chore(renovate): require manual review for privileged CI image digest bumps

Adds a packageRule matching .gitea/workflows/** digest updates with
automerge: false. Digest bumps for images running --privileged --pid=host
have root-equivalent host access and must not be auto-merged.

Addresses Nora's review concern on #537.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.gitea/workflows/ci.yml

@@ -39,6 +39,48 @@ jobs:
       - name: Run unit and component tests
         run: npm test
         working-directory: frontend
+        env:
+          TZ: Europe/Berlin
+
+      - name: Run coverage (server + client)
+        run: npm run test:coverage
+        working-directory: frontend
+        env:
+          TZ: Europe/Berlin
+
+      - name: Upload coverage reports
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: coverage-reports
+          path: frontend/coverage/
+
+      - name: Build frontend
+        run: npm run build
+        working-directory: frontend
+
+      # ── Prerender output is exactly the public help page ───────────────────
+      # SvelteKit prerender + crawl follows nav links and bakes "redirect to
+      # /login" HTML for every protected route, served BEFORE runtime hooks
+      # (see #514). With `crawl: false` only the explicit entry should land
+      # in build/prerendered/. Anything else is a regression — fail the build.
+      - name: Assert prerender output is only /hilfe/transkription
+        run: |
+          cd frontend
+          set -e
+          extra=$(find build/prerendered -type f \
+                  -not -path 'build/prerendered/hilfe/*' \
+                  -not -name '*.br' -not -name '*.gz' \
+                  || true)
+          if [ -n "$extra" ]; then
+            echo "FAIL: unexpected prerendered files (would shadow runtime hooks):"
+            echo "$extra"
+            exit 1
+          fi
+          # And the help page must still be there.
+          test -f build/prerendered/hilfe/transkription.html \
+            || { echo "FAIL: /hilfe/transkription.html missing from prerender output"; exit 1; }
+          echo "PASS: only /hilfe/transkription.html prerendered."
 
       - name: Upload screenshots
         if: always()
@@ -74,6 +116,8 @@ jobs:
     runs-on: ubuntu-latest
     env:
       DOCKER_API_VERSION: "1.43"  # NAS runner runs Docker 24.x (max API 1.43); Testcontainers 2.x defaults to 1.44
+      DOCKER_HOST: unix:///var/run/docker.sock
+      TESTCONTAINERS_RYUK_DISABLED: "true"
     steps:
       - uses: actions/checkout@v4
 
@@ -93,4 +137,123 @@ jobs:
         run: |
           chmod +x mvnw
           ./mvnw clean test
-        working-directory: backend
+        working-directory: backend
+
+  # ─── fail2ban Regex Regression ────────────────────────────────────────────────
+  # The filter parses Caddy's JSON access log; a Caddy upgrade that reorders
+  # the JSON keys would silently break it (fail2ban-regex would return
+  # "0 matches", fail2ban would stop banning, no error surface). This job
+  # pins the contract against a deterministic sample line.
+  fail2ban-regex:
+    name: fail2ban Regex
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install fail2ban
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y fail2ban
+
+      - name: Matches /api/auth/login 401
+        run: |
+          echo '{"level":"info","ts":1700000000.12,"logger":"http.log.access","msg":"handled request","request":{"remote_ip":"203.0.113.42","method":"POST","host":"archiv.raddatz.cloud","uri":"/api/auth/login"},"status":401}' > /tmp/sample.log
+          out=$(fail2ban-regex /tmp/sample.log infra/fail2ban/filter.d/familienarchiv-auth.conf)
+          echo "$out"
+          echo "$out" | grep -qE '1 matched' \
+            || { echo "expected 1 match for /api/auth/login 401"; exit 1; }
+
+      - name: Matches /api/auth/login 429
+        run: |
+          echo '{"level":"info","ts":1700000000.12,"logger":"http.log.access","msg":"handled request","request":{"remote_ip":"203.0.113.42","method":"POST","host":"archiv.raddatz.cloud","uri":"/api/auth/login"},"status":429}' > /tmp/sample.log
+          out=$(fail2ban-regex /tmp/sample.log infra/fail2ban/filter.d/familienarchiv-auth.conf)
+          echo "$out"
+          echo "$out" | grep -qE '1 matched' \
+            || { echo "expected 1 match for /api/auth/login 429"; exit 1; }
+
+      - name: Matches /api/auth/forgot-password 401
+        run: |
+          echo '{"level":"info","ts":1700000000.12,"logger":"http.log.access","msg":"handled request","request":{"remote_ip":"203.0.113.42","method":"POST","host":"archiv.raddatz.cloud","uri":"/api/auth/forgot-password"},"status":401}' > /tmp/sample.log
+          out=$(fail2ban-regex /tmp/sample.log infra/fail2ban/filter.d/familienarchiv-auth.conf)
+          echo "$out"
+          echo "$out" | grep -qE '1 matched' \
+            || { echo "expected 1 match for /api/auth/forgot-password 401"; exit 1; }
+
+      - name: Does not match /api/auth/login 200
+        run: |
+          echo '{"level":"info","ts":1700000000.12,"logger":"http.log.access","msg":"handled request","request":{"remote_ip":"203.0.113.42","method":"POST","host":"archiv.raddatz.cloud","uri":"/api/auth/login"},"status":200}' > /tmp/sample.log
+          out=$(fail2ban-regex /tmp/sample.log infra/fail2ban/filter.d/familienarchiv-auth.conf)
+          echo "$out"
+          echo "$out" | grep -qE '0 matched' \
+            || { echo "expected 0 matches for /api/auth/login 200"; exit 1; }
+
+      - name: Does not match /api/documents (unrelated 401)
+        run: |
+          echo '{"level":"info","ts":1700000000.12,"logger":"http.log.access","msg":"handled request","request":{"remote_ip":"203.0.113.42","method":"GET","host":"archiv.raddatz.cloud","uri":"/api/documents"},"status":401}' > /tmp/sample.log
+          out=$(fail2ban-regex /tmp/sample.log infra/fail2ban/filter.d/familienarchiv-auth.conf)
+          echo "$out"
+          echo "$out" | grep -qE '0 matched' \
+            || { echo "expected 0 matches for /api/documents 401"; exit 1; }
+
+      # ── Backend resolves to file-polling, not systemd ─────────────────────
+      # The Debian/Ubuntu fail2ban package ships defaults-debian.conf with
+      # `[DEFAULT] backend = systemd`. Without `backend = polling` in our
+      # jail, the daemon loads the jail but reads from journald and never
+      # touches /var/log/caddy/access.log — i.e. the regex above passes in
+      # isolation while the live jail is inert. See issue #503.
+      - name: Jail resolves with polling backend (not inherited systemd)
+        run: |
+          sudo ln -sfn "$PWD/infra/fail2ban/jail.d/familienarchiv.conf"       /etc/fail2ban/jail.d/familienarchiv.conf
+          sudo ln -sfn "$PWD/infra/fail2ban/filter.d/familienarchiv-auth.conf" /etc/fail2ban/filter.d/familienarchiv-auth.conf
+          dump=$(sudo fail2ban-client -d 2>&1)
+          echo "$dump" | grep -E "add.*familienarchiv-auth" || true
+          echo "$dump" | grep -qE "\['add', 'familienarchiv-auth', 'polling'\]" \
+            || { echo "FAIL: familienarchiv-auth jail did not resolve to 'polling' backend"; exit 1; }
+
+  # ─── Compose Bucket-Bootstrap Idempotency ─────────────────────────────────────
+  # docker-compose.prod.yml's create-buckets service runs on every
+  # `docker compose up` (one-shot, no restart). Must be idempotent — a
+  # re-deploy must not fail just because the bucket / user / policy
+  # already exists. Validated by running create-buckets twice against a
+  # throwaway minio stack and asserting both invocations exit 0.
+  compose-idempotency:
+    name: Compose Bucket Idempotency
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Write stub env file
+        run: |
+          cat > .env.test <<'EOF'
+          TAG=test
+          PORT_BACKEND=18080
+          PORT_FRONTEND=13000
+          APP_DOMAIN=localhost
+          POSTGRES_PASSWORD=stub
+          MINIO_PASSWORD=stubrootpassword
+          MINIO_APP_PASSWORD=stubapppassword
+          OCR_TRAINING_TOKEN=stub
+          APP_ADMIN_USERNAME=admin@local
+          APP_ADMIN_PASSWORD=stub
+          MAIL_HOST=mailpit
+          MAIL_PORT=1025
+          APP_MAIL_FROM=noreply@local
+          EOF
+
+      - name: Bring up minio
+        run: |
+          docker compose -f docker-compose.prod.yml -p test-idem --env-file .env.test up -d --wait minio
+
+      - name: First create-buckets run
+        run: |
+          docker compose -f docker-compose.prod.yml -p test-idem --env-file .env.test run --rm create-buckets
+
+      - name: Second create-buckets run (idempotency check)
+        run: |
+          docker compose -f docker-compose.prod.yml -p test-idem --env-file .env.test run --rm create-buckets
+
+      - name: Teardown
+        if: always()
+        run: |
+          docker compose -f docker-compose.prod.yml -p test-idem --env-file .env.test down -v
+          rm -f .env.test

.gitea/workflows/nightly.yml

@@ -0,0 +1,171 @@
+name: nightly
+
+# Builds and deploys the staging environment from main every night.
+# Runs on the self-hosted runner using Docker-out-of-Docker (the docker
+# socket is mounted in), so `docker compose build` produces images on
+# the host daemon and `docker compose up` consumes them directly — no
+# registry hop.
+#
+# Operational assumptions (see docs/DEPLOYMENT.md §3 for the full setup):
+#
+#   1. Single-tenant self-hosted runner. The "Write staging env file" step
+#      writes every secret to .env.staging on the runner filesystem; the
+#      `if: always()` cleanup step removes it. A multi-tenant runner
+#      would need to switch to docker compose --env-file <(stdin) instead.
+#
+#   2. Host docker layer cache is authoritative. There is no
+#      actions/cache; we rely on the host daemon to keep Maven and npm
+#      layers warm between runs. A `docker system prune` on the host
+#      will cause the next nightly build to be cold (5–10 min slower).
+#
+# Staging environment isolation:
+#   - project name: archiv-staging
+#   - host ports:   backend 8081, frontend 3001
+#   - profile:      staging (starts mailpit instead of a real SMTP relay)
+#
+# Required Gitea secrets:
+#   STAGING_POSTGRES_PASSWORD
+#   STAGING_MINIO_PASSWORD
+#   STAGING_MINIO_APP_PASSWORD
+#   STAGING_OCR_TRAINING_TOKEN
+#   STAGING_APP_ADMIN_USERNAME
+#   STAGING_APP_ADMIN_PASSWORD
+
+on:
+  schedule:
+    - cron: "0 2 * * *"
+  workflow_dispatch:
+
+env:
+  # Ensures the backend Dockerfile's `RUN --mount=type=cache` lines are
+  # honoured (Maven cache survives between runs).
+  DOCKER_BUILDKIT: "1"
+
+jobs:
+  deploy-staging:
+    # `ubuntu-latest` matches our self-hosted runner's advertised label
+    # (the runner has labels: ubuntu-latest / ubuntu-24.04 / ubuntu-22.04).
+    # `self-hosted` would never match — no runner advertises it — so the
+    # job parks in the queue forever. ADR-011's "single-tenant" promise
+    # is at the repo level; sharing this runner between CI and deploys
+    # for the same repo is within that boundary.
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Write staging env file
+        run: |
+          cat > .env.staging <<EOF
+          TAG=nightly
+          PORT_BACKEND=8081
+          PORT_FRONTEND=3001
+          APP_DOMAIN=staging.raddatz.cloud
+          POSTGRES_PASSWORD=${{ secrets.STAGING_POSTGRES_PASSWORD }}
+          MINIO_PASSWORD=${{ secrets.STAGING_MINIO_PASSWORD }}
+          MINIO_APP_PASSWORD=${{ secrets.STAGING_MINIO_APP_PASSWORD }}
+          OCR_TRAINING_TOKEN=${{ secrets.STAGING_OCR_TRAINING_TOKEN }}
+          APP_ADMIN_USERNAME=${{ secrets.STAGING_APP_ADMIN_USERNAME }}
+          APP_ADMIN_PASSWORD=${{ secrets.STAGING_APP_ADMIN_PASSWORD }}
+          MAIL_HOST=mailpit
+          MAIL_PORT=1025
+          MAIL_USERNAME=
+          MAIL_PASSWORD=
+          MAIL_SMTP_AUTH=false
+          MAIL_STARTTLS_ENABLE=false
+          APP_MAIL_FROM=noreply@staging.raddatz.cloud
+          EOF
+
+      - name: Build images
+        # `--pull` forces re-fetching pinned base images so a CVE
+        # re-publication of the same tag (e.g. node:20.19.0-alpine3.21,
+        # postgres:16-alpine) is picked up instead of being served
+        # from the host's stale Docker layer cache.
+        run: |
+          docker compose \
+            -f docker-compose.prod.yml \
+            -p archiv-staging \
+            --env-file .env.staging \
+            --profile staging \
+            build --pull
+
+      - name: Deploy staging
+        run: |
+          docker compose \
+            -f docker-compose.prod.yml \
+            -p archiv-staging \
+            --env-file .env.staging \
+            --profile staging \
+            up -d --wait --remove-orphans
+
+      - name: Reload Caddy
+        # Apply any committed Caddyfile changes before smoke-testing the
+        # public surface. Without this step, a Caddyfile edit lands in the
+        # repo but Caddy keeps serving the previous config until someone
+        # reloads it manually — the smoke test would then catch a stale
+        # header or a still-proxied /actuator route rather than confirming
+        # the current config is live.
+        #
+        # The runner executes job steps inside Docker containers (DooD).
+        # `systemctl` is not present in container images and cannot reach
+        # the host's systemd directly. We use the Docker socket (mounted
+        # into every job container via runner-config.yaml) to spin up a
+        # privileged sibling container in the host PID namespace; nsenter
+        # then enters the host's namespaces so systemctl talks to the real
+        # host systemd daemon. No sudoers entry is required — the Docker
+        # socket already grants root-equivalent host access.
+        #
+        # Alpine is used: ~5 MB vs ~70 MB for ubuntu, no unnecessary
+        # tooling, and the digest is pinned so any upstream change requires
+        # an explicit bump PR. util-linux (which ships nsenter) is installed
+        # at run time; apk add takes ~1 s on the warm VPS cache.
+        #
+        # `reload` not `restart`: reload sends SIGHUP so Caddy re-reads its
+        # config in-process without dropping TLS connections. `restart`
+        # would briefly stop the service, losing in-flight requests.
+        #
+        # If Caddy is not running this step fails fast before the smoke test
+        # issues a misleading "port 443 refused" error.
+        run: |
+          docker run --rm --privileged --pid=host \
+            alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d65087abc07d \
+            sh -c 'apk add --no-cache util-linux -q && nsenter -t 1 -m -u -n -p -i -- /bin/systemctl reload caddy'
+
+      - name: Smoke test deployed environment
+        # Healthchecks confirm containers are healthy; they do NOT confirm the
+        # public surface works. This step catches: Caddy not reloaded, HSTS
+        # header dropped, /actuator block bypassed.
+        #
+        # --resolve pins staging.raddatz.cloud to the runner's loopback so we
+        # do NOT depend on the host router doing hairpin NAT (many SOHO
+        # routers do not, or do so only after a firmware update). SNI still
+        # uses the public hostname so the cert validates correctly.
+        run: |
+          set -e
+          HOST="staging.raddatz.cloud"
+          URL="https://$HOST"
+          RESOLVE="--resolve $HOST:443:127.0.0.1"
+          echo "Smoke test: $URL (pinned to 127.0.0.1)"
+          curl -fsS $RESOLVE --max-time 10 "$URL/login" -o /dev/null
+          # Pin the preload-list-eligible HSTS value, not just header presence:
+          # a degraded `max-age=1` or a dropped `includeSubDomains; preload` must
+          # fail this check rather than pass it silently.
+          curl -fsS $RESOLVE --max-time 10 -I "$URL/" \
+            | grep -Eqi 'strict-transport-security:[[:space:]]*max-age=31536000.*includeSubDomains.*preload'
+          # Permissions-Policy denies APIs the app does not use (camera,
+          # microphone, geolocation). A regression that loosens or drops the
+          # header now fails the smoke step.
+          curl -fsS $RESOLVE --max-time 10 -I "$URL/" \
+            | grep -Eqi 'permissions-policy:[[:space:]]*camera=\(\),[[:space:]]*microphone=\(\),[[:space:]]*geolocation=\(\)'
+          status=$(curl -s $RESOLVE -o /dev/null -w "%{http_code}" --max-time 10 "$URL/actuator/health")
+          [ "$status" = "404" ] || { echo "expected 404 from /actuator/health, got $status"; exit 1; }
+          echo "All smoke checks passed"
+
+      - name: Cleanup env file
+        # LOAD-BEARING: `if: always()` is the linchpin of the ADR-011
+        # single-tenant runner trust model. Every secret in .env.staging
+        # is plain text on the runner filesystem until this step runs.
+        # If a future refactor drops `if: always()`, a failed deploy
+        # leaves the env-file behind. Do not remove this conditional
+        # without first re-evaluating ADR-011.
+        if: always()
+        run: rm -f .env.staging

.gitea/workflows/release.yml

@@ -0,0 +1,140 @@
+name: release
+
+# Builds and deploys the production environment on `v*` tag push.
+# Runs on the self-hosted runner via Docker-out-of-Docker; images are
+# tagged with the actual git tag (e.g. v1.0.0) so rollback is
+#   `TAG=<previous> docker compose -f docker-compose.prod.yml -p archiv-production up -d --wait`
+#
+# Operational assumptions (see docs/DEPLOYMENT.md §3 for the full setup):
+#
+#   1. Single-tenant self-hosted runner. The "Write production env file"
+#      step writes every secret to .env.production on the runner
+#      filesystem; the `if: always()` cleanup step removes it. A
+#      multi-tenant runner would need to switch to
+#      `docker compose --env-file <(stdin)` instead.
+#
+#   2. Host docker layer cache is authoritative. There is no
+#      actions/cache; we rely on the host daemon to keep Maven and npm
+#      layers warm between runs. A `docker system prune` on the host
+#      will cause the next release build to be cold (5–10 min slower).
+#
+# Production environment:
+#   - project name: archiv-production
+#   - host ports:   backend 8080, frontend 3000
+#   - profile:      (none) — mailpit is excluded; real SMTP relay is used
+#
+# Required Gitea secrets:
+#   PROD_POSTGRES_PASSWORD
+#   PROD_MINIO_PASSWORD
+#   PROD_MINIO_APP_PASSWORD
+#   PROD_OCR_TRAINING_TOKEN
+#   PROD_APP_ADMIN_USERNAME       (CRITICAL: see docs/DEPLOYMENT.md)
+#   PROD_APP_ADMIN_PASSWORD       (CRITICAL: locked in on first deploy)
+#   MAIL_HOST
+#   MAIL_PORT
+#   MAIL_USERNAME
+#   MAIL_PASSWORD
+
+on:
+  push:
+    tags:
+      - "v*"
+
+env:
+  DOCKER_BUILDKIT: "1"
+
+jobs:
+  deploy-production:
+    # See nightly.yml — same rationale: `ubuntu-latest` matches the
+    # advertised label of our single-tenant self-hosted runner.
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Write production env file
+        run: |
+          cat > .env.production <<EOF
+          TAG=${{ gitea.ref_name }}
+          PORT_BACKEND=8080
+          PORT_FRONTEND=3000
+          APP_DOMAIN=archiv.raddatz.cloud
+          POSTGRES_PASSWORD=${{ secrets.PROD_POSTGRES_PASSWORD }}
+          MINIO_PASSWORD=${{ secrets.PROD_MINIO_PASSWORD }}
+          MINIO_APP_PASSWORD=${{ secrets.PROD_MINIO_APP_PASSWORD }}
+          OCR_TRAINING_TOKEN=${{ secrets.PROD_OCR_TRAINING_TOKEN }}
+          APP_ADMIN_USERNAME=${{ secrets.PROD_APP_ADMIN_USERNAME }}
+          APP_ADMIN_PASSWORD=${{ secrets.PROD_APP_ADMIN_PASSWORD }}
+          MAIL_HOST=${{ secrets.MAIL_HOST }}
+          MAIL_PORT=${{ secrets.MAIL_PORT }}
+          MAIL_USERNAME=${{ secrets.MAIL_USERNAME }}
+          MAIL_PASSWORD=${{ secrets.MAIL_PASSWORD }}
+          MAIL_SMTP_AUTH=true
+          MAIL_STARTTLS_ENABLE=true
+          APP_MAIL_FROM=noreply@raddatz.cloud
+          EOF
+
+      - name: Build images
+        # `--pull` forces re-fetching pinned base images so a CVE
+        # re-publication of the same tag is picked up rather than served
+        # from the host's stale Docker layer cache.
+        run: |
+          docker compose \
+            -f docker-compose.prod.yml \
+            -p archiv-production \
+            --env-file .env.production \
+            build --pull
+
+      - name: Deploy production
+        run: |
+          docker compose \
+            -f docker-compose.prod.yml \
+            -p archiv-production \
+            --env-file .env.production \
+            up -d --wait --remove-orphans
+
+      - name: Reload Caddy
+        # See nightly.yml — same rationale and mechanism: DooD job containers
+        # cannot call systemctl directly; nsenter via a privileged sibling
+        # container reaches the host systemd. Must run after deploy (so the
+        # latest Caddyfile is on disk) and before the smoke test (so the
+        # public surface reflects the current config). Alpine with pinned
+        # digest; reload not restart — see nightly.yml for full rationale.
+        run: |
+          docker run --rm --privileged --pid=host \
+            alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d65087abc07d \
+            sh -c 'apk add --no-cache util-linux -q && nsenter -t 1 -m -u -n -p -i -- /bin/systemctl reload caddy'
+
+      - name: Smoke test deployed environment
+        # See nightly.yml — same three checks, against the prod vhost.
+        # --resolve pins archiv.raddatz.cloud to the runner's loopback so
+        # the smoke test does NOT depend on hairpin NAT on the host router.
+        run: |
+          set -e
+          HOST="archiv.raddatz.cloud"
+          URL="https://$HOST"
+          RESOLVE="--resolve $HOST:443:127.0.0.1"
+          echo "Smoke test: $URL (pinned to 127.0.0.1)"
+          curl -fsS $RESOLVE --max-time 10 "$URL/login" -o /dev/null
+          # Pin the preload-list-eligible HSTS value, not just header presence:
+          # a degraded `max-age=1` or a dropped `includeSubDomains; preload` must
+          # fail this check rather than pass it silently.
+          curl -fsS $RESOLVE --max-time 10 -I "$URL/" \
+            | grep -Eqi 'strict-transport-security:[[:space:]]*max-age=31536000.*includeSubDomains.*preload'
+          # Permissions-Policy denies APIs the app does not use (camera,
+          # microphone, geolocation). A regression that loosens or drops the
+          # header now fails the smoke step.
+          curl -fsS $RESOLVE --max-time 10 -I "$URL/" \
+            | grep -Eqi 'permissions-policy:[[:space:]]*camera=\(\),[[:space:]]*microphone=\(\),[[:space:]]*geolocation=\(\)'
+          status=$(curl -s $RESOLVE -o /dev/null -w "%{http_code}" --max-time 10 "$URL/actuator/health")
+          [ "$status" = "404" ] || { echo "expected 404 from /actuator/health, got $status"; exit 1; }
+          echo "All smoke checks passed"
+
+      - name: Cleanup env file
+        # LOAD-BEARING: `if: always()` is the linchpin of the ADR-011
+        # single-tenant runner trust model. Every secret in
+        # .env.production is plain text on the runner filesystem until
+        # this step runs. If a future refactor drops `if: always()`, a
+        # failed deploy leaves the env-file behind. Do not remove this
+        # conditional without first re-evaluating ADR-011.
+        if: always()
+        run: rm -f .env.production

backend/src/main/java/org/raddatz/familienarchiv/document/DocumentRepository.java

