Release: merge development into beta

.github/workflows/code-quality.yml

@@ -302,10 +302,23 @@ jobs:
       # `check:specs` is the aggregate (json-strict + manifest + register),
       # listed as ONE leg rather than three.
       # Measured on this tree before enabling: `check:specs` PASSES (with
-      # warnings), `test:l10n` FAILS on source strings missing from
-      # l10n/en.json, e.g. "Write an audit-trail entry for every step"
-      # (src/views/settings/sections/FlowConfiguration.vue:29). A real
-      # pre-existing defect; the gate is what makes it visible.
+      # warnings), `test:l10n` FAILED on source strings missing from the English
+      # catalogue. That defect is now fixed and the leg PASSES: the gate was
+      # asserting frontend t() calls against l10n/en.json, a file no frontend
+      # code path reads, and now targets l10n/en.js.
+      #
+      # `test:l10n:parity` is listed as a SEPARATE leg from `test:l10n` because
+      # they answer different questions and fail for different reasons:
+      #   test:l10n         -> does en.js cover every t()/n() call in src/?
+      #   test:l10n:parity  -> does every FINISHED locale cover every en.js key?
+      # Without the second one, adding an English string silently leaves 17
+      # finished locales one key short and nothing notices — which is how en.js
+      # came to sit ~700 keys ahead of the locales in the first place. Locales
+      # still being translated are reported as a backlog and do not fail the
+      # build; empty values and wrong plural arity fail for every locale, because
+      # those render blank at runtime.
+      # Measured on this tree before enabling: PASSES — 17 locales at full
+      # key-for-key parity (2051 keys each), 19 in progress.
       # `test` is NOT listed: "Frontend Tests (unit)" already runs it.
       #
       # `format` (prettier --check) is listed because the shared workflow has NO
@@ -324,7 +337,16 @@ jobs:
       # renders the English source inside an otherwise translated form, silently.
       # The fleet had 30,459 such strings, so this records the current count and
       # fails only when it GROWS — burning it down stays an ordinary PR.
-      frontend-checks: '["check:specs", "test:l10n", "format", "check:schema-l10n", "check:l10n-js"]'
+      #
+      # `test:l10n:parity` is a SEPARATE leg from `test:l10n` because they answer
+      # different questions: `test:l10n` asks whether en.js covers every t()/n()
+      # call in src/, while `test:l10n:parity` asks whether every locale matches
+      # en.js key-for-key. One passing tells you nothing about the other.
+      #
+      # `check:l10n-js` is a THIRD question again: whether the generated browser
+      # catalogues (l10n/*.js) are in step with their .json sources. Both sides
+      # of this merge added one of these; neither replaces the other.
+      frontend-checks: '["check:specs", "test:l10n", "test:l10n:parity", "format", "check:schema-l10n", "check:l10n-js"]'
       # ── Cost controls (see ConductionNL/.github#596, #599) ───────────────
       #
       # The fleet's CI was not slow, it was QUEUED. A Code Quality run does

.gitignore

@@ -13,6 +13,10 @@
 
 /node_modules/
 /docs/node_modules/
+
+# Third-party hunspell dictionaries for the l10n audit spell pass. Fetched by
+# `npm run l10n:fetchdicts`; tens of MB and licensed variously, so not vendored.
+/scripts/l10n/dicts/
 /website/.docusaurus/
 /docs/build/
 /docs/.docusaurus/
@@ -127,6 +131,9 @@ tests/e2e/test-results-ci/
 tests/e2e/.mdm-seed.json
 .hydra
 
+# harvest.js writes its candidate list next to itself, one file per locale. It is
+# scratch input for a translation pass, not something a locale commit should carry.
+scripts/l10n/harvest-*.json
 # Agent/test scratch and tool caches — generated, never source.
 # Added by the 2026-08-25 fleet hygiene sweep (ADR-100 Decision 2).
 .stale/

