocumentPosition(DOCUMENT_POSITION_FOLLOWING) checks DOM tree position
directly and is not affected by non-visible text.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace the '#753 deferred' placeholder with the actual implementation:
JOURNEY edit route now opens JourneyEditor (item list, drag/move-up/down,
add bar, sidebar) while STORY route continues to use GeschichteEditor.
Add item-mutation endpoints to the relation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Interlude is a first-class editorial concept introduced by the JourneyEditor
(#753): a note-only JourneyItem with no backing document, stored in the note
column and distinguished visually by --color-interlude-* CSS tokens.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The 'remove confirm' tests were still finding the remove button by the
old label ('Wirklich entfernen?'). After the aria-label change the
button is named 'Eintrag entfernen'; the visible confirmation text
'Wirklich entfernen?' in the DOM is unaffected.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Both error-handling sections now list JOURNEY_ITEM_NOT_IN_JOURNEY,
JOURNEY_NOTE_TOO_LONG, JOURNEY_DOCUMENT_ALREADY_ADDED, and
GESCHICHTE_TYPE_IMMUTABLE alongside the existing security codes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Removed the inline vi.unstubAllGlobals() call from the end of the
'reveals picker' test body and added it to the shared afterEach hook
so every test in the file gets the same global cleanup regardless of
which test runs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add move-up and move-down tests that verify PUT /items/reorder is called
with the swapped ID order; 50ms delay accounts for two await levels before
csrfFetch is called (click → handleMoveUp → handleReorder → csrfFetch)
- Replace vacuous 'isDirty stays false' test (was asserting a dialog that
never renders) with a meaningful publish-button-enabled assertion after
adding an item
- Update remove button query from 'Wirklich entfernen?' to 'Eintrag entfernen'
to match the new journey_remove_item_aria aria-label
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