@@ -100,7 +100,45 @@ public interface DocumentRepository extends JpaRepository<Document, UUID>, JpaSp
             ORDER BY ts_rank(d.search_vector, q.pq) DESC,
                      d.meta_date DESC NULLS LAST
             """)
-    List<UUID> findRankedIdsByFts(@Param("query") String query);
+    // Unpaged path — for bulk-edit "select all" and density chart
+    List<UUID> findAllMatchingIdsByFts(@Param("query") String query);
+
+    /**
+     * Returns one page of FTS-ranked document IDs with the total match count.
+     *
+     * <p>Each row contains (in column order):
+     * <ol>
+     *   <li>UUID   — document id</li>
+     *   <li>double — ts_rank score</li>
+     *   <li>long   — COUNT(*) OVER () — full match count, not page count</li>
+     * </ol>
+     *
+     * <p>Returns an empty list when the query matches no documents (including
+     * stopword-only queries where websearch_to_tsquery returns an empty tsquery).
+     * Use findAllMatchingIdsByFts for the unpaged bulk-edit path.
+     */
+    @Query(nativeQuery = true, value = """
+            WITH q AS (
+                SELECT CASE WHEN websearch_to_tsquery('german', :query)::text <> ''
+                            THEN to_tsquery('simple', regexp_replace(
+                                     websearch_to_tsquery('german', :query)::text,
+                                     '''([^'']+)''',
+                                     '''\\1'':*',
+                                     'g'))
+                       END AS pq
+            ), matches AS (
+                SELECT d.id, ts_rank(d.search_vector, q.pq) AS rank
+                FROM documents d, q
+                WHERE d.search_vector @@ q.pq
+            )
+            SELECT id, rank, COUNT(*) OVER () AS total
+            FROM matches
+            ORDER BY rank DESC, id
+            OFFSET :offset LIMIT :limit
+            """)
+    List<Object[]> findFtsPageRaw(@Param("query") String query,
+                                  @Param("offset") int offset,
+                                  @Param("limit") int limit);
 
     /**
      * Returns match-enrichment data for a set of documents identified by their IDs.

backend/src/main/java/org/raddatz/familienarchiv/document/DocumentService.java

@@ -162,7 +162,7 @@ public class DocumentService {
      */
     private List<UUID> resolveFtsIds(String text) {
         if (!StringUtils.hasText(text)) return null;
-        return documentRepository.findRankedIdsByFts(text);
+        return documentRepository.findAllMatchingIdsByFts(text);
     }
 
     /** Loads matching documents and projects to non-null {@link LocalDate}s. */
@@ -485,7 +485,7 @@ public class DocumentService {
         boolean hasText = StringUtils.hasText(text);
         List<UUID> rankedIds = null;
         if (hasText) {
-            rankedIds = documentRepository.findRankedIdsByFts(text);
+            rankedIds = documentRepository.findAllMatchingIdsByFts(text);
             if (rankedIds.isEmpty()) return List.of();
         }
 
@@ -645,39 +645,43 @@ public class DocumentService {
     // 1. Allgemeine Suche (für das Suchfeld im Frontend)
     public DocumentSearchResult searchDocuments(String text, LocalDate from, LocalDate to, UUID sender, UUID receiver, List<String> tags, String tagQ, DocumentStatus status, DocumentSort sort, String dir, TagOperator tagOperator, Pageable pageable) {
         boolean hasText = StringUtils.hasText(text);
-        List<UUID> rankedIds = null;
 
+        // Pure-text RELEVANCE: push pagination into SQL — skip findAllMatchingIdsByFts entirely (ADR-008).
+        if (isPureTextRelevance(hasText, sort, from, to, sender, receiver, tags, tagQ, status)) {
+            return relevanceSortedPageFromSql(text, pageable);
+        }
+
+        List<UUID> rankedIds = null;
         if (hasText) {
-            rankedIds = documentRepository.findRankedIdsByFts(text);
+            rankedIds = documentRepository.findAllMatchingIdsByFts(text);
             if (rankedIds.isEmpty()) return DocumentSearchResult.of(List.of());
         }
 
         Specification<Document> spec = buildSearchSpec(
                 hasText, rankedIds, from, to, sender, receiver, tags, tagQ, status, tagOperator);
 
-        // SENDER, RECEIVER and RELEVANCE sorts load the full match set and slice in memory.
+        // SENDER and RECEIVER sorts load the full match set and slice in-memory.
         // JPA's Sort.by("sender.lastName") generates an INNER JOIN that silently drops
-        // documents with null sender/receivers; RELEVANCE maps a DB order to an external
-        // rank list. Cost scales linearly with match count — acceptable while documents
-        // stays under ~10k rows. Past that, replace with SQL-level LEFT JOIN sort.
+        // documents with null sender/receivers. Cost scales with match count —
+        // acceptable while documents stays under ~10k rows. (ADR-008)
         if (sort == DocumentSort.RECEIVER) {
+            // In-memory sort on page slice (≤ page size rows) — acceptable
             List<Document> sorted = sortByFirstReceiver(documentRepository.findAll(spec), dir);
             return buildResultPaged(pageSlice(sorted, pageable), text, pageable, sorted.size());
         }
         if (sort == DocumentSort.SENDER) {
+            // In-memory sort on page slice (≤ page size rows) — acceptable
             List<Document> sorted = sortBySender(documentRepository.findAll(spec), dir);
             return buildResultPaged(pageSlice(sorted, pageable), text, pageable, sorted.size());
         }
 
-        // RELEVANCE: default when text present and no explicit sort given
+        // RELEVANCE with active filters: load filtered subset and sort in-memory by rank.
         boolean useRankOrder = hasText && (sort == null || sort == DocumentSort.RELEVANCE);
         if (useRankOrder) {
-            List<Document> results = documentRepository.findAll(spec);
             Map<UUID, Integer> rankMap = new HashMap<>();
             for (int i = 0; i < rankedIds.size(); i++) rankMap.put(rankedIds.get(i), i);
-            List<Document> sorted = results.stream()
-                    .sorted(Comparator.comparingInt(
-                            doc -> rankMap.getOrDefault(doc.getId(), Integer.MAX_VALUE)))
+            List<Document> sorted = documentRepository.findAll(spec).stream()
+                    .sorted(Comparator.comparingInt(doc -> rankMap.getOrDefault(doc.getId(), Integer.MAX_VALUE)))
                     .toList();
             return buildResultPaged(pageSlice(sorted, pageable), text, pageable, sorted.size());
         }
@@ -688,6 +692,39 @@ public class DocumentService {
         return buildResultPaged(page.getContent(), text, pageable, page.getTotalElements());
     }
 
+    private static boolean isPureTextRelevance(boolean hasText, DocumentSort sort,
+            LocalDate from, LocalDate to, UUID sender, UUID receiver,
+            List<String> tags, String tagQ, DocumentStatus status) {
+        return hasText && (sort == null || sort == DocumentSort.RELEVANCE)
+                && from == null && to == null && sender == null && receiver == null
+                && (tags == null || tags.isEmpty()) && (tagQ == null || tagQ.isBlank()) && status == null;
+    }
+
+    /**
+     * Pure-text RELEVANCE path — pagination and ts_rank ordering pushed into SQL.
+     * Called when no non-text filters are active (ADR-008).
+     */
+    private DocumentSearchResult relevanceSortedPageFromSql(String text, Pageable pageable) {
+        long rawOffset = pageable.getOffset();
+        if (rawOffset > Integer.MAX_VALUE) return DocumentSearchResult.of(List.of());
+        int offset = (int) rawOffset;
+        int limit = pageable.getPageSize();
+        FtsPage ftsPage = toFtsPage(documentRepository.findFtsPageRaw(text, offset, limit));
+        if (ftsPage.hits().isEmpty()) return DocumentSearchResult.of(List.of());
+
+        // Preserve ts_rank order from SQL across the JPA findAllById call.
+        Map<UUID, Integer> rankMap = new HashMap<>();
+        List<UUID> pageIds = new ArrayList<>();
+        for (int i = 0; i < ftsPage.hits().size(); i++) {
+            rankMap.put(ftsPage.hits().get(i).id(), i);
+            pageIds.add(ftsPage.hits().get(i).id());
+        }
+        List<Document> docs = documentRepository.findAllById(pageIds).stream()
+                .sorted(Comparator.comparingInt(d -> rankMap.getOrDefault(d.getId(), Integer.MAX_VALUE)))
+                .toList();
+        return buildResultPaged(docs, text, pageable, ftsPage.total());
+    }
+
     private static <T> List<T> pageSlice(List<T> sorted, Pageable pageable) {
         int from = Math.min((int) pageable.getOffset(), sorted.size());
         int to = Math.min(from + pageable.getPageSize(), sorted.size());
@@ -1013,6 +1050,28 @@ public class DocumentService {
         return result;
     }
 
+    private static final int COL_ID = 0;
+    private static final int COL_RANK = 1;
+    private static final int COL_TOTAL = 2;
+
+    /**
+     * Maps raw Object[] rows from {@link DocumentRepository#findFtsPageRaw} to an
+     * {@link FtsPage}. Uses pattern-matching UUID cast to guard against driver-level
+     * type variance (some JDBC drivers return UUID as String).
+     */
+    private static FtsPage toFtsPage(List<Object[]> rows) {
+        if (rows.isEmpty()) return new FtsPage(List.of(), 0);
+        long total = ((Number) rows.get(0)[COL_TOTAL]).longValue();
+        List<FtsHit> hits = rows.stream()
+                .map(r -> {
+                    UUID id = r[COL_ID] instanceof UUID u ? u : UUID.fromString(r[COL_ID].toString());
+                    double rank = ((Number) r[COL_RANK]).doubleValue();
+                    return new FtsHit(id, rank);
+                })
+                .toList();
+        return new FtsPage(hits, total);
+    }
+
     /** Clean text + highlight offsets parsed from a {@code ts_headline} sentinel-delimited string. */
     public record ParsedHighlight(String cleanText, List<MatchOffset> offsets) {}
 

backend/src/main/java/org/raddatz/familienarchiv/document/FtsHit.java

@@ -0,0 +1,6 @@
+package org.raddatz.familienarchiv.document;
+
+import java.util.UUID;
+
+/** A single document hit from a paginated FTS query — id and its ts_rank score. */
+record FtsHit(UUID id, double rank) {}

backend/src/main/java/org/raddatz/familienarchiv/document/FtsPage.java

@@ -0,0 +1,6 @@
+package org.raddatz.familienarchiv.document;
+
+import java.util.List;
+
+/** One page of FTS results — the ranked hit list for this page and the total match count. */
+record FtsPage(List<FtsHit> hits, long total) {}

backend/src/main/java/org/raddatz/familienarchiv/document/comment/CommentController.java

@@ -27,7 +27,9 @@ public class CommentController {
     // ─── Block (transcription) comments ────────────────────────────────────────
 
     @GetMapping("/api/documents/{documentId}/transcription-blocks/{blockId}/comments")
-    public List<DocumentComment> getBlockComments(@PathVariable UUID blockId) {
+    public List<DocumentComment> getBlockComments(
+            @PathVariable UUID documentId,
+            @PathVariable UUID blockId) {
         return commentService.getCommentsForBlock(blockId);
     }
 
@@ -48,6 +50,7 @@ public class CommentController {
     @RequirePermission({Permission.ANNOTATE_ALL, Permission.WRITE_ALL})
     public DocumentComment replyToBlockComment(
             @PathVariable UUID documentId,
+            @PathVariable UUID blockId,
             @PathVariable UUID commentId,
             @RequestBody CreateCommentDTO dto,
             Authentication authentication) {

backend/src/main/java/org/raddatz/familienarchiv/security/AuthTokenCookieFilter.java

@@ -0,0 +1,137 @@
+package org.raddatz.familienarchiv.security;
+
+import jakarta.servlet.FilterChain;
+import jakarta.servlet.ServletException;
+import jakarta.servlet.http.Cookie;
+import jakarta.servlet.http.HttpServletRequest;
+import jakarta.servlet.http.HttpServletRequestWrapper;
+import jakarta.servlet.http.HttpServletResponse;
+import org.springframework.core.annotation.Order;
+import org.springframework.http.HttpHeaders;
+import org.springframework.stereotype.Component;
+import org.springframework.web.filter.OncePerRequestFilter;
+
+import java.io.IOException;
+import java.net.URLDecoder;
+import java.nio.charset.StandardCharsets;
+import java.util.Collections;
+import java.util.Enumeration;
+
+/**
+ * Promotes the {@code auth_token} cookie to an {@code Authorization} header
+ * so that browser-side requests to {@code /api/*} authenticate the same way
+ * SSR fetches do.
+ *
+ * <p>The SvelteKit login action stores the full HTTP Basic header value
+ * ({@code "Basic <base64>"}) in an HttpOnly cookie. SSR fetches from
+ * {@code hooks.server.ts} read the cookie and pass it explicitly as the
+ * {@code Authorization} header. In the dev environment, Vite's proxy does
+ * the same on every {@code /api/*} request (see {@code vite.config.ts}).
+ * In production, Caddy proxies {@code /api/*} straight to the backend and
+ * does NOT translate the cookie — so client-side {@code fetch} and
+ * {@code EventSource} calls reach the backend without auth, get
+ * {@code 401 WWW-Authenticate: Basic}, and the browser pops a native dialog.
+ *
+ * <p>This filter closes that gap: if a request has an {@code auth_token}
+ * cookie but no explicit {@code Authorization} header, promote the cookie
+ * value (URL-decoded) into the header before Spring Security inspects it.
+ * Explicit {@code Authorization} headers are preserved unchanged.
+ *
+ * <p>See #520. Filter runs at {@code Ordered.HIGHEST_PRECEDENCE} so it
+ * mutates the request before any Spring Security filter sees it.
+ *
+ * <p><b>Scope:</b> only {@code /api/*} requests are touched. The
+ * {@code /actuator/*} block in Caddy plus the open auth/reset paths in
+ * {@link SecurityConfig} must NOT receive a promoted Authorization.
+ *
+ * <p><b>⚠ Log-leakage warning:</b> the wrapped request exposes the
+ * Authorization header via {@code getHeaderNames}/{@code getHeaders}. Any
+ * filter or interceptor that iterates request headers will see the live
+ * Basic credential. Do NOT add a request-header logger downstream of this
+ * filter without explicitly scrubbing the {@code Authorization} field.
+ */
+@Component
+@Order(org.springframework.core.Ordered.HIGHEST_PRECEDENCE)
+public class AuthTokenCookieFilter extends OncePerRequestFilter {
+
+    static final String COOKIE_NAME = "auth_token";
+    static final String SCOPE_PREFIX = "/api/";
+
+    @Override
+    protected void doFilterInternal(HttpServletRequest request,
+                                    HttpServletResponse response,
+                                    FilterChain chain) throws ServletException, IOException {
+        // Scope: only /api/* needs cookie promotion. /actuator/health (open),
+        // /api/auth/forgot-password (open), /login etc. don't.
+        if (!request.getRequestURI().startsWith(SCOPE_PREFIX)) {
+            chain.doFilter(request, response);
+            return;
+        }
+        // An explicit Authorization header wins — this is the SSR fetch path
+        // (hooks.server.ts builds the header itself).
+        if (request.getHeader(HttpHeaders.AUTHORIZATION) != null) {
+            chain.doFilter(request, response);
+            return;
+        }
+        Cookie[] cookies = request.getCookies();
+        if (cookies == null) {
+            chain.doFilter(request, response);
+            return;
+        }
+        for (Cookie c : cookies) {
+            if (COOKIE_NAME.equals(c.getName()) && c.getValue() != null && !c.getValue().isBlank()) {
+                String decoded;
+                try {
+                    decoded = URLDecoder.decode(c.getValue(), StandardCharsets.UTF_8);
+                } catch (IllegalArgumentException malformed) {
+                    // Malformed percent-encoding — refuse to forward a bogus
+                    // Authorization header. Spring Security will treat the
+                    // request as unauthenticated.
+                    chain.doFilter(request, response);
+                    return;
+                }
+                chain.doFilter(new AuthHeaderRequest(request, decoded), response);
+                return;
+            }
+        }
+        chain.doFilter(request, response);
+    }
+
+    /**
+     * Adds (or overrides) the {@code Authorization} header on a wrapped request.
+     * All other headers pass through unchanged.
+     */
+    static final class AuthHeaderRequest extends HttpServletRequestWrapper {
+        private final String authorization;
+
+        AuthHeaderRequest(HttpServletRequest request, String authorization) {
+            super(request);
+            this.authorization = authorization;
+        }
+
+        @Override
+        public String getHeader(String name) {
+            if (HttpHeaders.AUTHORIZATION.equalsIgnoreCase(name)) {
+                return authorization;
+            }
+            return super.getHeader(name);
+        }
+
+        @Override
+        public Enumeration<String> getHeaders(String name) {
+            if (HttpHeaders.AUTHORIZATION.equalsIgnoreCase(name)) {
+                return Collections.enumeration(Collections.singletonList(authorization));
+            }
+            return super.getHeaders(name);
+        }
+
+        @Override
+        public Enumeration<String> getHeaderNames() {
+            Enumeration<String> base = super.getHeaderNames();
+            java.util.Set<String> names = new java.util.LinkedHashSet<>();
+            while (base.hasMoreElements()) names.add(base.nextElement());
+            names.add(HttpHeaders.AUTHORIZATION);
+            return Collections.enumeration(names);
+        }
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/security/SecurityConfig.java

@@ -37,12 +37,20 @@ public class SecurityConfig {
     @Bean
     public SecurityFilterChain securityFilterChain(HttpSecurity http) throws Exception {
         http
-                // CSRF is intentionally disabled: every request from the SvelteKit frontend
-                // carries an explicit Authorization header (Basic Auth token injected by
-                // hooks.server.ts). Browsers block cross-origin requests from setting custom
-                // headers, so cross-site request forgery via a third-party page is not
-                // possible with this auth scheme. If the auth model ever changes to
-                // cookie-based sessions, CSRF protection must be re-enabled.
+                // CSRF is intentionally disabled. With the cookie-promotion model
+                // (auth_token cookie → Authorization header via AuthTokenCookieFilter,
+                // see #520), every authenticated request to /api/* now carries the
+                // credential automatically once the cookie is set. The CSRF defence
+                // for state-changing endpoints is therefore LOAD-BEARING on:
+                //
+                //   1. SameSite=strict on the auth_token cookie (login/+page.server.ts).
+                //      A cross-site POST from evil.com cannot include the cookie.
+                //   2. CORS — Spring's default rejects cross-origin requests with
+                //      credentials unless explicitly allowed (no allowedOrigins config).
+                //
+                // If either of those is ever weakened (e.g. cookie flipped to
+                // SameSite=lax, CORS allowedOrigins expanded), CSRF protection
+                // MUST be re-enabled here.
                 .csrf(csrf -> csrf.disable())
 