CLAUDE.md

@@ -0,0 +1,141 @@
+# openregister — l10n
+
+App id is **`openregister`**. Every user-visible string is wrapped:
+
+```js
+t('openregister', 'Some string')
+n('openregister', 'object', 'objects', count)
+n('openregister', '{count} object', '{count} objects', count, { count })
+```
+
+## Two catalogues, different consumers
+
+| Files | Set | Read by |
+| --- | --- | --- |
+| `l10n/*.js` | **frontend** | `OC.L10N.register` → `t()` / `n()` |
+| `l10n/*.json` | **backend** | PHP `IL10N` |
+
+Not two renderings of one source — separate catalogues, separate consumers. **A `t()` call
+in `.vue`/`.js` belongs in `en.js`, never in `en.json`.** Mixing them up is the bug that
+once left `en.js` ~700 keys behind while every gate stayed green. There is no scanner for
+the backend set; it is maintained by hand.
+
+## Commands
+
+| You want to… | Run |
+| --- | --- |
+| Check / view / edit a key | `node scripts/l10n-ai.js has\|get\|find\|add\|set\|rm\|rename` |
+| Audit `en.js` vs `src/` (missing / unused / unwrapped) | `npm run check:l10n` |
+| **CI gate** — does `en.js` cover every `t()`/`n()` call? | `npm run test:l10n` |
+| **CI gate** — every locale complete and well-formed? | `npm run test:l10n:parity` |
+| Extract new keys into `en.js` | `npm run test:l10n:write` |
+| State of one locale | `npm run l10n:status -- <loc>` |
+| Write a patch (gated, dry-run by default) | `npm run l10n:apply -- <loc> patch.json [--apply]` |
+| Verify one locale before committing | `npm run l10n:selfcheck -- <loc>` |
+| See what actually renders | `npm run l10n:runtime -- <loc>` |
+| Prove the gates still refuse | `npm run l10n:gatetest -- <loc>` |
+
+Never hand-edit `l10n/*.js` — 37 files that must stay in sync, no validation, and reading
+one into context costs hundreds of tokens per call. **`apply.js` is the only writer.**
+
+## Rules
+
+**Every locale is key-for-key identical to `en.js`.** `test:l10n:parity` fails on a missing
+key, an empty value or wrong plural arity, for every locale, with no exemption list and no
+env override. Add an English string and you translate it or you unwrap it. Never add an
+exemption to get a green build.
+
+**An untranslatable string should not be wrapped at all.** An input placeholder or example
+value (`sk-...`, `myapp`, `https://example.com/webhook`) gets unwrapped in `src/` *and*
+deleted from all 37 bundles including `en.js`, in one commit. Deleting it from the locales
+alone leaves `check:l10n` reporting it unused forever.
+
+**A genuine cognate is written out and recorded.** `CSV`, `PDF`, `URL`, `Flows` in nl/de/da:
+write the value so the locale keeps parity, and record why in `locales/<loc>.json` under
+`"cognates"`. `apply.js` refuses an identical value without a record; parity fails on an
+unjustified one *and* on a stale record whose value is no longer identical. Enforcement is
+opt-in per locale, keyed on `locales/<loc>.json` existing — the gate prints which locales
+are enforced and which are merely unreviewed, so a green run is not evidence of review.
+
+Measure that split, never eyeball it: a key is untranslatable only if **no locale has ever
+carried a value differing from it**. `node scripts/l10n-ai.js get <key>` answers it.
+`value === key` without a record is the worst option available — absent falls back to
+English and stays visibly untranslated, identical is indistinguishable from finished work
+and so never gets revisited.
+
+**An `n()` call's key is NEITHER source string.** It is `"_<singular>_::_<plural>_"` —
+`pluralIdentifier` in `scripts/l10n/lib.js`. Storing forms under the bare singular renders
+for `count === 1` and falls back to English everywhere else; that shipped in all 37 bundles
+while passing every gate.
+
+**Plural arrays match that locale's own `nplurals`, and are never copied between
+languages.** An array shorter than the index the runtime asks for renders **blank** — the
+one defect you cannot see by reading the file. Equal form counts do not mean equal
+boundaries. **Read `docs/l10n-workflow.md` §7.1 before writing an array.**
+
+At runtime the form index comes from the library's own `getPlural`, not the file's
+`plural=` header: the header governs the arity gate, the library governs which element
+renders. `npm run l10n:runtime -- <loc>` is the only check that catches a wrong boundary,
+and **nothing** catches a wrong noun form.
+
+**`{plural}` is banned** — in a key, in a value, and inside a translation call, including
+the `? 's' : ''` spelling. Four gates enforce it; `l10n:gatetest` proves they refuse.
+Use `n()`. Why: `docs/l10n-workflow.md` §7.4.
+
+**A plural form must not also be a catalogue key.** `translatePlural` re-translates the form
+it selects, so such a form renders the *other* key's value, invisibly. Parity fails on it.
+
+**Never overwrite a real translation.** `l10n-ai.js` refuses without `--force`, `apply.js`
+without `--allow-replace`. Trust the refusal; replace only what is genuinely wrong, and say
+why in the commit.
+
+**But a locale pass must grammatically audit the pre-existing values and fix what is bad.**
+That rule guards against changes of taste, not against fixing grammar. No gate sees a
+wrongly-inflected value: it is not empty, not identical to English, has the right arity, and
+reads as finished work. Method in `docs/l10n-workflow.md` §6.9; what past passes found, and
+how big to expect it to be, in `docs/l10n-audit-findings.md`.
+
+**Adding a string:** `en` is required (identical to the key is correct — `en` *is* the
+source), other locales optional, `--locales=` narrows. A new English string puts every
+finished locale one key short, which parity treats as fatal — procedure in
+`docs/l10n-workflow.md` §6.15.
+
+**Commit one language at a time**, so a bad locale can be reverted alone.
+
+## Gotchas
+
+- **`clean:l10n` is a dry run on purpose.** It removes keys from all 37 files, and some
+  candidates are live UI prose nobody has wrapped yet. Cross-check against
+  `find:unwrapped`, then remove by hand, matching the whole quoted literal.
+- **Locale files are not linted or formatted, by design.** `serializeJs` emits the exact
+  Nextcloud/Transifex layout; `.prettierignore` excludes `l10n/`.
+- `l10n-ai.js rename` does not rewrite call sites — grep `src/` afterwards. `set` refuses
+  pluralized (array) keys.
+- `find:unwrapped` is deliberately high-recall (~1500 candidates). Audit by hand; do not
+  tighten the heuristic until real strings are missed.
+- **A `SKIP` or `NOTE` from `selfcheck` / `runtime-check` is not a bug to tighten away.**
+  See "Two traps in the verification scripts" in `scripts/l10n/README.md` first.
+- `scripts/l10n/lib.js` is the **origin** copy; `openconnector` vendors it. Keep them in
+  sync — the only intended divergence is `DYNAMIC_KEYS`. openconnector is currently behind:
+  old `scripts/lib/l10n.js` path, plurals stored under the bare singular, and a
+  `check-l10n-parity.js` predating the arity, identical-value and finished-set gates.
+
+## Known state
+
+`npm run test:l10n` is **red at HEAD, and not because of l10n work**: a `development` merge
+replaced the Dutch GDPR source terms with English ones and added flow strings, leaving 17
+keys used in `src/` but missing from `en.js`. That is the §6.15 procedure and its own
+commit. Everything else is green.
+
+Do not trust key or locale counts written down anywhere — `npm run test:l10n:parity` prints
+the current parity / enforced / unreviewed split, and that is the only number that cannot go
+stale.
+
+## Further reading
+
+| Doc | What |
+| --- | --- |
+| `docs/l10n-workflow.md` | The runbook: the pass in order, every gate refusal, the traps catalogue, per-locale plural data (§7.1) |
+| `scripts/l10n/README.md` | Tooling layout and what each script refuses |
+| `docs/l10n-ui-translation.md` | The non-mechanical parts: register, button conventions, per-locale decisions |
+| `docs/l10n-audit-findings.md` | What past audits actually found — defect rates, where to look first |

