fix(integrations): stop naming apps that were renamed out from under us

[rubenvdlinde]

Every DSO beschikking has been a text placeholder, not a PDF, since the docudesk rename

This is the finding to act on, so it goes first.

BeschikkingGenerationService asked the DI container for OCA\Docudesk\Service\DocumentService. Filinq is docudesk renamed, and the rename moved the PHP namespace along with the app id, so no instance has registered that class for months. The lookup threw, the throw was caught, a debug line nobody reads was written, and the method fell through to createStubBijlage(), which attaches a text placeholder and returns 'success' => true. The controller returns 201. Nothing anywhere reports a problem.

So every permit decision generated through the DSO flow since the rename has been a stub rather than the rendered PDF, and it has looked like a success to the caseworker, to the API and to the logs.

Fixing the namespace alone would not have fixed it. generateFromTemplate(), the method this service called, has never existed on filinq's DocumentService either. It would have thrown into the same catch and produced the same stub, behind a diff that read as a repair. The call moves to the published generateDocument(templateId, dataRefs, options), and attachBijlageToZaak() now reads the output.fileId and output.name that filinq actually returns instead of top-level keys that were never in that array. Generating into Files needs an owner, so the service takes IUserSession; without a session it falls back to the stub and says so.

What this PR fixes

The fleet renamed its apps (openconnector to integriq, docudesk to filinq, decidesk to decidiq, hrmq to humaniq), the directory names did not move, and several of dossiq's bindings still named an app by an id or a namespace nothing answers to.

None of them errored. Every cross-app reference here is a duck-typed runtime lookup: a stale id makes isInstalled() answer false, a stale namespace makes class_exists() answer false and container->get() throw, and each one sits behind a guard that degrades gracefully. The integrations did not break loudly, they went dark.

The other live defects

PdokService gated on isInstalled('openconnector') and read pdok.feature_flag out of the openconnector app-config namespace. IAppConfig keys on the app id, and integriq's own MigrateAppConfigKeys repair step copies that key to integriq, so the flag read permanently off and getServiceStatus() reported "not installed" about an app that was. Both resolve through the existing FleetAppId helper now.

The same method then called linkToRoute('openconnector.pdok.parcel'). That route does not exist under either name. Integriq publishes exactly four PDOK routes:

GET /api/pdok/suggest      GET /api/pdok/lookup/{id}
GET /api/pdok/free         GET /api/pdok/reverse

The stale guard is the only thing keeping the crash off: on a current instance the old id resolves false, the method returns early, and nobody reaches the route. Correcting the id alone would have turned a quiet empty list into an uncaught RouteNotFoundException, while the diff read as a fix. searchParcel() reports the gap now instead of calling an endpoint that is not there. Integriq does ship Sources\Pdok\PdokWfsSourceAdapter, which can query kadastralekaart:perceel, but it is not exposed over HTTP; closing the gap properly means integriq publishing a parcel route.

src/services/pdokService.js hit /apps/openconnector/api/pdok from the browser and 404d on every current instance. The shim reads a 404 as "the app is absent" and hands the caller an empty result, so address autocomplete simply stopped suggesting and nothing reported an error. It resolves the app segment from OC.appswebroots now, the browser-side equivalent of FleetAppId. It also asked for /lookup?id=, where integriq mounts /lookup/{id}, so the id moves into the path.

Why the tests did not catch any of it

PdokServiceTest mocked isInstalled to answer true for 'openconnector' and mocked linkToRoute to return a URL, so the suite was green against a name no instance answers to and a route that does not exist. Every BeschikkingGenerationServiceTest case made container->get() throw, so no test ever reached the generation call.

Both now assert against the names and contracts the other apps really ship, and each new assertion was mutation-checked: reverting the canonical name, reverting the method name, pinning the app-config read to a literal id, trimming the JS candidate list to one name, and moving the lookup id back to a query parameter each redden the intended assertion, and all restore green.

Also in this PR

Admin- and user-facing copy that still named openconnector, Docudesk or Decidesk (settings labels, DigiD and eHerkenning activation hints, the BRP, KvK, ZGW and ZTC adapter note values, two dialogs), and two settings links pointing at docs.procest.nl, which is not where these docs are published; docusaurus.config.js declares procest.conduction.nl and 41 other links in src/ agree.

Renaming a user-facing string renames its translation key, so all 38 l10n catalogues are updated. The keys are renamed in place, and each locale's own human translation carries over with the product name swapped, rather than extracting the new keys as English-to-English and abandoning eleven real translations. The diff is 233 added and 233 removed lines on each side, which is what a pure rename looks like: no reserialisation, no reordering. l10n/*.js is rebuilt from the JSON.

Deliberately not changed

• procest as dossiq's OWN legacy id: migration repair steps, group names, X-Procest-Signature, the SOURCE_APP literal decidiq matches on. Frozen by design.
• Dual-registered OCA\Decidesk / OCA\Decidiq event names in WorkflowListenerRegistrar, ContractDecisionDelegationService and CommitteeDelegationService. Already correct, both spellings listed newest first.
• mandaat_decidesk_connection and its decidesk-default value, and the decidesk-decisions leaf id. Stored configuration, not lookups.

Verification

lint, phpcs, phpmd, psalm, phpstan and test:all (3478 tests) each exit 0, read as $? rather than off the summary line. Full vitest 1082 passing. test:l10n, check:l10n-js, check:schema-l10n, check:manifest, eslint and prettier --check all exit 0. run-hydra-gates.sh: all 76 applicable gates green.

One check cannot run locally and CI is the instrument for it. PHPUnit's beStrictAboutCoverageMetadata plus failOnRisky only fires when a coverage driver is loaded. This machine has neither xdebug nor pcov and cannot install one, so the first push went red in all six matrix cells with OK, but there were issues! and exit 1 while the local run was honestly green. The five risky tests each named OCA\Dossiq\Support\FleetAppId as executed but undeclared; both test classes now declare it, following the precedent and the warning already written into WOORedactionServiceTest from when this happened in #1863. Please read the PHPUnit cells on this push rather than my local result for that one.

[github-actions]

Quality Report — ConductionNL/dossiq @ 49b491a

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
check-manifest		✅
check-vue3-compile		✅
test-l10n		❌
format		❌
check-schema-l10n		✅
check-l10n-js		✅
composer			✅	✅ 106/106
npm			✅	✅ 549/549
app:check-code	⏭️
info.xml	✅
REUSE	❌
PHPUnit					❌
Newman					⏭️
Playwright					❌
Hydra gates					❌

---

Quality workflow — 2026-09-09 09:33 UTC

Download the full PDF report from the workflow artifacts.

[github-actions]

Quality Report — ConductionNL/dossiq @ e97916b

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
check-manifest		✅
check-vue3-compile		✅
test-l10n		✅
format		✅
check-schema-l10n		✅
check-l10n-js		✅
composer			✅	✅ 106/106
npm			✅	✅ 549/549
app:check-code	⏭️
info.xml	✅
REUSE	❌
PHPUnit					✅
Newman					⏭️
Playwright					❌
Hydra gates					✅

---

Quality workflow — 2026-09-09 10:08 UTC

Download the full PDF report from the workflow artifacts.