                 .authorizeHttpRequests(auth -> {

backend/src/main/java/org/raddatz/familienarchiv/user/AppUser.java

@@ -88,7 +88,8 @@ public class AppUser {
     };
 
     public static String computeColor(UUID id) {
-        return PALETTE[Math.abs(id.hashCode()) % PALETTE.length];
+        // Math.floorMod avoids the Integer.MIN_VALUE overflow trap in Math.abs(hashCode())
+        return PALETTE[Math.floorMod(id.hashCode(), PALETTE.length)];
     }
 
     @PrePersist

backend/src/main/java/org/raddatz/familienarchiv/user/UserDataInitializer.java

@@ -20,6 +20,7 @@ import org.springframework.boot.CommandLineRunner;
 import org.springframework.context.annotation.Bean;
 import org.springframework.context.annotation.Configuration;
 import org.springframework.context.annotation.Profile;
+import org.springframework.core.env.Environment;
 import org.springframework.security.crypto.password.PasswordEncoder;
 
 import java.time.LocalDate;
@@ -31,26 +32,51 @@ import java.util.Set;
 @DependsOn("flyway")
 public class UserDataInitializer {
 
-    @Value("${app.admin.email:admin@familyarchive.local}")
+    static final String DEFAULT_ADMIN_EMAIL = "admin@familienarchiv.local";
+    static final String DEFAULT_ADMIN_PASSWORD = "admin123";
+
+    @Value("${app.admin.email:" + DEFAULT_ADMIN_EMAIL + "}")
     private String adminEmail;
 
-    @Value("${app.admin.password:admin123}")
+    @Value("${app.admin.password:" + DEFAULT_ADMIN_PASSWORD + "}")
     private String adminPassword;
 
     private final AppUserRepository userRepository;
     private final UserGroupRepository groupRepository;
+    private final Environment environment;
 
     @Bean
     public CommandLineRunner initAdminUser(PasswordEncoder passwordEncoder) {
         return args -> {
             if (userRepository.findByEmail(adminEmail).isEmpty()) {
+                // Fail-closed in production: refuse to seed with the well-known
+                // defaults. Otherwise an operator who forgets APP_ADMIN_USERNAME
+                // / APP_ADMIN_PASSWORD locks production to admin@…/admin123 PERMANENTLY
+                // (UserDataInitializer only seeds when the row is missing — see #513).
+                // Allowed in dev/test/e2e because those run without secrets configured.
+                boolean isLocalProfile = environment.matchesProfiles("dev", "test", "e2e");
+                if (!isLocalProfile
+                    && (DEFAULT_ADMIN_EMAIL.equals(adminEmail)
+                        || DEFAULT_ADMIN_PASSWORD.equals(adminPassword))) {
+                    throw new IllegalStateException(
+                        "Refusing to seed admin user with default credentials outside "
+                        + "the dev/test/e2e profiles. Set APP_ADMIN_USERNAME and "
+                        + "APP_ADMIN_PASSWORD to non-default values before first boot — "
+                        + "this lock-in is permanent."
+                    );
+                }
                 log.info("Kein Admin-User '{}' gefunden. Erstelle Default-Admin...", adminEmail);
 
-                UserGroup adminGroup = UserGroup.builder()
-                        .name("Administrators")
-                        .permissions(Set.of("ADMIN", "READ_ALL", "WRITE_ALL", "ANNOTATE_ALL", "ADMIN_USER", "ADMIN_TAG", "ADMIN_PERMISSION"))
-                        .build();
-                groupRepository.save(adminGroup);
+                // Reuse the Administrators group if it already exists (e.g. a
+                // previous boot seeded the group but failed before creating
+                // the admin user, or the operator deleted just the user row
+                // to retry the seed with a new email). Blind-INSERTing would
+                // violate user_groups_name_key and abort the context. See #518.
+                UserGroup adminGroup = groupRepository.findByName("Administrators")
+                        .orElseGet(() -> groupRepository.save(UserGroup.builder()
+                                .name("Administrators")
+                                .permissions(Set.of("ADMIN", "READ_ALL", "WRITE_ALL", "ANNOTATE_ALL", "ADMIN_USER", "ADMIN_TAG", "ADMIN_PERMISSION"))
+                                .build()));
 
                 AppUser admin = AppUser.builder()
                         .email(adminEmail)

backend/src/main/java/org/raddatz/familienarchiv/user/UserService.java

@@ -271,9 +271,10 @@ public class UserService {
 
     @Transactional
     public UserGroup createGroup(GroupDTO dto) {
-        UserGroup group = new UserGroup();
-        group.setName(dto.getName());
-        group.setPermissions(dto.getPermissions());
+        UserGroup group = UserGroup.builder()
+                .name(dto.getName())
+                .permissions(dto.getPermissions() != null ? dto.getPermissions() : new HashSet<>())
+                .build();
         return groupRepository.save(group);
     }
 

backend/src/main/resources/application.yaml

@@ -38,6 +38,12 @@ spring:
           starttls:
             enable: true
 
+server:
+  # Behind Caddy/reverse proxy: trust X-Forwarded-{Proto,For,Host} so that
+  # request.getScheme(), redirect URLs, and Spring Session "Secure" cookies
+  # reflect the original https client request, not the http hop from Caddy.
+  forward-headers-strategy: native
+
 management:
   health:
     mail:
@@ -63,7 +69,11 @@ app:
     from: ${APP_MAIL_FROM:noreply@familienarchiv.local}
 
   admin:
-    username: ${APP_ADMIN_USERNAME:admin}
+    # Key must be `email`, not `username` — UserDataInitializer reads
+    # `${app.admin.email:...}`. The env-var name stays APP_ADMIN_USERNAME
+    # to match the existing Gitea secrets and DEPLOYMENT.md §3.3.
+    # See #513.
+    email: ${APP_ADMIN_USERNAME:admin@familienarchiv.local}
     password: ${APP_ADMIN_PASSWORD:admin123}
 
   import:

backend/src/main/resources/db/migration/V62__index_fk_columns.sql

@@ -0,0 +1,8 @@
+-- Speeds up "documents by sender" queries used on /persons/[id] Korrespondenz-Überblick (#306),
+-- /briefwechsel, and bulk-edit flows.
+CREATE INDEX IF NOT EXISTS idx_documents_sender_id
+    ON documents(sender_id);
+
+-- Speeds up "comments by author" queries on admin user detail and (future) contributor profile.
+CREATE INDEX IF NOT EXISTS idx_comments_author_id
+    ON document_comments(author_id);

backend/src/main/resources/db/migration/V63__deduplicate_group_permissions.sql

@@ -0,0 +1,7 @@
+-- Remove duplicate (group_id, permission) rows that accumulated without a UNIQUE constraint.
+-- Keeps the row with the smallest ctid (earliest physical insertion order).
+DELETE FROM group_permissions a
+USING group_permissions b
+WHERE a.ctid < b.ctid
+  AND a.group_id = b.group_id
+  AND a.permission = b.permission;

backend/src/main/resources/db/migration/V64__group_permissions_primary_key.sql

@@ -0,0 +1,11 @@
+-- Add NOT NULL and PRIMARY KEY to group_permissions.
+-- Requires V63 to have run first (no duplicates can remain).
+--
+-- After this migration, future seed migrations can use:
+--   INSERT INTO group_permissions ... ON CONFLICT DO NOTHING
+-- instead of the INSERT ... WHERE NOT EXISTS pattern used before V64.
+ALTER TABLE group_permissions
+    ALTER COLUMN permission SET NOT NULL;
+
+ALTER TABLE group_permissions
+    ADD CONSTRAINT pk_group_permissions PRIMARY KEY (group_id, permission);

backend/src/main/resources/db/migration/V65__tbmp_promote_unique_to_pk.sql

@@ -0,0 +1,8 @@
+-- Promote the de-facto unique constraint on transcription_block_mentioned_persons to a named PK.
+-- uq_tbmp_block_person (added in V57) is backed by a B-tree index identical to a PK;
+-- this rename makes the naming convention explicit (pk_* vs uq_*).
+ALTER TABLE transcription_block_mentioned_persons
+    DROP CONSTRAINT uq_tbmp_block_person;
+
+ALTER TABLE transcription_block_mentioned_persons
+    ADD CONSTRAINT pk_tbmp PRIMARY KEY (block_id, person_id);

backend/src/test/java/org/raddatz/familienarchiv/MigrationIntegrationTest.java

@@ -399,6 +399,86 @@ class MigrationIntegrationTest {
               AND dc.annotation_id IS NOT NULL
             """;
 
+    // ─── V62: indexes on FK columns ──────────────────────────────────────────
+
+    @Test
+    void v62_idx_documents_sender_id_exists() {
+        Integer count = jdbc.queryForObject(
+                "SELECT COUNT(*) FROM pg_catalog.pg_indexes WHERE tablename = 'documents' AND indexname = 'idx_documents_sender_id'",
+                Integer.class);
+        assertThat(count).isEqualTo(1);
+    }
+
+    @Test
+    void v62_idx_comments_author_id_exists() {
+        Integer count = jdbc.queryForObject(
+                "SELECT COUNT(*) FROM pg_catalog.pg_indexes WHERE tablename = 'document_comments' AND indexname = 'idx_comments_author_id'",
+                Integer.class);
+        assertThat(count).isEqualTo(1);
+    }
+
+    // ─── V63+V64: group_permissions dedup + primary key ──────────────────────
+
+    @Test
+    void v64_pk_group_permissions_exists() {
+        Integer count = jdbc.queryForObject(
+                """
+                SELECT COUNT(*) FROM pg_catalog.pg_constraint c
+                JOIN pg_catalog.pg_class t ON c.conrelid = t.oid
+                WHERE t.relname = 'group_permissions'
+                  AND c.conname = 'pk_group_permissions'
+                  AND c.contype = 'p'
+                """,
+                Integer.class);
+        assertThat(count).isEqualTo(1);
+    }
+
+    @Test
+    void v64_permission_column_isNotNullable() {
+        Integer count = jdbc.queryForObject(
+                """
+                SELECT COUNT(*) FROM information_schema.columns
+                WHERE table_schema = 'public'
+                  AND table_name = 'group_permissions'
+                  AND column_name = 'permission'
+                  AND is_nullable = 'NO'
+                """,
+                Integer.class);
+        assertThat(count).isEqualTo(1);
+    }
+
+    @Test
+    @Transactional(propagation = Propagation.NOT_SUPPORTED)
+    void v64_rejectsDuplicateGroupPermission() {
+        UUID groupId = createUserGroup("DuplicateTestGroup-" + UUID.randomUUID());
+        try {
+            jdbc.update("INSERT INTO group_permissions (group_id, permission) VALUES (?, 'READ_ALL')", groupId);
+
+            assertThatThrownBy(() ->
+                    jdbc.update("INSERT INTO group_permissions (group_id, permission) VALUES (?, 'READ_ALL')", groupId)
+            ).isInstanceOf(DataIntegrityViolationException.class);
+        } finally {
+            jdbc.update("DELETE FROM group_permissions WHERE group_id = ?", groupId);
+            jdbc.update("DELETE FROM user_groups WHERE id = ?", groupId);
+        }
+    }
+
+    // ─── V65: tbmp UNIQUE promoted to PRIMARY KEY ─────────────────────────────
+
+    @Test
+    void v65_pk_tbmp_exists() {
+        Integer count = jdbc.queryForObject(
+                """
+                SELECT COUNT(*) FROM pg_catalog.pg_constraint c
+                JOIN pg_catalog.pg_class t ON c.conrelid = t.oid
+                WHERE t.relname = 'transcription_block_mentioned_persons'
+                  AND c.conname = 'pk_tbmp'
+                  AND c.contype = 'p'
+                """,
+                Integer.class);
+        assertThat(count).isEqualTo(1);
+    }
+
     // ─── helpers ─────────────────────────────────────────────────────────────
 
     private UUID createPerson(String firstName, String lastName) {
@@ -482,4 +562,10 @@ class MigrationIntegrationTest {
                 """, id, recipientId, docId, commentId);
         return id;
     }
+
+    private UUID createUserGroup(String name) {
+        UUID id = UUID.randomUUID();
+        jdbc.update("INSERT INTO user_groups (id, name) VALUES (?, ?)", id, name);
+        return id;
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/config/ForwardHeadersConfigurationTest.java

@@ -0,0 +1,48 @@
+package org.raddatz.familienarchiv.config;
+
+import org.junit.jupiter.api.Test;
+import org.springframework.beans.factory.config.YamlPropertiesFactoryBean;
+import org.springframework.boot.web.server.autoconfigure.ServerProperties.ForwardHeadersStrategy;
+import org.springframework.boot.context.properties.bind.Binder;
+import org.springframework.boot.context.properties.source.ConfigurationPropertySources;
+import org.springframework.core.env.PropertiesPropertySource;
+import org.springframework.core.io.ClassPathResource;
+
+import java.util.Properties;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+/**
+ * Binds {@code server.forward-headers-strategy} from {@code application.yaml} into
+ * Spring Boot's typed {@link ForwardHeadersStrategy} enum. The binder rejects any
+ * value that is not a valid enum constant ({@code BindException}), so a typo
+ * ({@code "nativ"}, {@code "Native"}, {@code "framework "}) or a future Spring
+ * rename of the property fails the test, not silently degrades to {@code NONE}.
+ *
+ * <p>No Spring context, no embedded server, no Testcontainers — this is the
+ * cheapest test that pins the contract "Caddy's X-Forwarded-Proto is trusted".
+ */
+class ForwardHeadersConfigurationTest {
+
+    @Test
+    void forward_headers_strategy_binds_to_NATIVE() {
+        YamlPropertiesFactoryBean yaml = new YamlPropertiesFactoryBean();
+        yaml.setResources(new ClassPathResource("application.yaml"));
+        Properties props = yaml.getObject();
+        assertThat(props).as("application.yaml must be on the classpath").isNotNull();
+
+        Binder binder = new Binder(ConfigurationPropertySources.from(
+            new PropertiesPropertySource("application", props)));
+
+        ForwardHeadersStrategy strategy = binder
+            .bind("server.forward-headers-strategy", ForwardHeadersStrategy.class)
+            .orElseThrow(() -> new AssertionError(
+                "server.forward-headers-strategy is missing from application.yaml"));
+
+        assertThat(strategy)
+            .as("Spring must trust X-Forwarded-Proto from Caddy so that "
+                + "request.getScheme(), redirect URLs, and the Spring Session "
+                + "'Secure' cookie reflect the original https client request.")
+            .isEqualTo(ForwardHeadersStrategy.NATIVE);
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/document/DocumentFtsPagedIntegrationTest.java

@@ -0,0 +1,109 @@
+package org.raddatz.familienarchiv.document;
+
+import jakarta.persistence.EntityManager;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.PostgresContainerConfig;
+import org.raddatz.familienarchiv.config.FlywayConfig;
+import org.raddatz.familienarchiv.document.DocumentRepository;
+import org.raddatz.familienarchiv.document.Document;
+import org.raddatz.familienarchiv.document.DocumentStatus;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.boot.data.jpa.test.autoconfigure.DataJpaTest;
+import org.springframework.boot.jdbc.test.autoconfigure.AutoConfigureTestDatabase;
+import org.springframework.context.annotation.Import;
+import org.springframework.test.annotation.DirtiesContext;
+
+import java.util.List;
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatNoException;
+
+/**
+ * Repository-level integration tests for {@code findFtsPageRaw}: verifies that the
+ * paginated FTS query returns exactly page-size rows and that the window-function
+ * total reflects the full match count, not just the page count.
+ *
+ * <p>Uses real Postgres via Testcontainers so the GIN index, tsvector trigger, and
+ * {@code websearch_to_tsquery} semantics are identical to production.
+ *
+ * <p>{@code AFTER_CLASS} dirty-context keeps the Spring context alive for all tests
+ * in this class and rebuilds it once at the end, rather than after every test.
+ */
+@DataJpaTest
+@AutoConfigureTestDatabase(replace = AutoConfigureTestDatabase.Replace.NONE)
+@Import({PostgresContainerConfig.class, FlywayConfig.class})
+@DirtiesContext(classMode = DirtiesContext.ClassMode.AFTER_CLASS)
+class DocumentFtsPagedIntegrationTest {
+
+    @Autowired DocumentRepository documentRepository;
+    @Autowired EntityManager em;
+
+    // 60 docs match "Walter"; 10 docs with "Hans" do not.
+    private static final int WALTER_COUNT = 60;
+    private static final int PAGE_SIZE = 50;
+
+    @BeforeEach
+    void seed() {
+        documentRepository.deleteAll();
+        em.flush();
+        for (int i = 0; i < WALTER_COUNT; i++) {
+            documentRepository.saveAndFlush(doc("Brief von Walter Nr. " + i));
+        }
+        for (int i = 0; i < 10; i++) {
+            documentRepository.saveAndFlush(doc("Brief von Hans Nr. " + i));
+        }
+        em.clear();
+    }
+
+    @Test
+    void findFtsPageRaw_firstPage_returnsPageSizeRows() {
+        List<Object[]> rows = documentRepository.findFtsPageRaw("Walter", 0, PAGE_SIZE);
+
+        assertThat(rows).hasSize(PAGE_SIZE);
+    }
+
+    @Test
+    void findFtsPageRaw_windowTotal_equalsFullMatchCount_notPageSize() {
+        List<Object[]> rows = documentRepository.findFtsPageRaw("Walter", 0, PAGE_SIZE);
+
+        long total = ((Number) rows.get(0)[2]).longValue();
+        assertThat(total).isEqualTo(WALTER_COUNT);
+    }
+
+    @Test
+    void findFtsPageRaw_lastPage_returnsRemainder() {
+        int remainder = WALTER_COUNT % PAGE_SIZE; // 60 % 50 = 10
+        List<Object[]> rows = documentRepository.findFtsPageRaw("Walter", PAGE_SIZE, PAGE_SIZE);
+
+        assertThat(rows).hasSize(remainder);
+        long total = ((Number) rows.get(0)[2]).longValue();
+        assertThat(total).isEqualTo(WALTER_COUNT);
+    }
+
+    @Test
+    void findFtsPageRaw_noMatches_returnsEmptyList() {
+        List<Object[]> rows = documentRepository.findFtsPageRaw("XYZ_KEIN_TREFFER", 0, PAGE_SIZE);
+
+        assertThat(rows).isEmpty();
+    }
+
+    @Test
+    void findFtsPageRaw_stopwordOnlyQuery_returnsEmptyList_noException() {
+        assertThatNoException().isThrownBy(() -> {
+            List<Object[]> rows = documentRepository.findFtsPageRaw("der die das und", 0, PAGE_SIZE);
+            assertThat(rows).isEmpty();
+        });
+    }
+
+    // ─── Helper ───────────────────────────────────────────────────────────────
+
+    private Document doc(String title) {
+        return Document.builder()
+                .title(title)
+                .originalFilename(title.replace(" ", "_") + ".pdf")
+                .status(DocumentStatus.UPLOADED)
+                .build();
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/document/DocumentFtsTest.java

@@ -69,7 +69,7 @@ class DocumentFtsTest {
         documentRepository.saveAndFlush(document("Alter Brief"));
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Brief");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Brief");
 
         assertThat(ids).hasSize(1);
     }
@@ -79,7 +79,7 @@ class DocumentFtsTest {
         documentRepository.saveAndFlush(document("Alter Brief"));
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Briefe");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Briefe");
 
         assertThat(ids).hasSize(1);
     }
@@ -89,7 +89,7 @@ class DocumentFtsTest {
         documentRepository.saveAndFlush(document("Ein furchtbarer Brief"));
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("furchtb");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("furchtb");
 
         assertThat(ids).hasSize(1);
     }
@@ -99,7 +99,7 @@ class DocumentFtsTest {
         documentRepository.saveAndFlush(document("Familienfoto"));
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Brief");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Brief");
 
         assertThat(ids).isEmpty();
     }
@@ -115,7 +115,7 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("schreiben");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("schreiben");
 
         assertThat(ids).contains(doc.getId());
     }
@@ -125,14 +125,14 @@ class DocumentFtsTest {
         Document doc = documentRepository.saveAndFlush(document("Leeres Dokument"));
         em.clear();
 
-        assertThat(documentRepository.findRankedIdsByFts("Grundbuch")).isEmpty();
+        assertThat(documentRepository.findAllMatchingIdsByFts("Grundbuch")).isEmpty();
 
         UUID annotationId = annotation(doc.getId());
         blockRepository.saveAndFlush(block(doc.getId(), annotationId, "Grundbuch Eintrag 1923", 0));
         em.flush();
         em.clear();
 
-        assertThat(documentRepository.findRankedIdsByFts("Grundbuch")).contains(doc.getId());
+        assertThat(documentRepository.findAllMatchingIdsByFts("Grundbuch")).contains(doc.getId());
     }
 
     @Test
@@ -144,13 +144,13 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        assertThat(documentRepository.findRankedIdsByFts("Grundbuch")).contains(doc.getId());
+        assertThat(documentRepository.findAllMatchingIdsByFts("Grundbuch")).contains(doc.getId());
 
         blockRepository.deleteById(block.getId());
         em.flush();
         em.clear();
 
-        assertThat(documentRepository.findRankedIdsByFts("Grundbuch")).doesNotContain(doc.getId());
+        assertThat(documentRepository.findAllMatchingIdsByFts("Grundbuch")).doesNotContain(doc.getId());
     }
 