appinfo/info.xml

@@ -40,7 +40,7 @@ Open Register drijft apps zoals OpenCatalogi, Procest, Pipelinq en Software Cata
 
 Vrij en open source onder de EUPL-licentie.
     ]]></description>
-    <version>1.1.9-unstable.20260829214947</version>
+    <version>1.1.11-unstable.20260830083706</version>
     <licence>EUPL-1.2</licence>
     <author mail="info@conduction.nl" homepage="https://www.conduction.nl/">Conduction</author>
     <namespace>OpenRegister</namespace>

composer.json

@@ -102,7 +102,7 @@
 		"php": "^8.3",
 		"adbario/php-dot-notation": "^3.3.0",
 		"bamarni/composer-bin-plugin": "^1.8",
-		"cweagans/composer-patches": "^1.7",
+		"cweagans/composer-patches": "^2.0",
 		"ddn/sapp": "dev-feat/chained-filter-text-replace#5c406e91254d6936f44372db35f1cc15e5a06c56",
 		"dompdf/dompdf": "^3.1",
 		"dragonmantank/cron-expression": "^3.3",
@@ -124,7 +124,7 @@
 		"symfony/uid": "^6.4",
 		"symfony/workflow": "^6.4",
 		"symfony/yaml": "^6.4 || ^7.0",
-		"theodo-group/llphant": "^0.9.3",
+		"theodo-group/llphant": "^1.0",
 		"twig/twig": "^3.27.0",
 		"web-token/jwt-library": "^4.1.9",
 		"webonyx/graphql-php": "^15.0",
@@ -163,7 +163,6 @@
 		}
 	},
 	"extra": {
-		"composer-exit-on-patch-failure": true,
 		"patches": {
 			"theodo-group/llphant": {
 				"Forward think:false + keep_alive:-1 to the Ollama API (qwen3 ~5x speedup, no idle-unload wedge) \u2014 drop when LLPhant adds a per-call keep_alive/think model option": "patches/llphant-ollama-think-keepalive.patch",