     // ─── Ranking ───────────────────────────────────────────────────────────────
@@ -166,7 +166,7 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Grundbuch");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Grundbuch");
 
         assertThat(ids).hasSize(2);
         assertThat(ids.get(0)).isEqualTo(docA.getId());
@@ -179,7 +179,7 @@ class DocumentFtsTest {
         documentRepository.saveAndFlush(document("Ein Brief von der Oma"));
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("der die das und");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("der die das und");
 
         assertThat(ids).isEmpty();
     }
@@ -195,7 +195,7 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Wille");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Wille");
 
         assertThat(ids).contains(doc.getId());
     }
@@ -205,7 +205,7 @@ class DocumentFtsTest {
         documentRepository.saveAndFlush(document("Brief"));
         em.clear();
 
-        assertThatNoException().isThrownBy(() -> documentRepository.findRankedIdsByFts("((("));
+        assertThatNoException().isThrownBy(() -> documentRepository.findAllMatchingIdsByFts("((("));
     }
 
     // ─── Weight C: sender/receiver names ───────────────────────────────────────
@@ -223,7 +223,7 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Schmidt");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Schmidt");
 
         assertThat(ids).contains(doc.getId());
     }
@@ -241,7 +241,7 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Raddatz");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Raddatz");
 
         assertThat(ids).contains(doc.getId());
     }
@@ -260,7 +260,7 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        List<UUID> ids = documentRepository.findRankedIdsByFts("Familiengeschichte");
+        List<UUID> ids = documentRepository.findAllMatchingIdsByFts("Familiengeschichte");
 
         assertThat(ids).hasSize(1);
     }
@@ -278,7 +278,7 @@ class DocumentFtsTest {
         em.flush();
         em.clear();
 
-        List<UUID> rankedIds = documentRepository.findRankedIdsByFts("Grundbuch");
+        List<UUID> rankedIds = documentRepository.findAllMatchingIdsByFts("Grundbuch");
         Specification<Document> spec = Specification.where(hasIds(rankedIds))
                 .and(hasStatus(DocumentStatus.UPLOADED));
 

backend/src/test/java/org/raddatz/familienarchiv/document/DocumentServiceSortTest.java

@@ -21,17 +21,22 @@ import org.springframework.data.domain.Pageable;
 import org.springframework.data.jpa.domain.Specification;
 
 import java.time.LocalDate;
+import java.util.ArrayList;
 import java.util.List;
 import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
 import static org.mockito.ArgumentMatchers.any;
+import static org.mockito.ArgumentMatchers.anyInt;
+import static org.mockito.ArgumentMatchers.anyString;
+import static org.mockito.Mockito.never;
+import static org.mockito.Mockito.verify;
 import static org.mockito.Mockito.when;
 
 @ExtendWith(MockitoExtension.class)
 class DocumentServiceSortTest {
 
-    private static final Pageable UNPAGED = org.springframework.data.domain.PageRequest.of(0, 10_000);
+    private static final Pageable PAGE = org.springframework.data.domain.PageRequest.of(0, 10_000);
 
     @Mock DocumentRepository documentRepository;
     @Mock PersonService personService;
@@ -43,12 +48,12 @@ class DocumentServiceSortTest {
     @Mock TranscriptionBlockQueryService transcriptionBlockQueryService;
     @InjectMocks DocumentService documentService;
 
-    // ─── searchDocuments — DATE sort ──────────────────────────────────────────
+    // ─── DATE sort ────────────────────────────────────────────────────────────
 
     @Test
     void searchDocuments_with_DATE_sort_and_text_sorts_chronologically_not_by_relevance() {
-        UUID id1 = UUID.randomUUID();  // rank position 0 (higher relevance, older doc)
-        UUID id2 = UUID.randomUUID();  // rank position 1 (lower relevance, newer doc)
+        UUID id1 = UUID.randomUUID();  // higher relevance, older doc
+        UUID id2 = UUID.randomUUID();  // lower relevance, newer doc
 
         Document older = Document.builder().id(id1)
                 .title("Brief").status(DocumentStatus.UPLOADED)
@@ -57,38 +62,48 @@ class DocumentServiceSortTest {
                 .title("Brief").status(DocumentStatus.UPLOADED)
                 .documentDate(LocalDate.of(1960, 1, 1)).build();
 
-        // FTS returns id1 first (higher rank), id2 second
-        when(documentRepository.findRankedIdsByFts("Brief")).thenReturn(List.of(id1, id2));
-        // findAll(spec, pageable) — the correct date path — returns date-DESC order
+        when(documentRepository.findAllMatchingIdsByFts("Brief")).thenReturn(List.of(id1, id2));
         when(documentRepository.findAll(any(Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of(newer, older)));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                "Brief", null, null, null, null, null, null, null, DocumentSort.DATE, "DESC", null, UNPAGED);
+                "Brief", null, null, null, null, null, null, null, DocumentSort.DATE, "DESC", null, PAGE);
 
-        // Expect: date order (newer 1960 first), NOT rank order (older 1940 first)
         assertThat(result.items()).hasSize(2);
-        assertThat(result.items().get(0).document().getId()).isEqualTo(id2);  // newer doc first
+        assertThat(result.items().get(0).document().getId()).isEqualTo(id2);  // newer first
     }
 
-    // ─── searchDocuments — RELEVANCE sort ─────────────────────────────────────
+    // ─── RELEVANCE sort — pure text (no filters) ──────────────────────────────
+
+    @Test
+    void searchDocuments_relevance_pureText_calls_findFtsPageRaw_not_findAllMatchingIds() {
+        UUID id1 = UUID.randomUUID();
+        List<Object[]> ftsRows = ftsRows(id1, 0.5d, 1L);
+        when(documentRepository.findFtsPageRaw(anyString(), anyInt(), anyInt())).thenReturn(ftsRows);
+        when(documentRepository.findAllById(any()))
+                .thenReturn(List.of(doc(id1)));
+
+        documentService.searchDocuments(
+                "Brief", null, null, null, null, null, null, null, DocumentSort.RELEVANCE, null, null, PAGE);
+
+        verify(documentRepository).findFtsPageRaw(anyString(), anyInt(), anyInt());
+        verify(documentRepository, never()).findAllMatchingIdsByFts(anyString());
+    }
 
     @Test
     void searchDocuments_with_RELEVANCE_sort_and_text_preserves_fts_rank_order() {
-        UUID id1 = UUID.randomUUID();  // rank position 0
-        UUID id2 = UUID.randomUUID();  // rank position 1
+        UUID id1 = UUID.randomUUID();  // higher rank — must appear first
+        UUID id2 = UUID.randomUUID();  // lower rank
 
-        Document doc1 = Document.builder().id(id1).title("Brief").status(DocumentStatus.UPLOADED).build();
-        Document doc2 = Document.builder().id(id2).title("Brief").status(DocumentStatus.UPLOADED).build();
-
-        when(documentRepository.findRankedIdsByFts("Brief")).thenReturn(List.of(id1, id2));
-        when(documentRepository.findAll(any(Specification.class)))
-                .thenReturn(List.of(doc2, doc1));  // unordered from DB
+        List<Object[]> ftsRows = new ArrayList<>();
+        ftsRows.add(new Object[]{id1, 0.8d, 2L});
+        ftsRows.add(new Object[]{id2, 0.3d, 2L});
+        when(documentRepository.findFtsPageRaw(anyString(), anyInt(), anyInt())).thenReturn(ftsRows);
+        when(documentRepository.findAllById(any())).thenReturn(List.of(doc(id2), doc(id1))); // unordered from JPA
 
         DocumentSearchResult result = documentService.searchDocuments(
-                "Brief", null, null, null, null, null, null, null, DocumentSort.RELEVANCE, null, null, UNPAGED);
+                "Brief", null, null, null, null, null, null, null, DocumentSort.RELEVANCE, null, null, PAGE);
 
-        // Expect: rank order restored (id1 first)
         assertThat(result.items().get(0).document().getId()).isEqualTo(id1);
     }
 
@@ -97,16 +112,82 @@ class DocumentServiceSortTest {
         UUID id1 = UUID.randomUUID();
         UUID id2 = UUID.randomUUID();
 
-        Document doc1 = Document.builder().id(id1).title("Brief").status(DocumentStatus.UPLOADED).build();
-        Document doc2 = Document.builder().id(id2).title("Brief").status(DocumentStatus.UPLOADED).build();
-
-        when(documentRepository.findRankedIdsByFts("Brief")).thenReturn(List.of(id1, id2));
-        when(documentRepository.findAll(any(Specification.class)))
-                .thenReturn(List.of(doc2, doc1));
+        List<Object[]> ftsRows = new ArrayList<>();
+        ftsRows.add(new Object[]{id1, 0.8d, 2L});
+        ftsRows.add(new Object[]{id2, 0.3d, 2L});
+        when(documentRepository.findFtsPageRaw(anyString(), anyInt(), anyInt())).thenReturn(ftsRows);
+        when(documentRepository.findAllById(any())).thenReturn(List.of(doc(id2), doc(id1)));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                "Brief", null, null, null, null, null, null, null, null, null, null, UNPAGED);
+                "Brief", null, null, null, null, null, null, null, null, null, null, PAGE);
 
         assertThat(result.items().get(0).document().getId()).isEqualTo(id1);
     }
+
+    // ─── RELEVANCE sort — overflow guard ─────────────────────────────────────
+
+    @Test
+    void searchDocuments_relevance_returns_empty_when_offset_exceeds_maxInt() {
+        // offset = pageNumber * pageSize; choose values so offset > Integer.MAX_VALUE
+        Pageable hugePage = org.springframework.data.domain.PageRequest.of(Integer.MAX_VALUE / 10 + 1, 10);
+
+        DocumentSearchResult result = documentService.searchDocuments(
+                "Brief", null, null, null, null, null, null, null,
+                DocumentSort.RELEVANCE, null, null, hugePage);
+
+        assertThat(result.items()).isEmpty();
+        verify(documentRepository, never()).findFtsPageRaw(anyString(), anyInt(), anyInt());
+    }
+
+    // ─── toFtsPage — UUID-as-String JDBC driver variance ────────────────────
+
+    @Test
+    void searchDocuments_relevance_handles_string_uuid_from_jdbc_driver() {
+        String stringId = "11111111-1111-1111-1111-111111111111";
+        UUID uuidId = UUID.fromString(stringId);
+        // Simulate a JDBC driver that returns the id column as String instead of UUID
+        List<Object[]> ftsRows = new ArrayList<>();
+        ftsRows.add(new Object[]{stringId, 0.5d, 1L});
+        when(documentRepository.findFtsPageRaw(anyString(), anyInt(), anyInt())).thenReturn(ftsRows);
+        when(documentRepository.findAllById(any())).thenReturn(List.of(doc(uuidId)));
+
+        DocumentSearchResult result = documentService.searchDocuments(
+                "Brief", null, null, null, null, null, null, null,
+                DocumentSort.RELEVANCE, null, null, PAGE);
+
+        assertThat(result.items()).hasSize(1);
+        assertThat(result.items().get(0).document().getId()).isEqualTo(uuidId);
+    }
+
+    // ─── RELEVANCE sort — text + active filter ────────────────────────────────
+
+    @Test
+    void searchDocuments_relevance_with_active_filter_uses_inMemory_path() {
+        UUID id1 = UUID.randomUUID();
+        UUID id2 = UUID.randomUUID();
+
+        when(documentRepository.findAllMatchingIdsByFts("Brief")).thenReturn(List.of(id1, id2));
+        when(documentRepository.findAll(any(Specification.class)))
+                .thenReturn(List.of(doc(id2), doc(id1)));
+
+        // sender filter is active → triggers in-memory path, not findFtsPageRaw
+        LocalDate from = LocalDate.of(1900, 1, 1);
+        documentService.searchDocuments(
+                "Brief", from, null, null, null, null, null, null, DocumentSort.RELEVANCE, null, null, PAGE);
+
+        verify(documentRepository, never()).findFtsPageRaw(anyString(), anyInt(), anyInt());
+        verify(documentRepository).findAllMatchingIdsByFts("Brief");
+    }
+
+    // ─── Helpers ──────────────────────────────────────────────────────────────
+
+    private static Document doc(UUID id) {
+        return Document.builder().id(id).title("Brief").status(DocumentStatus.UPLOADED).build();
+    }
+
+    private static List<Object[]> ftsRows(UUID id, double rank, long total) {
+        List<Object[]> rows = new ArrayList<>();
+        rows.add(new Object[]{id, rank, total});
+        return rows;
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/document/DocumentServiceTest.java

@@ -1620,9 +1620,10 @@ class DocumentServiceTest {
         // chr(1)=\u0001 marks start, chr(2)=\u0002 marks end of highlighted term
         List<Object[]> rows = Collections.singletonList(new Object[]{docId, "\u0001Brief\u0002 an Anna", null, false, null, null, null});
 
-        when(documentRepository.findRankedIdsByFts("Brief")).thenReturn(List.of(docId));
-        when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class)))
-                .thenReturn(List.of(doc));
+        List<Object[]> ftsRows = new java.util.ArrayList<>();
+        ftsRows.add(new Object[]{docId, 0.5d, 1L});
+        when(documentRepository.findFtsPageRaw(anyString(), anyInt(), anyInt())).thenReturn(ftsRows);
+        when(documentRepository.findAllById(any())).thenReturn(List.of(doc));
         when(documentRepository.findEnrichmentData(any(), eq("Brief"))).thenReturn(rows);
 
         DocumentSearchResult result = documentService.searchDocuments(
@@ -1654,9 +1655,10 @@ class DocumentServiceTest {
         String snippetHeadline = "Hier ist der \u0001Brief\u0002 aus Berlin";
         List<Object[]> rows = Collections.singletonList(new Object[]{docId, "Dok", snippetHeadline, false, null, null, null});
 
-        when(documentRepository.findRankedIdsByFts("Brief")).thenReturn(List.of(docId));
-        when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class)))
-                .thenReturn(List.of(doc));
+        List<Object[]> snippetFtsRows = new java.util.ArrayList<>();
+        snippetFtsRows.add(new Object[]{docId, 0.5d, 1L});
+        when(documentRepository.findFtsPageRaw(anyString(), anyInt(), anyInt())).thenReturn(snippetFtsRows);
+        when(documentRepository.findAllById(any())).thenReturn(List.of(doc));
         when(documentRepository.findEnrichmentData(any(), eq("Brief"))).thenReturn(rows);
 
         DocumentSearchResult result = documentService.searchDocuments(
@@ -2202,7 +2204,7 @@ class DocumentServiceTest {
 
     @Test
     void findIdsForFilter_returnsEmpty_whenFtsHasNoMatches() {
-        when(documentRepository.findRankedIdsByFts("xyz")).thenReturn(List.of());
+        when(documentRepository.findAllMatchingIdsByFts("xyz")).thenReturn(List.of());
 
         List<UUID> result = documentService.findIdsForFilter(
                 "xyz", null, null, null, null, null, null, null, null);
@@ -2386,7 +2388,7 @@ class DocumentServiceTest {
 
     @Test
     void getDensity_shortCircuits_whenFtsReturnsNoMatches() {
-        when(documentRepository.findRankedIdsByFts("xyz")).thenReturn(List.of());
+        when(documentRepository.findAllMatchingIdsByFts("xyz")).thenReturn(List.of());
 
         DocumentDensityResult result = documentService.getDensity(
                 new DensityFilters("xyz", null, null, null, null, null, null));

backend/src/test/java/org/raddatz/familienarchiv/document/ThumbnailServiceIntegrationTest.java

@@ -10,6 +10,7 @@ import org.raddatz.familienarchiv.document.DocumentStatus;
 import org.raddatz.familienarchiv.document.DocumentRepository;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.boot.test.context.SpringBootTest;
+import org.springframework.test.context.ActiveProfiles;
 import org.springframework.context.annotation.Import;
 import org.springframework.test.context.DynamicPropertyRegistry;
 import org.springframework.test.context.DynamicPropertySource;
@@ -41,6 +42,7 @@ import static org.assertj.core.api.Assertions.assertThat;
  * test pyramid mocks at the FileService boundary.
  */
 @SpringBootTest
+@ActiveProfiles("test")
 @Import(PostgresContainerConfig.class)
 class ThumbnailServiceIntegrationTest {
 

backend/src/test/java/org/raddatz/familienarchiv/document/comment/CommentControllerTest.java

@@ -44,6 +44,14 @@ class CommentControllerTest {
 
     // ─── Block comment endpoints ─────────────────────────────────────────────
 
+    @Test
+    @WithMockUser
+    void getBlockComments_returns400_when_documentId_is_not_a_UUID() throws Exception {
+        UUID blockId = UUID.randomUUID();
+        mockMvc.perform(get("/api/documents/NOT-A-UUID/transcription-blocks/" + blockId + "/comments"))
+                .andExpect(status().isBadRequest());
+    }
+
     @Test
     @WithMockUser
     void getBlockComments_returns200() throws Exception {
@@ -115,6 +123,15 @@ class CommentControllerTest {
 
     // ─── Block reply endpoints ───────────────────────────────────────────────
 
+    @Test
+    @WithMockUser(authorities = "ANNOTATE_ALL")
+    void replyToBlockComment_returns400_when_blockId_is_not_a_UUID() throws Exception {
+        mockMvc.perform(post("/api/documents/" + DOC_ID + "/transcription-blocks/NOT-A-UUID"
+                                + "/comments/" + COMMENT_ID + "/replies")
+                        .contentType(MediaType.APPLICATION_JSON).content(COMMENT_JSON))
+                .andExpect(status().isBadRequest());
+    }
+
     @Test
     void replyToBlockComment_returns401_whenUnauthenticated() throws Exception {
         UUID blockId = UUID.randomUUID();

backend/src/test/java/org/raddatz/familienarchiv/security/AuthTokenCookieFilterTest.java

@@ -0,0 +1,134 @@
+package org.raddatz.familienarchiv.security;
+
+import jakarta.servlet.FilterChain;
+import jakarta.servlet.http.Cookie;
+import jakarta.servlet.http.HttpServletRequest;
+import jakarta.servlet.http.HttpServletResponse;
+import org.junit.jupiter.api.Test;
+import org.mockito.ArgumentCaptor;
+import org.springframework.mock.web.MockHttpServletRequest;
+import org.springframework.mock.web.MockHttpServletResponse;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.mockito.Mockito.mock;
+import static org.mockito.Mockito.times;
+import static org.mockito.Mockito.verify;
+
+/**
+ * The filter must turn a browser-side {@code Cookie: auth_token=Basic%20<base64>}
+ * into {@code Authorization: Basic <base64>} (URL-decoded) so that Spring's
+ * Basic-auth filter accepts it. Skips when the request already has an explicit
+ * {@code Authorization} header, or when no {@code auth_token} cookie is present.
+ *
+ * <p>See #520.
+ */
+class AuthTokenCookieFilterTest {
+
+    private final AuthTokenCookieFilter filter = new AuthTokenCookieFilter();
+
+    @Test
+    void promotes_url_encoded_auth_token_cookie_to_decoded_Authorization_header() throws Exception {
+        MockHttpServletRequest req = new MockHttpServletRequest();
+        req.setRequestURI("/api/users/me");
+        req.setCookies(new Cookie("auth_token", "Basic%20YWRtaW5AZmFtaWx5YXJjaGl2ZS5sb2NhbDpzZWNyZXQ%3D"));
+        MockHttpServletResponse res = new MockHttpServletResponse();
+        FilterChain chain = mock(FilterChain.class);
+
+        filter.doFilter(req, res, chain);
+
+        ArgumentCaptor<HttpServletRequest> captor = ArgumentCaptor.forClass(HttpServletRequest.class);
+        verify(chain, times(1)).doFilter(captor.capture(), org.mockito.ArgumentMatchers.any(HttpServletResponse.class));
+
+        HttpServletRequest forwarded = captor.getValue();
+        assertThat(forwarded.getHeader("Authorization"))
+            .as("Authorization must be URL-decoded so Spring's Basic parser sees a literal space")
+            .isEqualTo("Basic YWRtaW5AZmFtaWx5YXJjaGl2ZS5sb2NhbDpzZWNyZXQ=");
+    }
+
+    @Test
+    void preserves_explicit_Authorization_header_and_ignores_cookie() throws Exception {
+        MockHttpServletRequest req = new MockHttpServletRequest();
+        req.setRequestURI("/api/users/me");
+        req.addHeader("Authorization", "Basic explicit-header-wins");
+        req.setCookies(new Cookie("auth_token", "Basic%20cookie-would-have-promoted"));
+        MockHttpServletResponse res = new MockHttpServletResponse();
+        FilterChain chain = mock(FilterChain.class);
+
+        filter.doFilter(req, res, chain);
+
+        // Forwards the original request unchanged — same instance, no wrapping.
+        verify(chain).doFilter(req, res);
+    }
+
+    @Test
+    void passes_through_when_no_cookies_at_all() throws Exception {
+        MockHttpServletRequest req = new MockHttpServletRequest();
+        req.setRequestURI("/api/users/me");
+        MockHttpServletResponse res = new MockHttpServletResponse();
+        FilterChain chain = mock(FilterChain.class);
+
+        filter.doFilter(req, res, chain);
+
+        verify(chain).doFilter(req, res);
+    }
+
+    @Test
+    void passes_through_when_auth_token_cookie_is_absent() throws Exception {
+        MockHttpServletRequest req = new MockHttpServletRequest();
+        req.setRequestURI("/api/users/me");
+        req.setCookies(new Cookie("some_other_cookie", "value"));
+        MockHttpServletResponse res = new MockHttpServletResponse();
+        FilterChain chain = mock(FilterChain.class);
+
+        filter.doFilter(req, res, chain);
+
+        verify(chain).doFilter(req, res);
+    }
+
+    @Test
+    void passes_through_when_auth_token_cookie_is_empty() throws Exception {
+        MockHttpServletRequest req = new MockHttpServletRequest();
+        req.setRequestURI("/api/users/me");
+        req.setCookies(new Cookie("auth_token", ""));
+        MockHttpServletResponse res = new MockHttpServletResponse();
+        FilterChain chain = mock(FilterChain.class);
+
+        filter.doFilter(req, res, chain);
+
+        verify(chain).doFilter(req, res);
+    }
+
+    @Test
+    void passes_through_unchanged_when_request_is_outside_api_scope() throws Exception {
+        MockHttpServletRequest req = new MockHttpServletRequest();
+        // /actuator/health and similar must NOT receive a promoted Authorization
+        // header — they have their own access rules and should never be authed
+        // via the cookie.
+        req.setRequestURI("/actuator/health");
+        req.setCookies(new Cookie("auth_token", "Basic%20YWR=="));
+        MockHttpServletResponse res = new MockHttpServletResponse();
+        FilterChain chain = mock(FilterChain.class);
+
+        filter.doFilter(req, res, chain);
+
+        // Forwards the original request unchanged — same instance, no wrapping.
+        verify(chain).doFilter(req, res);
+    }
+
+    @Test
+    void passes_through_unchanged_when_cookie_value_is_malformed_percent_encoding() throws Exception {
+        MockHttpServletRequest req = new MockHttpServletRequest();
+        req.setRequestURI("/api/users/me");
+        // Lone "%" without two hex digits → URLDecoder throws → filter must
+        // refuse to forward a bogus Authorization header.
+        req.setCookies(new Cookie("auth_token", "Basic%2"));
+        MockHttpServletResponse res = new MockHttpServletResponse();
+        FilterChain chain = mock(FilterChain.class);
+
+        filter.doFilter(req, res, chain);
+
+        // Forwards the original request unchanged — Spring Security treats it
+        // as unauthenticated rather than crashing on bad input.
+        verify(chain).doFilter(req, res);
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/user/AdminSeedFailClosedTest.java

@@ -0,0 +1,174 @@
+package org.raddatz.familienarchiv.user;
+
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.extension.ExtendWith;
+import org.mockito.Mock;
+import org.mockito.junit.jupiter.MockitoExtension;
+import org.springframework.boot.CommandLineRunner;
+import org.springframework.core.env.Environment;
+import org.springframework.security.crypto.password.PasswordEncoder;
+import org.springframework.test.util.ReflectionTestUtils;
+
+import java.util.Optional;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
+import static org.mockito.ArgumentMatchers.any;
+import static org.mockito.ArgumentMatchers.anyString;
+import static org.mockito.ArgumentMatchers.eq;
+import static org.mockito.Mockito.never;
+import static org.mockito.Mockito.verify;
+import static org.mockito.Mockito.when;
+
+/**
+ * UserDataInitializer must refuse to seed the admin user with the hardcoded
+ * dev defaults when running outside the {@code dev} profile.
+ *
+ * <p>Why this matters: per DEPLOYMENT.md §3.5 and ADR-011, the admin password
+ * is permanently locked on first deploy (UserDataInitializer only seeds when
+ * the row is missing). If an operator forgets to set {@code APP_ADMIN_USERNAME}
+ * / {@code APP_ADMIN_PASSWORD}, prod silently boots with the well-known dev
+ * defaults — a credential-disclosure foot-gun, not a config typo. See #513.
+ */
+@ExtendWith(MockitoExtension.class)
+class AdminSeedFailClosedTest {
+
+    @Mock AppUserRepository userRepository;
+    @Mock UserGroupRepository groupRepository;
+    @Mock Environment environment;
+    @Mock PasswordEncoder passwordEncoder;
+
+    UserDataInitializer initializer;
+
+    @BeforeEach
+    void setUp() {
+        initializer = new UserDataInitializer(userRepository, groupRepository, environment);
+    }
+
+    @Test
+    void refuses_to_seed_when_email_is_default_and_profile_is_not_dev() throws Exception {
+        when(userRepository.findByEmail(anyString())).thenReturn(Optional.empty());
+        when(environment.matchesProfiles("dev", "test", "e2e")).thenReturn(false);
+        ReflectionTestUtils.setField(initializer, "adminEmail", UserDataInitializer.DEFAULT_ADMIN_EMAIL);
+        ReflectionTestUtils.setField(initializer, "adminPassword", "operator-set-this-one");
+
+        CommandLineRunner runner = initializer.initAdminUser(passwordEncoder);
+
+        assertThatThrownBy(() -> runner.run())
+            .isInstanceOf(IllegalStateException.class)
+            .hasMessageContaining("default credentials")
+            .hasMessageContaining("permanent");
+
+        verify(userRepository, never()).save(org.mockito.ArgumentMatchers.any());
+    }
+
+    @Test
+    void refuses_to_seed_when_password_is_default_and_profile_is_not_dev() throws Exception {
+        when(userRepository.findByEmail(anyString())).thenReturn(Optional.empty());
+        when(environment.matchesProfiles("dev", "test", "e2e")).thenReturn(false);
+        ReflectionTestUtils.setField(initializer, "adminEmail", "admin@archiv.raddatz.cloud");
+        ReflectionTestUtils.setField(initializer, "adminPassword", UserDataInitializer.DEFAULT_ADMIN_PASSWORD);
+
+        CommandLineRunner runner = initializer.initAdminUser(passwordEncoder);
+
+        assertThatThrownBy(() -> runner.run())
+            .isInstanceOf(IllegalStateException.class)
+            .hasMessageContaining("default credentials");
+    }
+
+    @Test
+    void allows_seed_when_both_values_are_set_and_profile_is_not_dev() throws Exception {
+        when(userRepository.findByEmail(anyString())).thenReturn(Optional.empty());
+        when(groupRepository.findByName("Administrators")).thenReturn(Optional.empty());
+        when(groupRepository.save(any(UserGroup.class))).thenAnswer(inv -> inv.getArgument(0));
+        when(environment.matchesProfiles("dev", "test", "e2e")).thenReturn(false);
+        when(passwordEncoder.encode(anyString())).thenReturn("$2a$10$stub");
+        ReflectionTestUtils.setField(initializer, "adminEmail", "admin@archiv.raddatz.cloud");
+        ReflectionTestUtils.setField(initializer, "adminPassword", "a-real-strong-password");
+
+        CommandLineRunner runner = initializer.initAdminUser(passwordEncoder);
+        runner.run();
+
+        verify(userRepository).save(any(AppUser.class));
+    }
+
+    @Test
+    void allows_seed_with_defaults_when_profile_is_dev() throws Exception {
+        when(userRepository.findByEmail(anyString())).thenReturn(Optional.empty());
+        when(groupRepository.findByName("Administrators")).thenReturn(Optional.empty());
+        when(groupRepository.save(any(UserGroup.class))).thenAnswer(inv -> inv.getArgument(0));
+        when(environment.matchesProfiles("dev", "test", "e2e")).thenReturn(true);
+        when(passwordEncoder.encode(anyString())).thenReturn("$2a$10$stub");
+        ReflectionTestUtils.setField(initializer, "adminEmail", UserDataInitializer.DEFAULT_ADMIN_EMAIL);
+        ReflectionTestUtils.setField(initializer, "adminPassword", UserDataInitializer.DEFAULT_ADMIN_PASSWORD);
+
+        CommandLineRunner runner = initializer.initAdminUser(passwordEncoder);
+        runner.run();
+
+        verify(userRepository).save(any(AppUser.class));
+    }
+
+    @Test
+    void does_not_check_defaults_when_admin_already_exists() throws Exception {
+        AppUser existing = AppUser.builder()
+            .email("someone@example.com")
+            .password("$2a$10$stub")
+            .build();
+        when(userRepository.findByEmail(anyString())).thenReturn(Optional.of(existing));
+        ReflectionTestUtils.setField(initializer, "adminEmail", UserDataInitializer.DEFAULT_ADMIN_EMAIL);
+        ReflectionTestUtils.setField(initializer, "adminPassword", UserDataInitializer.DEFAULT_ADMIN_PASSWORD);
+
+        CommandLineRunner runner = initializer.initAdminUser(passwordEncoder);
+        runner.run();
+
+        verify(userRepository, never()).save(org.mockito.ArgumentMatchers.any());
+        // Importantly, no IllegalStateException — re-deploys must not panic over
+        // historical default-seeded data they cannot retroactively fix.
+    }
+
+    @Test
+    void reuses_existing_Administrators_group_when_seeding_a_new_admin() throws Exception {
+        // Setup: admin user does not exist, but the Administrators group does
+        // (e.g. previous boot seeded the group then failed; operator deleted
+        // the bad user row to retry with a corrected APP_ADMIN_USERNAME). The
+        // re-seed must reuse the group, not blind-INSERT a duplicate. See #518.
+        UserGroup existingGroup = UserGroup.builder()
+            .name("Administrators")
+            .build();
+        when(userRepository.findByEmail(anyString())).thenReturn(Optional.empty());
+        when(groupRepository.findByName("Administrators")).thenReturn(Optional.of(existingGroup));
+        when(environment.matchesProfiles("dev", "test", "e2e")).thenReturn(false);
+        when(passwordEncoder.encode(anyString())).thenReturn("$2a$10$stub");
+        ReflectionTestUtils.setField(initializer, "adminEmail", "admin@archiv.raddatz.cloud");
+        ReflectionTestUtils.setField(initializer, "adminPassword", "a-real-strong-password");
+
+        CommandLineRunner runner = initializer.initAdminUser(passwordEncoder);
+        runner.run();
+
+        // Group must not be re-inserted — that would violate user_groups_name_key.
+        verify(groupRepository, never()).save(any(UserGroup.class));
+        // But the admin user IS created, with the existing group attached.
+        org.mockito.ArgumentCaptor<AppUser> captor = org.mockito.ArgumentCaptor.forClass(AppUser.class);
+        verify(userRepository).save(captor.capture());
+        assertThat(captor.getValue().getGroups()).containsExactly(existingGroup);
+    }
+
+    @Test
+    void creates_Administrators_group_when_seeding_admin_on_a_fresh_database() throws Exception {
+        when(userRepository.findByEmail(anyString())).thenReturn(Optional.empty());
+        when(groupRepository.findByName("Administrators")).thenReturn(Optional.empty());
+        when(groupRepository.save(any(UserGroup.class))).thenAnswer(inv -> inv.getArgument(0));
+        when(environment.matchesProfiles("dev", "test", "e2e")).thenReturn(false);
+        when(passwordEncoder.encode(anyString())).thenReturn("$2a$10$stub");
+        ReflectionTestUtils.setField(initializer, "adminEmail", "admin@archiv.raddatz.cloud");
+        ReflectionTestUtils.setField(initializer, "adminPassword", "a-real-strong-password");
+
+        CommandLineRunner runner = initializer.initAdminUser(passwordEncoder);
+        runner.run();
+
+        // Group should be inserted exactly once.
+        verify(groupRepository).save(any(UserGroup.class));
+        verify(userRepository).save(any(AppUser.class));
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/user/AdminSeedPropertyKeyTest.java

@@ -0,0 +1,95 @@
+package org.raddatz.familienarchiv.user;
+
+import org.junit.jupiter.api.Test;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.beans.factory.config.YamlPropertiesFactoryBean;
+import org.springframework.boot.context.properties.bind.Binder;
+import org.springframework.boot.context.properties.source.ConfigurationPropertySources;
+import org.springframework.core.env.PropertiesPropertySource;
+import org.springframework.core.io.ClassPathResource;
+
+import java.lang.reflect.Field;
+import java.util.Properties;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+/**
+ * Pins the admin-seed property key contract. {@code UserDataInitializer} reads
+ * {@code @Value("${app.admin.email:...}")} and {@code @Value("${app.admin.password:...}")}.
+ * The yaml MUST expose those exact keys, not e.g. {@code app.admin.username}, or
+ * the env vars {@code APP_ADMIN_USERNAME} / {@code APP_ADMIN_PASSWORD} are
+ * silently ignored and the admin user gets seeded with the hardcoded defaults.
+ *
+ * <p>Discovered as a HIGH bug during the production-deploy bootstrap (#513): on
+ * first deploy the prod admin password is permanently locked to whatever ends
+ * up in the database, so a key-name mismatch would lock prod to the dev defaults
+ * {@code admin@familyarchive.local} / {@code admin123}.
+ *
+ * <p>No Spring context — Binder reads application.yaml directly.
+ */
+class AdminSeedPropertyKeyTest {
+
+    @Test
+    void admin_email_key_binds_from_yaml() {
+        Binder binder = binderFromApplicationYaml();
+
+        String email = binder.bind("app.admin.email", String.class)
+            .orElseThrow(() -> new AssertionError(
+                "app.admin.email is missing from application.yaml. "
+                + "UserDataInitializer reads this exact key; if the yaml uses "
+                + "a different name (e.g. 'username'), the env var "
+                + "APP_ADMIN_USERNAME is silently ignored."));
+
+        assertThat(email)
+            .as("app.admin.email must resolve from APP_ADMIN_USERNAME or its default")
+            .isNotBlank();
+    }
+
+    @Test
+    void admin_password_key_binds_from_yaml() {
+        Binder binder = binderFromApplicationYaml();
+
+        String password = binder.bind("app.admin.password", String.class)
+            .orElseThrow(() -> new AssertionError(
+                "app.admin.password is missing from application.yaml. "
+                + "UserDataInitializer reads this exact key."));
+
+        assertThat(password)
+            .as("app.admin.password must resolve from APP_ADMIN_PASSWORD or its default")
+            .isNotBlank();
+    }
+
+    @Test
+    void userDataInitializer_reads_app_admin_email_not_username() throws NoSuchFieldException {
+        // Pin the Java side too: a future rename of the @Value placeholder
+        // (e.g. back to `${app.admin.username:...}`) would silently break the
+        // binding while the yaml-side assertions above still pass. See #513.
+        Field field = UserDataInitializer.class.getDeclaredField("adminEmail");
+        Value annotation = field.getAnnotation(Value.class);
+        assertThat(annotation)
+            .as("UserDataInitializer.adminEmail must be @Value-annotated")
+            .isNotNull();
+        assertThat(annotation.value())
+            .as("UserDataInitializer must read app.admin.email — not username or any other key")
+            .startsWith("${app.admin.email:");
+    }
+
+    @Test
+    void userDataInitializer_reads_app_admin_password() throws NoSuchFieldException {
+        Field field = UserDataInitializer.class.getDeclaredField("adminPassword");
+        Value annotation = field.getAnnotation(Value.class);
+        assertThat(annotation).isNotNull();
+        assertThat(annotation.value())
+            .as("UserDataInitializer must read app.admin.password")
+            .startsWith("${app.admin.password:");
+    }
+
+    private Binder binderFromApplicationYaml() {
+        YamlPropertiesFactoryBean yaml = new YamlPropertiesFactoryBean();
+        yaml.setResources(new ClassPathResource("application.yaml"));
+        Properties props = yaml.getObject();
+        assertThat(props).as("application.yaml must be on the classpath").isNotNull();
+        return new Binder(ConfigurationPropertySources.from(
+            new PropertiesPropertySource("application", props)));
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/user/AppUserTest.java

@@ -35,4 +35,15 @@ class AppUserTest {
             .count();
         assertThat(distinct).isGreaterThan(1);
     }
+
+    @Test
+    void computeColor_returnsValidPaletteColorForIntegerMinValueHash() {
+        // UUID "80000000-0000-0000-0000-000000000000" has hashCode() == Integer.MIN_VALUE.
+        // Math.abs(Integer.MIN_VALUE) overflows back to Integer.MIN_VALUE (negative), making
+        // Math.abs(hashCode()) % n unsafe for palette sizes that don't evenly divide MIN_VALUE.
+        // Math.floorMod eliminates this edge case entirely.
+        UUID minHashId = UUID.fromString("80000000-0000-0000-0000-000000000000");
+        assertThat(minHashId.hashCode()).isEqualTo(Integer.MIN_VALUE);
+        assertThat(EXPECTED_PALETTE).contains(AppUser.computeColor(minHashId));
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/user/UserServiceTest.java

@@ -902,4 +902,18 @@ class UserServiceTest {
         assertThat(result.getName()).isEqualTo("Familie");
         assertThat(result.getPermissions()).containsExactlyInAnyOrder("READ_ALL", "WRITE_ALL");
     }
+
+    @Test
+    void createGroup_withNullPermissions_savesGroupWithEmptyPermissionSet() {
+        org.raddatz.familienarchiv.user.GroupDTO dto = new org.raddatz.familienarchiv.user.GroupDTO();
+        dto.setName("Leser");
+        dto.setPermissions(null);
+
+        UserGroup saved = UserGroup.builder().id(UUID.randomUUID()).name("Leser").build();
+        when(groupRepository.save(any())).thenReturn(saved);
+
+        userService.createGroup(dto);
+
+        verify(groupRepository).save(argThat(g -> g.getPermissions() != null && g.getPermissions().isEmpty()));
+    }
 }

docker-compose.prod.yml

@@ -0,0 +1,231 @@
+# Production / staging Docker Compose for Familienarchiv.
+#
+# This is a self-contained file (not an overlay over docker-compose.yml).
+# All services for the prod stack live here. Environment isolation is
+# achieved via the docker compose project name:
+#
+#   production: docker compose -f docker-compose.prod.yml -p archiv-production ...
+#   staging:    docker compose -f docker-compose.prod.yml -p archiv-staging --profile staging ...
+#
+# Volumes, networks and containers are namespaced by the project name,
+# so the two environments cohabit cleanly on the same host.
+#
+# Required env vars (provided by .env.production / .env.staging in CI):
+#   TAG                         image tag (release tag or "nightly")
+#   PORT_BACKEND, PORT_FRONTEND host-side ports (bound to 127.0.0.1 only)
+#   APP_DOMAIN                  e.g. archiv.raddatz.cloud / staging.raddatz.cloud
+#   POSTGRES_PASSWORD           Postgres password
+#   MINIO_PASSWORD              MinIO root password (admin operations only)
+#   MINIO_APP_PASSWORD          MinIO application service-account password
+#                               (least-privilege scope: archive bucket only)
+#   OCR_TRAINING_TOKEN          token guarding ocr-service /train endpoint
+#   APP_ADMIN_USERNAME          seeded admin email (e.g. admin@archiv.raddatz.cloud)
+#   APP_ADMIN_PASSWORD          seeded admin password — CRITICAL: locked in on
+#                               first deploy because UserDataInitializer only
+#                               creates the account if the email does not exist
+#   MAIL_HOST, MAIL_PORT,       SMTP relay (production only; staging uses mailpit)
+#   MAIL_USERNAME, MAIL_PASSWORD
+#   APP_MAIL_FROM               sender address (e.g. noreply@raddatz.cloud)
+
+networks:
+  archiv-net:
+    driver: bridge
+
+volumes:
+  postgres-data:
+  minio-data:
+  ocr-models:
+  ocr-cache:
+
+services:
+  db:
+    image: postgres:16-alpine
+    restart: unless-stopped
+    environment:
+      POSTGRES_USER: archiv
+      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
+      POSTGRES_DB: archiv
+    volumes:
+      - postgres-data:/var/lib/postgresql/data
+    networks:
+      - archiv-net
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U archiv -d archiv"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+
+  minio:
+    # Pinned MinIO release for reproducible deploys. Bumped manually until
+    # Renovate is bootstrapped for these production images (see follow-up issue).
+    image: minio/minio:RELEASE.2025-02-28T09-55-16Z
+    restart: unless-stopped
+    command: server /data --console-address ":9001"
+    environment:
+      MINIO_ROOT_USER: archiv
+      MINIO_ROOT_PASSWORD: ${MINIO_PASSWORD}
+    volumes:
+      - minio-data:/data
+    networks:
+      - archiv-net
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]
+      interval: 30s
+      timeout: 20s
+      retries: 3
+
+  # Idempotent bucket bootstrap + service-account creation.
+  # Runs once per `docker compose up` and exits 0. The entrypoint is
+  # extracted to infra/minio/bootstrap.sh so the (non-trivial) idempotent
+  # logic is readable, reviewable, and unit-testable as a script rather
+  # than YAML-escaped shell.
+  create-buckets:
+    # Custom image bakes bootstrap.sh in at build time. A bind-mount fails on
+    # the Docker-out-of-Docker production runner because the host daemon
+    # resolves the relative path against the host filesystem, not the
+    # runner container's CWD. See #506 + infra/minio/Dockerfile.
+    build:
+      context: ./infra/minio
+    # Declare one-shot intent so `docker compose up -d --wait` treats
+    # exited(0) as success rather than "not running, fail". Pair with
+    # backend's `service_completed_successfully` dependency below. See #510.
+    restart: "no"
+    depends_on:
+      minio:
+        condition: service_healthy
+    networks:
+      - archiv-net
+    environment:
+      MINIO_PASSWORD: ${MINIO_PASSWORD}
+      MINIO_APP_PASSWORD: ${MINIO_APP_PASSWORD}
+
+  # Dev-only mail catcher; gated behind the staging profile so production
+  # never starts it. Staging workflow runs with `--profile staging`.
+  mailpit:
+    # Pinned for reproducibility; bumped manually until Renovate is bootstrapped.
+    image: axllent/mailpit:v1.29.7
+    restart: unless-stopped
+    profiles: ["staging"]
+    networks:
+      - archiv-net
+    healthcheck:
+      # TCP-port open check via BusyBox `nc`. The previous wget-based probe
+      # introduced a non-obvious binary dependency on the mailpit image; a
+      # future tag that ships without wget would silently disable the
+      # healthcheck. `nc` is part of BusyBox in the upstream image.
+      test: ["CMD-SHELL", "nc -z localhost 8025 || exit 1"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+
+  ocr-service:
+    build:
+      context: ./ocr-service
+    restart: unless-stopped
+    expose:
+      - "8000"
+    # Surya OCR loads ~5GB of transformer models at startup; first request
+    # triggers a further ~1GB Kraken model download into ocr-cache.
+    # CX42+ (16 GB RAM) honours the default. On a CX32 (8 GB) override with
+    # OCR_MEM_LIMIT=6g (slower first-request, fits the host).
+    mem_limit: ${OCR_MEM_LIMIT:-12g}
+    memswap_limit: ${OCR_MEM_LIMIT:-12g}
+    volumes:
+      - ocr-models:/app/models
+      - ocr-cache:/root/.cache
+    environment:
+      KRAKEN_MODEL_PATH: /app/models/german_kurrent.mlmodel
+      TRAINING_TOKEN: ${OCR_TRAINING_TOKEN}
+      OCR_CONFIDENCE_THRESHOLD: "0.3"
+      OCR_CONFIDENCE_THRESHOLD_KURRENT: "0.5"
+      # SSRF allowlist pinned explicitly to the internal MinIO hostname.
+      # In prod the OCR service only fetches PDFs from MinIO over the
+      # docker network; localhost/127.0.0.1 are dev-only sources and
+      # must NOT be reachable here. Do not widen to `*`.
+      ALLOWED_PDF_HOSTS: "minio"
+    networks:
+      - archiv-net
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
+      interval: 10s
+      timeout: 5s
+      retries: 12
+      start_period: 120s
+
+  backend:
+    image: familienarchiv/backend:${TAG:-nightly}
+    build:
+      context: ./backend
+    restart: unless-stopped
+    depends_on:
+      db:
+        condition: service_healthy
+      minio:
+        condition: service_healthy
+      ocr-service:
+        condition: service_healthy
+      # Gate startup on the bucket bootstrap. Without this, backend
+      # starts in parallel with create-buckets and may race the policy
+      # bind. Also tells compose's `up -d --wait` that create-buckets
+      # is a one-shot that must complete successfully. See #510.
+      create-buckets:
+        condition: service_completed_successfully
+    # Bound to localhost only — Caddy fronts external traffic.
+    ports:
+      - "127.0.0.1:${PORT_BACKEND}:8080"
+    environment:
+      SPRING_DATASOURCE_URL: jdbc:postgresql://db:5432/archiv
+      SPRING_DATASOURCE_USERNAME: archiv
+      SPRING_DATASOURCE_PASSWORD: ${POSTGRES_PASSWORD}
+      # Application uses the bucket-scoped service account, not MinIO root.
+      S3_ENDPOINT: http://minio:9000
+      S3_ACCESS_KEY: archiv-app
+      S3_SECRET_KEY: ${MINIO_APP_PASSWORD}
+      S3_BUCKET_NAME: familienarchiv
+      S3_REGION: us-east-1
+      # No SPRING_PROFILES_ACTIVE — base application.yaml is production-ready
+      # (Swagger disabled, show-sql off, open-in-view false).
+      APP_BASE_URL: https://${APP_DOMAIN}
+      APP_ADMIN_USERNAME: ${APP_ADMIN_USERNAME}
+      APP_ADMIN_PASSWORD: ${APP_ADMIN_PASSWORD}
+      APP_OCR_BASE_URL: http://ocr-service:8000
+      APP_OCR_TRAINING_TOKEN: ${OCR_TRAINING_TOKEN}
+      MAIL_HOST: ${MAIL_HOST}
+      MAIL_PORT: ${MAIL_PORT:-587}
+      MAIL_USERNAME: ${MAIL_USERNAME:-}
+      MAIL_PASSWORD: ${MAIL_PASSWORD:-}
+      APP_MAIL_FROM: ${APP_MAIL_FROM:-noreply@raddatz.cloud}
+      SPRING_MAIL_PROPERTIES_MAIL_SMTP_AUTH: ${MAIL_SMTP_AUTH:-true}
+      SPRING_MAIL_PROPERTIES_MAIL_SMTP_STARTTLS_ENABLE: ${MAIL_STARTTLS_ENABLE:-true}
+    networks:
+      - archiv-net
+    healthcheck:
+      test: ["CMD-SHELL", "wget -qO- http://localhost:8080/actuator/health | grep -q UP || exit 1"]
+      interval: 15s
+      timeout: 5s
+      retries: 10
+      start_period: 30s
+
+  frontend:
+    image: familienarchiv/frontend:${TAG:-nightly}
+    build:
+      context: ./frontend
+      target: production
+    restart: unless-stopped
+    depends_on:
+      backend:
+        condition: service_healthy
+    ports:
+      - "127.0.0.1:${PORT_FRONTEND}:3000"
+    environment:
+      # SSR fetches go inside the docker network; clients hit https://${APP_DOMAIN}
+      API_INTERNAL_URL: http://backend:8080
+      ORIGIN: https://${APP_DOMAIN}
+    networks:
+      - archiv-net
+    healthcheck:
+      test: ["CMD-SHELL", "wget -qO- http://127.0.0.1:3000/login >/dev/null 2>&1 || exit 1"]
+      interval: 15s
+      timeout: 5s
+      retries: 10
+      start_period: 20s

docker-compose.yml

@@ -13,7 +13,7 @@ services:
     ports:
       - "${PORT_DB}:5432"
     networks:
-      - archive-net
+      - archiv-net
     healthcheck:
       test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER} -d ${POSTGRES_DB}"]
       interval: 5s
@@ -35,7 +35,7 @@ services:
       - "${PORT_MINIO_API}:9000"      # API Port
       - "${PORT_MINIO_CONSOLE}:9001"  # Web-Oberfläche
     networks:
-      - archive-net
+      - archiv-net
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]
       interval: 30s
@@ -56,7 +56,7 @@ services:
       exit 0;
       "
     networks:
-      - archive-net
+      - archiv-net
 
   # --- Mail catcher: Mailpit (dev only) ---
   # Catches all outgoing emails and displays them in a web UI.
@@ -69,7 +69,7 @@ services:
       - "${PORT_MAILPIT_UI:-8025}:8025"   # Web UI
       - "${PORT_MAILPIT_SMTP:-1025}:1025" # SMTP
     networks:
-      - archive-net
+      - archiv-net
 
   # --- OCR: Python microservice (Surya + Kraken) ---
   # Single-node only: OCR training reloads the model in-process after each run.
@@ -99,7 +99,7 @@ services:
       OCR_CLAHE_TILE_SIZE: "8"      # CLAHE tile grid size (NxN tiles per page)
       OCR_MAX_CACHED_MODELS: "2"    # LRU cache; each model ~500 MB, so 2 = ~1 GB resident
     networks:
-      - archive-net
+      - archiv-net
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
       interval: 10s
@@ -150,7 +150,7 @@ services:
     ports:
       - "${PORT_BACKEND}:8080"
     networks:
-      - archive-net
+      - archiv-net
     healthcheck:
       test: ["CMD-SHELL", "wget -qO- http://localhost:8080/actuator/health | grep -q UP || exit 1"]
       interval: 15s
@@ -163,6 +163,7 @@ services:
     build:
       context: ./frontend
       dockerfile: Dockerfile
+      target: development  # Dockerfile is multi-stage; default would be the production stage
     container_name: archive-frontend
     restart: unless-stopped
     depends_on:
@@ -184,10 +185,10 @@ services:
     ports:
       - "${PORT_FRONTEND}:5173"
     networks:
-      - archive-net
+      - archiv-net
 
 networks:
-  archive-net:
+  archiv-net:
     driver: bridge
 
 volumes:

docs/DEPLOYMENT.md

@@ -27,20 +27,22 @@ This doc is the Day-1 checklist and operational reference. It links to the canon
 ```mermaid
 graph TD
     Browser -->|HTTPS| Caddy["Caddy (TLS termination)"]
-    Caddy -->|HTTP :5173| Frontend["Web Frontend\nSvelteKit / Node.js"]
+    Caddy -->|HTTP :3000| Frontend["Web Frontend\nSvelteKit Node adapter"]
     Caddy -->|HTTP :8080| Backend["API Backend\nSpring Boot / Jetty :8080"]
     Backend -->|JDBC :5432| DB[(PostgreSQL 16)]
-    Backend -->|S3 API :9000| MinIO[(MinIO / Hetzner OBS)]
+    Backend -->|S3 API :9000| MinIO[(MinIO)]
     Backend -->|HTTP :8000 internal| OCR["OCR Service\nPython FastAPI"]
     OCR -->|presigned URL| MinIO
-    Browser -->|SSE direct| Backend
+    Caddy -->|SSE proxy_pass| Backend
 ```
 
 **Key facts:**
-- Caddy terminates TLS and reverse-proxies to frontend and backend. See the Caddyfile in [`docs/infrastructure/production-compose.md`](infrastructure/production-compose.md).
-- The OCR service has **no external port** — reachable only on the internal Docker network from the backend.
-- SSE notifications go directly backend → browser (not via the SvelteKit SSR layer).
-- Management port 8081 (Spring Actuator / Prometheus scrape) is internal only — the Caddy config blocks `/actuator/*` externally.
+- Caddy terminates TLS and reverse-proxies to frontend (`:3000`) and backend (`:8080`). The Caddyfile is committed at [`infra/caddy/Caddyfile`](../infra/caddy/Caddyfile) and is installed on the host as `/etc/caddy/Caddyfile` (symlink).
+- The host binds all docker-published ports to `127.0.0.1` only; Caddy is the sole external entry point.
+- The OCR service has **no published port** — reachable only on the internal Docker network from the backend.
+- SSE notifications transit Caddy (browser → Caddy → backend); the backend is never reachable directly from the public internet. The SvelteKit SSR layer is bypassed for SSE, but Caddy is not.
+- The Caddyfile responds `404` on `/actuator/*` (defense in depth). Internal monitoring scrapes the backend on the docker network, not through Caddy.
+- Production and staging cohabit on the same host via docker compose project names: `archiv-production` (ports 8080/3000) and `archiv-staging` (ports 8081/3001).
 
 ### OCR memory requirements
 
@@ -52,19 +54,23 @@ The OCR service requires significant RAM for model loading. The dev compose sets
 | Hetzner CX32 | 8 GB | 6 GB | Accept reduced batch sizes and slower throughput |
 | Hetzner CX22 | 4 GB | — | Disable the OCR service (`profiles: [ocr]`); run OCR on demand only |
 
-A CX32 cannot honour a `mem_limit: 12g` — set it to `6g` in the prod overlay or use CX42.
+A CX32 cannot honour the default `mem_limit: 12g` — set the `OCR_MEM_LIMIT=6g` env var (in `.env.production` / `.env.staging`, or as a Gitea secret consumed by the workflow) before deploying on a CX32. The prod compose interpolates this var with a 12g default.
 
 ### Dev vs production differences
 
-| Concern | Dev compose | Prod overlay |
+| Concern | Dev (`docker-compose.yml`) | Prod (`docker-compose.prod.yml`) |
 |---|---|---|
-| MinIO image tag | `minio/minio:latest` (unpinned) | Pinned in prod overlay |
-| Data persistence | Bind mounts `./data/postgres`, `./data/minio` | Named Docker volumes |
-| Bucket creation | `create-buckets` helper container | Pre-created in Hetzner console |
-| Spring profile | `dev,e2e` (enables OpenAPI + Swagger UI) | `prod` |
-| Mail | Mailpit (local catcher) | Real SMTP |
+| MinIO image tag | `minio/minio:latest` | Pinned `minio/minio:RELEASE.…` |
+| Data persistence | Bind mounts `./data/postgres`, `./data/minio` | Named Docker volumes (`postgres-data`, `minio-data`) |
+| MinIO credentials for backend | Root user/password | Service account `archiv-app` with bucket-scoped rights |
+| Bucket creation | `create-buckets` helper | Same helper, plus service-account bootstrap on every up |
+| Spring profile | `dev,e2e` (Swagger + e2e overrides) | unset — base `application.yaml` is production-ready |
+| Mail | Mailpit (local catcher) | Real SMTP (production) / Mailpit via `profiles: [staging]` (staging) |
+| Frontend image | Dev server, `target: development`, port 5173 | Node adapter, `target: production`, port 3000 |
+| Host port binding | All published | Bound to `127.0.0.1` only; Caddy is the front door |
+| Deploy method | `docker compose up -d` (manual) | Gitea Actions: `nightly.yml` (staging, cron) and `release.yml` (production, on `v*` tag) — both use `up -d --wait` |
 
-Full prod overlay: [`docs/infrastructure/production-compose.md`](infrastructure/production-compose.md).
+Full prod compose: [`docker-compose.prod.yml`](../docker-compose.prod.yml). Workflow files: [`.gitea/workflows/nightly.yml`](../.gitea/workflows/nightly.yml), [`.gitea/workflows/release.yml`](../.gitea/workflows/release.yml).
 
 ---
 
@@ -112,9 +118,10 @@ All vars are set in `.env` at the repo root (copy from `.env.example`). The back
 
 | Variable | Purpose | Default | Required? | Sensitive? |
 |---|---|---|---|---|
-| `MINIO_ROOT_USER` | MinIO root username | `minio_admin` | YES | — |
-| `MINIO_ROOT_PASSWORD` | MinIO root password | `change-me` | YES | YES |
-| `MINIO_DEFAULT_BUCKETS` | Bucket name | `archive-documents` | YES | — |
+| `MINIO_ROOT_USER` | MinIO root username (dev compose only — prod compose hardcodes `archiv`) | `minio_admin` | YES (dev) | — |
+| `MINIO_ROOT_PASSWORD` / `MINIO_PASSWORD` | MinIO root password. **Used only by the `mc admin` bootstrap in prod, never by the backend.** | `change-me` | YES | YES |
+| `MINIO_APP_PASSWORD` | Password for the `archiv-app` service account that the backend uses. Bucket-scoped via `readwrite` policy on `familienarchiv`. Bootstrapped by `create-buckets`. | — | YES (prod) | YES |
+| `MINIO_DEFAULT_BUCKETS` | Bucket name (dev compose only — prod compose hardcodes `familienarchiv`) | `archive-documents` | YES (dev) | — |
 
 ### OCR service
 
@@ -124,53 +131,105 @@ All vars are set in `.env` at the repo root (copy from `.env.example`). The back
 | `ALLOWED_PDF_HOSTS` | SSRF protection — comma-separated list of allowed PDF source hosts. **Do not widen to `*`** | `minio,localhost,127.0.0.1` | YES | — |
 | `KRAKEN_MODEL_PATH` | Directory containing Kraken HTR models (populated by `download-kraken-models.sh`) | `/app/models/` | — | — |
 | `BLLA_MODEL_PATH` | Kraken baseline layout analysis model path | `/app/models/blla.mlmodel` | — | — |
+| `OCR_MEM_LIMIT` | Container memory cap for ocr-service in `docker-compose.prod.yml`. Set to `6g` on CX32 hosts; leave unset on CX42+ to use the 12g default | `12g` (prod compose default) | — | — |
 
 ---
 
 ## 3. Bootstrap from scratch
 
-> Full VPS provisioning steps are in [`docs/infrastructure/production-compose.md`](infrastructure/production-compose.md). This section covers the sequence and the security-critical steps.
+Production and staging deploy via Gitea Actions (`release.yml` on `v*` tag, `nightly.yml` on cron). The server itself only needs to host Caddy, Docker, and the runner — the workflows handle the rest.
 
-### Security checklist — complete before first boot
-
-> ⚠️ **These defaults ship in `.env.example` and `application.yaml`. Change them or you will have an insecure installation.**
-
-- [ ] Set `APP_ADMIN_PASSWORD` (default: `admin123` — change before starting the backend)
-- [ ] Set `APP_ADMIN_USERNAME` if you want a non-default admin login name (add to `.env` — not in `.env.example`)
-- [ ] Rotate `POSTGRES_PASSWORD` from `change-me`
-- [ ] Rotate `MINIO_ROOT_PASSWORD` from `change-me`
-- [ ] Set a strong `APP_OCR_TRAINING_TOKEN` (backend) and the matching `TRAINING_TOKEN` (OCR service) — both must be the same value (`python3 -c "import secrets; print(secrets.token_hex(32))"`)
-- [ ] Confirm `ALLOWED_PDF_HOSTS` is locked to your MinIO/S3 hostname — widening to `*` opens SSRF
-- [ ] Set `SPRING_PROFILES_ACTIVE=prod` in the prod overlay (not `dev,e2e` — that exposes Swagger UI and `/v3/api-docs`)
-- [ ] Use a dedicated MinIO service account for `S3_ACCESS_KEY` / `S3_SECRET_KEY`, not the root credentials
-
-### Bootstrap sequence
+### 3.1 Server one-time setup
 
 ```bash
-# 1. Copy and fill the env file
-cp .env.example .env
-# edit .env — complete the security checklist above first
+# Base hardening
+ufw default deny incoming && ufw allow 22/tcp && ufw allow 80/tcp && ufw allow 443/tcp && ufw enable
+# /etc/ssh/sshd_config: PasswordAuthentication no, PermitRootLogin no
 
-# 2. (Production only) Create the MinIO / Hetzner OBS bucket in the console
-#    The dev compose has a create-buckets helper; production does not.
-#    Create the bucket named $MINIO_DEFAULT_BUCKETS with private access.
+# Install Caddy 2 (https://caddyserver.com/docs/install#debian-ubuntu-raspbian)
+apt install caddy
 
-# 3. Start the stack (prod overlay — see docs/infrastructure/production-compose.md)
-#    docker-compose.prod.yml is NOT committed — create it from the guide above
-docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d
+# Use the Caddyfile from the repo (replace path with the runner's clone target)
+# CI DEPENDENCY: the nightly and release workflows run `systemctl reload caddy` to
+# pick up committed Caddyfile changes. They find the file via this symlink — if it
+# is absent or points elsewhere, the reload succeeds but serves stale config.
+ln -sf /opt/familienarchiv/infra/caddy/Caddyfile /etc/caddy/Caddyfile
+systemctl reload caddy
 
-# 4. Flyway migrations run automatically on backend start.
-#    Watch the backend log to confirm:
-docker compose logs --follow --tail=100 backend
+# fail2ban — protect /api/auth/login from credential stuffing.
+# Jail watches the Caddy JSON access log for 401 responses on
+# /api/auth/login. The jail (maxretry=10 / findtime=10m / bantime=30m)
+# and filter are committed under infra/fail2ban/ — symlink them in:
+apt install fail2ban
+ln -sf /opt/familienarchiv/infra/fail2ban/jail.d/familienarchiv.conf \
+       /etc/fail2ban/jail.d/familienarchiv.conf
+ln -sf /opt/familienarchiv/infra/fail2ban/filter.d/familienarchiv-auth.conf \
+       /etc/fail2ban/filter.d/familienarchiv-auth.conf
+systemctl reload fail2ban
+# Verify after first deploy with:
+#   fail2ban-client status familienarchiv-auth
+#   fail2ban-regex /var/log/caddy/access.log familienarchiv-auth
 
-# 5. Verify the stack is healthy
-curl http://localhost:8080/actuator/health
-# Expected: {"status":"UP"}
+# Tailscale — used by the backup pipeline to reach heim-nas (follow-up issue)
+curl -fsSL https://tailscale.com/install.sh | sh && tailscale up
 
-# 6. Open the app and log in with the admin credentials from .env
+# Self-hosted Gitea runner — register against the repo with a runner token.
+# This runner is assumed single-tenant: the deploy workflows write .env.*
+# files to disk during execution (cleaned up unconditionally on completion).
+# A multi-tenant runner would need to switch to stdin-piped env files.
+# (See https://docs.gitea.com/usage/actions/quickstart for the register step.)
 ```
 
-> **Do not use `docker-compose.ci.yml` locally** — it disables bind mounts that the dev workflow depends on.
+### 3.2 DNS records
+
+```
+archiv.raddatz.cloud   A   <server IP>
+staging.raddatz.cloud  A   <server IP>
+git.raddatz.cloud      A   <server IP>
+```
+
+### 3.3 Gitea secrets (Repo → Settings → Actions → Secrets)
+
+| Secret | Used by | Notes |
+|---|---|---|
+| `PROD_POSTGRES_PASSWORD` | release.yml | strong unique password |
+| `PROD_MINIO_PASSWORD` | release.yml | MinIO root password; used only at bootstrap |
+| `PROD_MINIO_APP_PASSWORD` | release.yml | application service-account password |
+| `PROD_OCR_TRAINING_TOKEN` | release.yml | `python3 -c "import secrets; print(secrets.token_hex(32))"` |
+| `PROD_APP_ADMIN_USERNAME` | release.yml | e.g. `admin@archiv.raddatz.cloud` |
+| `PROD_APP_ADMIN_PASSWORD` | release.yml | **⚠ locked permanently on first deploy** — see §3.5 |
+| `STAGING_POSTGRES_PASSWORD` | nightly.yml | different from prod |
+| `STAGING_MINIO_PASSWORD` | nightly.yml | different from prod |
+| `STAGING_MINIO_APP_PASSWORD` | nightly.yml | different from prod |
+| `STAGING_OCR_TRAINING_TOKEN` | nightly.yml | different from prod |
+| `STAGING_APP_ADMIN_USERNAME` | nightly.yml | e.g. `admin@staging.raddatz.cloud` |
+| `STAGING_APP_ADMIN_PASSWORD` | nightly.yml | locked on first staging deploy |
+| `MAIL_HOST` | release.yml | SMTP relay hostname (prod only) |
+| `MAIL_PORT` | release.yml | typically `587` |
+| `MAIL_USERNAME` | release.yml | SMTP user |
+| `MAIL_PASSWORD` | release.yml | SMTP password |
+
+### 3.4 First deploy
+
+```bash
+# 1. Trigger nightly.yml manually (Repo → Actions → nightly → "Run workflow")
+#    Expected: docker compose up -d --wait succeeds for archiv-staging, then
+#    the workflow's "Smoke test deployed environment" step asserts:
+#      - https://staging.raddatz.cloud/login returns 200
+#      - HSTS header is present
+#      - /actuator/health returns 404 (defense-in-depth check)
+# 2. (Optional) Re-verify manually
+curl -I https://staging.raddatz.cloud/
+#    Expected: 200 (login page) with HSTS + X-Content-Type-Options headers
+# 3. When staging looks healthy, push a v* tag to trigger release.yml
+git tag v1.0.0 && git push origin v1.0.0
+```
+
+### 3.5 ⚠ Admin password is locked on first deploy
+
+`UserDataInitializer` creates the admin user **only if the email does not exist**. The first successful deploy persists the admin password to the database. Changing `PROD_APP_ADMIN_PASSWORD` in Gitea secrets after that point has **no effect** — the secret is only consulted when the row is missing.
+
+Before the first deploy: rotate `PROD_APP_ADMIN_PASSWORD` to a strong value. After the first deploy: change the admin password via the in-app account settings, not via the Gitea secret.
 
 ---
 
@@ -224,7 +283,23 @@ docker exec -i archive-db psql -U ${POSTGRES_USER} ${POSTGRES_DB} < backup-YYYYM
 
 ### Planned — phase 5 of Production v1 milestone
 
-Automated backup (PostgreSQL WAL archiving + MinIO bucket replication) is planned in the Production v1 milestone phase 5. Until that ships: **manual backups are the only recovery option.**
+Automated backup (nightly `pg_dump` + MinIO `mc mirror` over Tailscale to `heim-nas`) is a follow-up issue. Until that ships: **manual backups are the only recovery option.**
+
+### Rollback
+
+Each release tag corresponds to a docker image tag on the host daemon (built via DooD; no registry). Rolling back to a previous tag is one command:
+
+```bash
+TAG=v1.0.0 docker compose \
+  -f docker-compose.prod.yml \
+  -p archiv-production \
+  --env-file /opt/familienarchiv/.env.production \
+  up -d --wait --remove-orphans
+```
+
+If the rollback target image is no longer present on the host (host disk pruned, etc.), re-trigger `release.yml` for that tag from Gitea Actions UI — it rebuilds and redeploys.
+
+**Flyway migrations are not auto-rolled-back.** If a release contained a destructive migration (drop column, rename table), a tag rollback brings the schema back to a previous app version but the data shape has already changed. For breaking schema changes, prefer a forward-only fix.
 
 ---
 

docs/GLOSSARY.md

@@ -107,6 +107,13 @@ _See also [Briefwechsel](#briefwechsel-user-facing)._
 
 ---
 
+## Infrastructure Terms
+
+**archiv-app** — the bucket-scoped MinIO service account the backend uses to read and write the `familienarchiv` bucket. Distinct from the MinIO root account (`archiv`, used only by the bootstrap container for admin operations). Defined and provisioned in [`infra/minio/bootstrap.sh`](../infra/minio/bootstrap.sh) and consumed by the backend as `S3_ACCESS_KEY` in [`docker-compose.prod.yml`](../docker-compose.prod.yml). The attached `archiv-app-policy` grants `s3:GetObject/PutObject/DeleteObject` on `familienarchiv/*` and `s3:ListBucket/GetBucketLocation` on the bucket only — not the built-in `readwrite` policy which would grant `s3:*` on all buckets.
+_See also [ADR-010 — MinIO stays self-hosted, not Hetzner OBS](./adr/010-minio-self-hosted-not-hetzner-obs.md)._
+
+---
+
 ## Pending Terms
 
 _Terms flagged as potentially ambiguous that have not yet been formally defined here. Add an entry above and remove it from this list when resolved._

docs/adr/008-fts-sql-pagination.md

@@ -0,0 +1,68 @@
+# ADR-008: SQL-level pagination for full-text search via window-function CTE
+
+## Status
+
+Accepted
+
+## Context
+
+`DocumentRepository.findAllMatchingIdsByFts` (formerly `findRankedIdsByFts`) returns all matching document IDs for a FTS query. `DocumentService.searchDocuments` then paginates in memory on the RELEVANCE sort path.
+
+A pre-production audit against 1,520 documents measured:
+
+```
+rows_per_call: 911 / call  (query: "walter")
+```
+
+At current scale this is acceptable — 911 UUIDs ≈ 14 KB, ms-level DB time. At 100 K+ documents two failure modes emerge:
+
+1. **Memory**: a broad query returns ~60 K UUIDs ≈ 1 MB per request, multiplied by concurrent users.
+2. **Latency**: the `LATERAL` join does work proportional to match-set size; at 60 K matches the FTS step alone exceeds 100 ms per query.
+
+Tracked as finding **F-31 (High)** in the pre-production architectural review.
+
+## Decision
+
+Push pagination and rank ordering into SQL for the RELEVANCE sort path when no non-text filters are active (pure full-text search):
+
+```sql
+WITH q AS (
+  SELECT CASE WHEN websearch_to_tsquery('german', :query)::text <> ''
+              THEN to_tsquery('simple', regexp_replace(
+                       websearch_to_tsquery('german', :query)::text,
+                       '''([^'']+)''', '''\\1'':*', 'g'))
+         END AS pq
+), matches AS (
+  SELECT d.id, ts_rank(d.search_vector, q.pq) AS rank
+  FROM documents d, q
+  WHERE d.search_vector @@ q.pq
+)
+SELECT id, rank, COUNT(*) OVER () AS total
+FROM matches
+ORDER BY rank DESC, id
+OFFSET :offset LIMIT :limit
+```
+
+`COUNT(*) OVER ()` returns the full match count alongside each page row in a single round-trip — no separate count query needed.
+
+`rows_per_call` for the FTS query drops from match-set size (911) to page size (≤ 50).
+
+When non-text filters (date range, sender, receiver, tags, status) are also active, the existing path is preserved: `findAllMatchingIdsByFts` returns all ranked IDs, which are passed as an `IN` clause to the JPA Specification, and `totalElements` comes from the JPA `Page.getTotalElements()`. This keeps the count accurate across the combined filter set.
+
+## Alternatives Considered
+
+**1. Two-query approach (separate COUNT + paged SELECT)**
+Correct, but doubles round-trips. The window function achieves the same result in one query.
+
+**2. Capped result set with a user-visible warning**
+Return at most N results (e.g. 500) and show "showing top 500 of many results". Simpler, but degrades UX for broad queries and doesn't reduce latency proportionally (still scans N rows).
+
+**3. Full SQL rewrite combining FTS + JPA Specification filters**
+Possible via a native query that embeds all filter predicates. Eliminates the in-memory SENDER/RECEIVER sort paths and the two-phase approach. High complexity, tight coupling to schema details, loses type-safe JPA Specification composition. Deferred to a future refactor if scale demands it.
+
+## Consequences
+
+- **`rows_per_call` for pure-text FTS searches drops to ≤ page size** — the primary metric.
+- **SENDER and RECEIVER sort paths stay in-memory** for combined text+filter queries. For pure-text queries with SENDER/RECEIVER sort, the current approach (fetch all matched IDs, build spec, load all matched entities, sort in-memory) still runs. This is acceptable while the archive stays under ~10 K documents.
+- **RELEVANCE sort with text+filters still loads the full filtered entity set in-memory.** The filtered set is typically much smaller than the raw FTS match set, so the cost is bounded by filter selectivity, not total match count.
+- **`findAllMatchingIdsByFts` is retained** for: (a) the bulk-edit "select all" fast path (`findIdsForFilter`), (b) the document density chart (`getDensity`), and (c) the SENDER/RECEIVER in-memory sort paths.

docs/adr/009-standalone-compose-not-overlay.md

@@ -0,0 +1,50 @@
+# ADR-009: Standalone `docker-compose.prod.yml`, not an overlay
+
+## Status
+
+Accepted
+
+## Context
+
+The repository's `docker-compose.yml` is a development stack: every service is built locally, ports are exposed on `0.0.0.0` for dev tooling, the frontend runs `npm run dev` with hot-reload, the backend is `spring-boot:run` with the dev profile, and there is no Caddy, no `archiv-app` service account, no admin-credential lock-in, no healthcheck-gated startup sequence. The dev stack reflects "single developer on a laptop", not "production on a single VPS".
+
+The pre-merge design (issue #497, comment #8331) sketched two ways to add a production stack:
+
+1. **Overlay** — keep `docker-compose.yml` as the base, add `docker-compose.prod.yml` as a `-f` overlay (`docker compose -f docker-compose.yml -f docker-compose.prod.yml up`). Compose merges the two files at runtime.
+2. **Standalone** — make `docker-compose.prod.yml` a fully self-contained file that does not reference or merge with `docker-compose.yml` at all. Project-name namespacing (`-p archiv-production`, `-p archiv-staging`) keeps multi-environment deploys clean on a single host.
+
+The earlier `docs/infrastructure/production-compose.md` notes assumed overlay because the original plan was to **remove** MinIO in production (replace with Hetzner Object Storage), so the prod file would only need to remove one service and add a few. With MinIO retained (see ADR-010), the prod stack diverges from dev in essentially every service: build vs pre-built image, target stage, port binding, env vars, healthcheck, restart policy, mem_limit, profile gating, service account, depends_on chain. Overlay would mostly be `override:` blocks that nullify the dev defaults — a fragile inversion.
+
+## Decision
+
+`docker-compose.prod.yml` is standalone. Production and staging both run it directly:
+
+```
+production: docker compose -f docker-compose.prod.yml -p archiv-production --env-file .env.production ...
+staging:    docker compose -f docker-compose.prod.yml -p archiv-staging    --env-file .env.staging --profile staging ...
+```
+
+Environment isolation is achieved via the Docker Compose project name (`-p`). Volumes, networks, and containers are namespaced by the project name, so production and staging cohabit cleanly on the same host without interfering.
+
+The dev `docker-compose.yml` is unchanged — `docker compose up` still works for developers, and its `frontend` service now specifies `target: development` explicitly so the new multi-stage Dockerfile builds the right stage.
+
+## Alternatives Considered
+
+| Alternative | Why rejected |
+|---|---|
+| Overlay (`-f base.yml -f prod.yml`) | With MinIO retained and most services differing across nearly every field, the overlay would consist mostly of `override:` blocks that null out dev defaults. Compose's merge semantics for nested keys (env, ports, healthcheck) are sharp — silent merges of port mappings, env-var entries, and depends_on edges cost reviewer hours. Standalone is one file the reader can hold in their head. |
+| Two fully separate files (dev + prod) but with shared YAML anchors via `extends:` | `extends:` works across files but is a niche feature and is increasingly discouraged in compose v2. Reviewer load is higher than reading two flat files. |
+| Generate prod compose from a template at deploy time (e.g. ytt, kustomize) | Adds a build-time step and a new tool to the operator toolchain. Justified for a fleet of 10+ environments; overkill for production + staging on one host. |
+| Single compose file with environment-specific profiles | Compose profiles select which *services* run, not which *configuration* a service runs with. Using profiles to swap "build locally" vs "pull image" would smear dev and prod across one file. |
+
+## Consequences
+
+- The prod file can be read top-to-bottom without cross-referencing `docker-compose.yml`. Onboarding and review cost drops.
+- Volume namespacing is automatic (`archiv-production_postgres-data`, `archiv-staging_postgres-data`) — no manual `volumes:` aliasing.
+- Dev compose churn (e.g. swapping a dev port) cannot accidentally affect production. The two files are independent.
+- The cost is duplication: identical environment variables (e.g. `POSTGRES_DB: archiv`) appear in both files. This duplication is bounded — there is no incentive to add more services that exist in both — and the alternative (overlay) carries its own duplication via `override:` boilerplate.
+- The retired `docs/infrastructure/production-compose.md` narrative is trimmed to a pointer at the live files. The cost/sizing rationale is preserved there.
+
+## Future Direction
+
+If the deployment fleet ever grows beyond two environments on one host (e.g. add a `demo` environment, or shard staging across two VPS for load testing), revisit the templating decision. At three+ environments the duplication starts to bite and a template engine (kustomize or ytt) becomes attractive.

docs/adr/010-minio-self-hosted-not-hetzner-obs.md

@@ -0,0 +1,53 @@
+# ADR-010: MinIO stays self-hosted on the production VPS
+
+## Status
+
+Accepted
+
+## Context
+
+`docs/infrastructure/production-compose.md` (pre-this-PR) sketched a production topology in which the application bucket migrates from in-cluster MinIO to Hetzner Object Storage (OBS, S3-compatible). The motivation was operational: one less service to back up, no MinIO RAM/disk pressure on the VPS, hand off durability to the hyperscaler.
+
+Two facts revisited at pre-merge review (issue #497, comment #8331) changed the answer:
+
+1. **Current data size is small.** The archive is ~13 GB of file uploads (Kurrent letters, scanned ODS files, attachment PDFs). Hetzner OBS billing on this size is dominated by the per-month base fee (~5 EUR/mo for the smallest unit), not capacity or egress. The break-even point against the VPS's existing disk is far above the current footprint.
+2. **MinIO is already production-grade.** The dev stack uses MinIO; the backend already drives it via the AWS SDK v2 with a generic `S3_ENDPOINT`. Switching providers is a runtime env-var change (`S3_ENDPOINT`, `S3_ACCESS_KEY`, `S3_SECRET_KEY`) plus an `mc mirror` to copy objects. There is no application-level rewrite cost waiting.
+
+If Hetzner OBS were a one-way-door (provider-specific SDK, complex IAM integration, multi-month migration), the decision would deserve a serious weighing. As reversible as the migration is, deferring it costs nothing.
+
+## Decision
+
+MinIO stays on the production VPS for the first launch. The application bucket is created and managed inside the docker-compose stack (`infra/minio/bootstrap.sh`). The backend uses a least-privilege service account (`archiv-app`) with a bucket-scoped IAM policy, not the MinIO root credentials.
+
+Hetzner Object Storage is **explicitly deferred**, not rejected. The migration path is documented as a runbook in `docs/DEPLOYMENT.md` (when the trigger fires): provision an OBS bucket, run `mc mirror local-minio:/familienarchiv obs:/familienarchiv`, rotate the three env vars, restart the backend, decommission the MinIO service from `docker-compose.prod.yml`.
+
+## Triggers to re-evaluate
+
+Revisit the decision when **any** of the following holds:
+
+- The `minio-data` volume exceeds 50 GB and is growing > 5 GB/month.
+- MinIO healthcheck latency exceeds 200 ms p95 (signal of disk pressure on the host).
+- The VPS upgrade required to keep MinIO healthy costs more per month than the equivalent OBS bucket + traffic.
+- Backup of the MinIO volume to `heim-nas` over Tailscale (deferred follow-up) is implemented and consistently runs > 30 min nightly. At that point durability-as-a-service starts paying for itself.
+
+The migration runbook in `docs/DEPLOYMENT.md` is the script for executing the swap when one of the triggers fires.
+
+## Alternatives Considered
+
+| Alternative | Why rejected (for now) |
+|---|---|
+| Migrate to Hetzner Object Storage in this PR | Premature. Adds an external dependency, locks the operator into the Hetzner ecosystem before the data has demonstrated it needs hyperscaler durability, blocks the PR on a migration that buys ~5 GB of headroom. |
+| Migrate to S3 (AWS) for HA across regions | Way over-spec for a family archive. Egress cost would dwarf any benefit; durability concerns at this size are addressed by nightly off-site backup, not by multi-region replication. |
+| Drop S3 abstraction entirely; store files directly on the VPS disk | Possible, but loses the bucket-policy IAM surface (least-privilege service account), loses presigned-URL flow (OCR service downloads files via short-lived URLs, not via shared filesystem), loses the migration path to OBS. The S3 indirection is cheap insurance. |
+| Self-hosted on-VPS plus periodic `mc mirror` to Hetzner OBS for off-site backup | This is the **target** for the backup pipeline follow-up. Treated as backup, not primary — primary stays MinIO. |
+
+## Consequences
+
+- The production VPS sizing (Hetzner CX42, 16 GB RAM, 80 GB disk) must accommodate MinIO's working set. Current footprint leaves ample headroom.
+- Backup of MinIO data is the operator's responsibility until the off-site `mc mirror` pipeline is implemented (deferred follow-up). The DEPLOYMENT.md rollback procedure explicitly flags this — manual backup is the only recovery option until the pipeline ships.
+- The backend never sees the MinIO root password; it uses the `archiv-app` service account with a bucket-scoped IAM policy (see `infra/minio/bootstrap.sh`). A backend RCE/SSRF cannot escalate beyond the `familienarchiv` bucket.
+- The migration to Hetzner OBS remains a small, well-understood runbook step rather than a major refactor. No application code, no SDK swap.
+
+## Future Direction
+
+When one of the triggers above fires, the migration is: provision OBS bucket → `mc mirror` → rotate three env vars → restart backend → remove MinIO service from compose. The bucket-scoped policy translates 1:1 to an OBS user policy (S3-compatible).

docs/adr/011-single-tenant-gitea-runner.md

@@ -0,0 +1,58 @@
+# ADR-011: Single-tenant Gitea runner with secrets-on-disk env-files
+
+## Status
+
+Accepted
+
+## Context
+
+The deploy workflows (`.gitea/workflows/nightly.yml`, `release.yml`) execute on a self-hosted Gitea Actions runner. The runner has Docker-out-of-Docker access (the host's Docker socket is mounted into the runner), so `docker compose build` produces images on the host daemon and `docker compose up` consumes them directly — no registry hop.
+
+Two workflow steps shape the security model:
+
+1. **"Write env file"** — the workflow writes every required secret to `.env.staging` or `.env.production` on the runner's filesystem so that `docker compose --env-file` can consume them. The file lives on disk for the duration of the workflow.
+2. **"Cleanup env file"** — the matching `if: always()` step deletes the env file after the workflow ends, regardless of success.
+
+This shape only works under one operational assumption: **the runner is single-tenant**. The runner is owned by the same operator who owns the secrets, no other repositories run jobs on the same runner, and no untrusted code is executed (no public fork PRs trigger workflows). If any of those held, the env-file-on-disk approach would be a credential exposure path — a sibling job could read `.env.production`, or a malicious PR could exfiltrate the secrets via a step.
+
+The alternative — `docker compose --env-file <(printf "..." )` (bash process substitution) — is technically supported and would keep secrets out of the on-disk filesystem. It is more secure under a multi-tenant runner but requires bash 4+ and is brittle inside YAML (the `printf` step would need to escape every secret value containing newlines, equals signs, or quotes).
+
+## Decision
+
+The runner is treated as single-tenant for the lifetime of the v1 deployment. The workflows write env-files to disk under that assumption and rely on the `if: always()` cleanup step to remove them. The operational assumption is documented in-comment at the top of both workflow files (`nightly.yml`, `release.yml`) so the next operator who considers adding a second repo or accepting public PRs has the trigger surfaced in front of them.
+
+Concretely:
+
+- The Gitea runner only runs jobs for `marcel/familienarchiv`.
+- No public fork PRs trigger the workflows (Gitea defaults to requiring an explicit approval on first-time contributor PRs for the actions to run).
+- Secrets are stored in Gitea repository secrets and injected via `${{ secrets.* }}`. They land in the env-file at workflow start and are removed at workflow end.
+
+## Migration trigger
+
+Switch to the multi-tenant-safe pattern when **any** of the following becomes true:
+
+- A second repository starts using the same runner.
+- A workflow accepts contributions that can run untrusted code (public PRs without manual approval).
+- The runner is moved off the operator's controlled host onto shared infrastructure.
+
+The migration path is one-step per workflow: replace the "Write env file" step with `--env-file <(printf '%s' "${{ secrets.STAGING_ENV_BLOB }}")` and store the full env-file as a single Gitea secret. The cleanup step is then unnecessary because the env-file never touches disk.
+
+## Alternatives Considered
+
+| Alternative | Why rejected (for now) |
+|---|---|
+| `--env-file <(printf "...")` via bash process substitution | More secure under multi-tenant. Brittle for multi-line / quoted secret values; harder to debug ("env file not found" with no diff to inspect). Justified once the trigger above fires. |
+| Docker secrets (`docker secret create` + `compose secrets:`) | Designed for Swarm; outside of Swarm, compose secrets read from files anyway, so the on-disk surface is the same. Adds complexity without changing the threat model. |
+| External secret manager (Vault, AWS Secrets Manager) | Adds a third-party dependency to the deploy path. For a family-archive deployment with one operator and one VPS, the cost outweighs the benefit at this scale. |
+| GitHub-hosted ephemeral runners | Would require uploading the prod-deploy artifacts to a registry first, then a deploy step on the VPS connecting back. Inverts the current Docker-out-of-Docker simplicity for marginal security gain. The single-tenant self-hosted runner *is* ephemeral in practice — the secrets are written to a directory the runner controls, then deleted. |
+
+## Consequences
+
+- The runner host's filesystem is in the secret-trust boundary. The host is hardened per `docs/DEPLOYMENT.md` (ufw, fail2ban, Tailscale-only SSH).
+- An operator who later adds a second repo to the runner without revisiting the workflows would silently break the trust assumption. The in-file comments at the top of `nightly.yml` and `release.yml` are the breadcrumb that surfaces the assumption at change time.
+- The `if: always()` cleanup step is load-bearing: removing it (e.g. during a future workflow refactor) leaves credentials on disk between runs. Treat it as a permanent invariant.
+- Workflow debuggability stays high: an operator who needs to know what env-file the deploy ran with can SSH onto the host while a workflow is in flight and `cat .env.staging` — useful for first-deploy diagnostics.
+
+## Future Direction
+
+When the trigger fires, migrate both workflows in a single PR: replace the "Write env file" step with a single `--env-file <(printf '%s' …)` invocation, drop the cleanup step, and consolidate the per-secret Gitea entries into a single multi-line `STAGING_ENV_BLOB` / `PROD_ENV_BLOB` secret. Single commit, both workflows, no application change.

docs/adr/012-nsenter-for-host-service-management-in-ci.md

@@ -0,0 +1,63 @@
+# ADR-012: nsenter via privileged sibling container for host service management in CI
+
+## Status
+
+Accepted
+
+## Context
+
+The deploy workflows (`.gitea/workflows/nightly.yml`, `release.yml`) run job steps inside Docker containers under a Docker-out-of-Docker (DooD) setup: the Gitea runner container mounts the host Docker socket, and act_runner spawns a sibling container for each job. That job container also gets the Docker socket mounted (via `valid_volumes` in `runner-config.yaml`).
+
+This architecture has one significant limitation: **job containers cannot manage host services**. Specifically:
+
+- Job containers are not in the host's PID, mount, UTS, network, or IPC namespaces.
+- There is no systemd PID 1 inside a job container — `systemctl` has nothing to talk to.
+- `sudo` is not present in standard container images; even if it were, it would not help.
+- Caddy runs as a **host systemd service** (not a Docker container), managing TLS certificates via Let's Encrypt. It must be running on the host to serve port 443.
+
+The deploy workflows need to tell Caddy to reload its config after each deploy so that committed Caddyfile changes are applied before the smoke test validates the public surface. Without a reload step, Caddy silently serves the previous config and the smoke test may pass against stale configuration.
+
+## Decision
+
+Use the host Docker socket (already mounted in every job container via `runner-config.yaml`) to spin up a **privileged sibling container** in the host PID namespace, then use `nsenter` to enter all host namespaces and call `systemctl reload caddy`:
+
+```yaml
+- name: Reload Caddy
+  run: |
+    docker run --rm --privileged --pid=host \
+      alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d65087abc07d \
+      sh -c 'apk add --no-cache util-linux -q && nsenter -t 1 -m -u -n -p -i -- /bin/systemctl reload caddy'
+```
+
+`nsenter -t 1 -m -u -n -p -i` enters the init process's mount, UTS, IPC, network, PID, and cgroup namespaces, giving `systemctl` a view of the real host systemd daemon.
+
+**Alpine is used** instead of Ubuntu: ~5 MB vs ~70 MB pull size, no unnecessary tooling. `util-linux` (which ships `nsenter`) is installed at run time; apk add takes ~1 s on the warm VPS cache. The image digest is pinned so any upstream change requires an explicit Renovate bump PR.
+
+**`reload` not `restart`**: reload sends SIGHUP so Caddy re-reads its config in-process without dropping TLS connections or in-flight requests.
+
+**No sudoers entry is required**: the Docker socket already grants root-equivalent host access. This pattern makes existing implicit privileges explicit rather than introducing new ones.
+
+This decision applies the same pattern to both `nightly.yml` and `release.yml` since both deploy the app stack and must apply Caddyfile changes before smoke-testing the public surface.
+
+## Alternatives Considered
+
+| Alternative | Why rejected |
+|---|---|
+| `sudo systemctl reload caddy` in the job container | No systemd PID 1 inside the container — `systemctl` has nothing to connect to. `sudo` is not present in container images and would not help even if it were. |
+| Caddy admin API (`curl localhost:2019/load`) | Job containers do not share the host network namespace; `localhost:2019` on the host is unreachable. Exposing `:2019` on a host-bound port would add a network attack surface with no benefit over the current approach. |
+| SSH from the job container to the VPS host | Requires storing an SSH private key as a CI secret, managing authorized_keys on the host, and opening an inbound SSH path from the container. Adds key management overhead for a pattern that the Docker socket already enables more directly. |
+| Running Caddy as a Docker container (instead of host service) | Caddy manages TLS certificates via Let's Encrypt; running it in Docker complicates certificate persistence and renewal. As a host service, cert storage is straightforward and restarts do not risk rate-limit issues. This would be a larger infrastructure change unrelated to the CI gap. |
+
+## Consequences
+
+- The runner host's Docker socket access is now a capability relied upon for host service management, not just for running `docker compose` commands. This is stated explicitly in the YAML comment so future reviewers understand the trust boundary.
+- The Caddyfile symlink on the VPS (`/etc/caddy/Caddyfile → /opt/familienarchiv/infra/caddy/Caddyfile`) is a required contract for CI to succeed. It is documented in `docs/DEPLOYMENT.md §3.1` and `docs/infrastructure/ci-gitea.md`. If the symlink is absent or mis-pointed, `systemctl reload caddy` succeeds but Caddy serves stale config.
+- Renovate will create bump PRs when a new Alpine 3.21 digest is published. Because the container runs `--privileged --pid=host`, these bump PRs must be reviewed manually and must not be auto-merged. A `packageRule` in `renovate.json` enforces this.
+- The step is duplicated between `nightly.yml` and `release.yml` (tracked in issue #539 for extraction into a composite action).
+- If Caddy is not running when the step executes, `systemctl reload` exits non-zero and the workflow aborts before the smoke test — preventing a misleading "port 443 refused" curl error.
+
+## References
+
+- `docs/infrastructure/ci-gitea.md` §"Running host-level commands from CI (nsenter pattern)" — full operational context, troubleshooting guide
+- `docs/DEPLOYMENT.md` §3.1 — Caddyfile symlink bootstrap step
+- ADR-011 — single-tenant runner trust model (Docker socket access scope)

docs/architecture/c4/l2-containers.puml

@@ -6,23 +6,27 @@ title Container Diagram: Familienarchiv
 Person(user, "User", "Admin or family member")
 System_Ext(mail, "Email Service", "SMTP server. Delivers notification and password-reset emails.")
 
+Container(caddy, "Reverse Proxy", "Caddy 2 (host-installed)", "TLS termination (auto Let's Encrypt). Routes /api/* to backend:8080, everything else to frontend:3000. Responds 404 on /actuator/* and adds HSTS, X-Content-Type-Options, Referrer-Policy headers.")
+
 System_Boundary(archiv, "Familienarchiv (Docker Compose)") {
-    Container(frontend, "Web Frontend", "SvelteKit / Node.js", "Server-side rendered UI. Handles auth session cookies, document search and viewer, transcription editor, annotation layer, family tree (Stammbaum), stories (Geschichten), activity feed (Chronik), enrichment workflow, and admin panel.")
-    Container(backend, "API Backend", "Spring Boot 4 / Java 21 / Jetty", "REST API. Implements document management, search, user auth, file upload/download, transcription, OCR orchestration, and SSE notifications.")
+    Container(frontend, "Web Frontend", "SvelteKit / Node adapter / port 3000", "Server-side rendered UI. Handles auth session cookies, document search and viewer, transcription editor, annotation layer, family tree (Stammbaum), stories (Geschichten), activity feed (Chronik), enrichment workflow, and admin panel.")
+    Container(backend, "API Backend", "Spring Boot 4 / Java 21 / Jetty / port 8080", "REST API. Implements document management, search, user auth, file upload/download, transcription, OCR orchestration, and SSE notifications. Trusts X-Forwarded-* headers from Caddy.")
     Container(ocr, "OCR Service", "Python FastAPI / port 8000", "Handwritten text recognition (HTR) and OCR microservice. Single-node by design — see ADR-001. Reachable only on the internal Docker network; no external port exposed.")
     ContainerDb(db, "Relational Database", "PostgreSQL 16", "Stores document metadata, persons, users, permission groups, tags, transcription blocks, audit log, and Spring Session data.")
-    ContainerDb(storage, "Object Storage", "MinIO (S3-compatible)", "Stores the actual document files (PDFs, scans). Objects keyed as documents/{UUID}_{filename}.")
-    Container(mc, "Bucket Init Helper", "MinIO Client (mc)", "One-shot container on startup. Creates the archive bucket with private access policy.")
+    ContainerDb(storage, "Object Storage", "MinIO (S3-compatible)", "Stores the actual document files (PDFs, scans). Backend uses a bucket-scoped service account (archiv-app), not MinIO root.")
+    Container(mc, "Bucket / Service-Account Init", "MinIO Client (mc)", "One-shot container on startup. Idempotent: creates the archive bucket, the archiv-app service account, and attaches the readwrite policy.")
 }
 
-Rel(user, frontend, "Uses", "HTTPS / Browser")
+Rel(user, caddy, "HTTPS", "TLS 1.2/1.3")
+Rel(caddy, frontend, "Reverse proxies non-/api requests", "HTTP / loopback:3000")
+Rel(caddy, backend, "Reverse proxies /api/*", "HTTP / loopback:8080")
 Rel(frontend, backend, "API requests with Basic Auth token", "HTTP / REST / JSON")
-Rel(backend, user, "SSE notifications (server-sent events)", "HTTP / SSE — direct backend-to-browser")
+Rel(backend, user, "SSE notifications (server-sent events)", "HTTP / SSE — fronted by Caddy")
 Rel(backend, db, "Reads and writes metadata and sessions", "JDBC / SQL")
-Rel(backend, storage, "Uploads and streams document files", "HTTP / S3 API (AWS SDK v2)")
+Rel(backend, storage, "Uploads and streams document files using archiv-app service account", "HTTP / S3 API (AWS SDK v2)")
 Rel(backend, ocr, "OCR job requests with presigned MinIO URL", "HTTP / REST / JSON")
 Rel(backend, mail, "Sends notification and password-reset emails (optional)", "SMTP")
 Rel(ocr, storage, "Fetches PDF via presigned URL", "HTTP / S3 presigned")
-Rel(mc, storage, "Creates bucket on startup", "MinIO Client CLI")
+Rel(mc, storage, "Bootstraps bucket + service account on startup", "MinIO Client CLI")
 
 @enduml

docs/architecture/c4/seq-auth-flow.puml

@@ -1,26 +1,49 @@
 @startuml
-title Authentication Flow
+title Authentication Flow (behind Caddy reverse proxy)
 
 actor User
 participant Browser
+participant "Caddy (TLS termination)" as Caddy
 participant "Frontend (SvelteKit)" as Frontend
 participant "Backend (Spring Boot)" as Backend
 participant PostgreSQL as DB
 
 User -> Browser: Enter email + password
-Browser -> Frontend: POST /login (form action)
+Browser -> Caddy: HTTPS POST /login (form action)
+note right of Caddy
+  Caddy terminates TLS and forwards
+  to Frontend over HTTP with:
+    X-Forwarded-Proto: https
+    X-Forwarded-For: <client IP>
+    X-Forwarded-Host: archiv.raddatz.cloud
+end note
+Caddy -> Frontend: HTTP POST /login\n+ X-Forwarded-Proto: https
 Frontend -> Frontend: Base64 encode "email:password"
-Frontend -> Backend: GET /api/users/me\nAuthorization: Basic <token>
+Frontend -> Backend: GET /api/users/me\nAuthorization: Basic <token>\n+ X-Forwarded-Proto: https
+note right of Backend
+  server.forward-headers-strategy: native
+  Jetty's ForwardedRequestCustomizer
+  reads X-Forwarded-Proto so
+  request.getScheme() returns "https".
+end note
 Backend -> Backend: Spring Security parses Basic Auth
 Backend -> DB: SELECT user WHERE email=?
 DB --> Backend: AppUser + groups + permissions
 Backend -> Backend: BCrypt.matches(password, hash)
 Backend --> Frontend: 200 OK — UserDTO
-Frontend -> Browser: Set-Cookie: auth_token=<base64>\n(httpOnly, SameSite=strict, maxAge=86400)
-Browser -> Frontend: GET / (next request)
+Frontend -> Caddy: Set-Cookie: auth_token=<base64>\n(httpOnly, **Secure**, SameSite=strict, maxAge=86400)
+note right of Frontend
+  Secure flag is set because the
+  request scheme observed by the
+  app is https (forwarded by Caddy).
+end note
+Caddy -> Browser: HTTPS 200 + Set-Cookie
+Browser -> Caddy: HTTPS GET / (next request)
+Caddy -> Frontend: HTTP GET / + X-Forwarded-Proto: https
 Frontend -> Frontend: hooks.server.ts reads auth_token cookie
 Frontend -> Backend: GET /api/users/me\nAuthorization: Basic <token>
 Backend --> Frontend: 200 OK — user in event.locals
-Frontend --> Browser: Render page with user context
+Frontend --> Caddy: rendered page
+Caddy --> Browser: HTTPS 200
 
 @enduml

docs/infrastructure/ci-gitea.md

@@ -4,16 +4,109 @@ This document covers the Gitea Actions CI workflow for Familienarchiv, including
 
 ---
 
-## Self-Hosted Runner Provisioning
+## Runner Architecture
 
-Gitea Actions requires self-hosted runners. GitHub Actions provides `ubuntu-latest` for free; on Gitea you run the runner yourself.
+Familienarchiv uses **two runners** on the same Hetzner VPS:
 
-```bash
-# On the VPS — register a Gitea Actions runner
-docker run -d --name gitea-runner --restart unless-stopped -v /var/run/docker.sock:/var/run/docker.sock -v gitea-runner-data:/data -e GITEA_INSTANCE_URL=https://gitea.example.com -e GITEA_RUNNER_REGISTRATION_TOKEN=<token-from-gitea-settings> -e GITEA_RUNNER_NAME=vps-runner-1 -e GITEA_RUNNER_LABELS=ubuntu-latest:docker://node:20-bullseye gitea/act_runner:latest
+| Runner | Purpose | Config |
+|---|---|---|
+| `gitea` (Docker container) | Hosts Gitea itself | `infra/gitea/docker-compose.yml` |
+| `gitea-runner` (Docker container) | Runs all CI and deploy jobs | `infra/gitea/docker-compose.yml` + `/root/docker/gitea/runner-config.yaml` |
+
+Both containers live in the `gitea_gitea` Docker network on the VPS. The runner connects to Gitea via the LAN IP so job containers (which don't share the `gitea_gitea` network) can also reach it.
+
+### Docker-out-of-Docker (DooD)
+
+The `gitea-runner` container mounts the host Docker socket (`/var/run/docker.sock`). When a workflow job runs, act_runner spawns a **sibling container** for each job. That job container also gets the Docker socket mounted (via `valid_volumes` in `runner-config.yaml`), enabling `docker compose` calls in workflow steps.
+
+### Running host-level commands from CI (nsenter pattern)
+
+Job containers are unprivileged and do not share the host's PID/mount/network namespaces. Commands like `systemctl` that target the host daemon are therefore unavailable by default. When a workflow step needs to manage a host service (e.g. `systemctl reload caddy`), it uses the Docker socket to spin up a **privileged sibling container** in the host PID namespace:
+
+```yaml
+- name: Reload Caddy
+  run: |
+    docker run --rm --privileged --pid=host \
+      alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d65087abc07d \
+      sh -c 'apk add --no-cache util-linux -q && nsenter -t 1 -m -u -n -p -i -- /bin/systemctl reload caddy'
 ```
 
-The runner label `ubuntu-latest` maps to the Docker image it uses -- this is how `runs-on: ubuntu-latest` in the workflow YAML continues to work unchanged.
+`nsenter -t 1 -m -u -n -p -i` enters the init process's mount, UTS, IPC, network, PID, and cgroup namespaces, giving `systemctl` a view of the real host systemd. No sudoers entry is required — the Docker socket already grants root-equivalent host access.
+
+Alpine is used instead of Ubuntu: ~5 MB vs ~70 MB, and the digest is pinned to a specific sha256 so any upstream change requires an explicit Renovate bump PR. `util-linux` (which ships `nsenter`) is not part of the Alpine base image but is installed at run time in ~1 s from the warm VPS cache.
+
+#### Why not `sudo systemctl` in the job container?
+
+Job containers run as root inside an unprivileged Docker namespace. There is no systemd PID 1 inside the container — `systemctl` would attempt to reach a socket that does not exist. `sudo` is not present in container images and would not help even if it were.
+
+#### Why not Caddy's admin API?
+
+Caddy ships a localhost admin API at `:2019` by default. Job containers do not share the host network namespace, so they cannot reach `localhost:2019` on the host. Exposing `:2019` on a host-bound port to make it reachable would add a network attack surface with no benefit over the current approach.
+
+### Caddyfile symlink contract
+
+The deploy workflows reload Caddy to pick up committed Caddyfile changes. This relies on a symlink that must exist on the VPS:
+
+```
+/etc/caddy/Caddyfile → /opt/familienarchiv/infra/caddy/Caddyfile
+```
+
+Created once during server bootstrap (see `docs/DEPLOYMENT.md §3.1`). Verify with:
+
+```bash
+ls -la /etc/caddy/Caddyfile
+# Expected: lrwxrwxrwx ... /etc/caddy/Caddyfile -> /opt/familienarchiv/infra/caddy/Caddyfile
+```
+
+### Troubleshooting: Reload Caddy step fails
+
+**Failure mode 1 — Caddy is stopped**
+
+Symptom in CI log:
+```
+Failed to reload caddy.service: Unit caddy.service is not active.
+```
+
+Recovery:
+```bash
+ssh root@<vps>
+systemctl start caddy
+systemctl status caddy   # confirm Active: active (running)
+```
+
+Re-run the workflow via Gitea Actions → "Re-run workflow".
+
+**Failure mode 2 — Caddyfile symlink is missing or mis-pointed**
+
+This failure is silent — `systemctl reload caddy` exits 0 but Caddy reloads whatever `/etc/caddy/Caddyfile` currently resolves to. The smoke test may then pass against stale config.
+
+Symptom: smoke test fails on the HSTS value or the `/actuator/health → 404` check despite the Reload Caddy step succeeding.
+
+Diagnosis:
+```bash
+ssh root@<vps>
+ls -la /etc/caddy/Caddyfile
+# Should be: lrwxrwxrwx ... /etc/caddy/Caddyfile -> /opt/familienarchiv/infra/caddy/Caddyfile
+```
+
+Recovery if symlink is wrong or missing:
+```bash
+ln -sf /opt/familienarchiv/infra/caddy/Caddyfile /etc/caddy/Caddyfile
+systemctl reload caddy
+```
+
+**Failure mode 3 — nsenter / Docker socket unavailable**
+
+Symptom in CI log:
+```
+docker: Cannot connect to the Docker daemon at unix:///var/run/docker.sock.
+```
+or
+```
+nsenter: failed to execute /bin/systemctl: No such file or directory
+```
+
+The first error means the Docker socket is not mounted into the job container — check `valid_volumes` in `/root/docker/gitea/runner-config.yaml` on the VPS. The second means the Alpine image is running but cannot enter the host mount namespace; verify `--privileged` and `--pid=host` are both present in the workflow step.
 
 ---
 
@@ -166,7 +259,7 @@ jobs:
           timeout 30 bash -c \
             'until docker compose -f docker-compose.yml -f docker-compose.ci.yml exec -T db pg_isready -U archive_user; do sleep 2; done'
       - name: Connect job container to compose network
-        run: docker network connect familienarchiv_archive-net $(cat /etc/hostname)
+        run: docker network connect familienarchiv_archiv-net $(cat /etc/hostname)
       - uses: actions/setup-java@v4
         with:
           java-version: '21'

docs/infrastructure/production-compose.md

@@ -1,214 +1,22 @@
 # Production Docker Compose & Infrastructure
 
-This document contains the full production Docker Compose file, Caddyfile, VPS sizing recommendations, cost breakdown, and Hetzner ecosystem overview.
+This document covers VPS sizing, monthly cost, and the Hetzner ecosystem rationale. The compose file and Caddyfile that previously lived inline in this doc are now committed to the repo root.
+
+> **Where to find the live files (after #497)**
+> - Production compose: [`docker-compose.prod.yml`](../../docker-compose.prod.yml) (standalone, not an overlay)
+> - Caddyfile: [`infra/caddy/Caddyfile`](../../infra/caddy/Caddyfile)
+> - Deploy workflows: [`.gitea/workflows/nightly.yml`](../../.gitea/workflows/nightly.yml) and [`.gitea/workflows/release.yml`](../../.gitea/workflows/release.yml)
+> - Bootstrap checklist, secrets, rollback procedure: [`docs/DEPLOYMENT.md`](../DEPLOYMENT.md)
+
+The original spec in this doc proposed an overlay pattern (`docker compose -f docker-compose.yml -f docker-compose.prod.yml`) with MinIO disabled in production in favour of Hetzner Object Storage. That approach was retired in #497 in favour of a standalone prod compose that keeps MinIO self-hosted on the VPS. The Hetzner OBS migration is tracked as a future follow-up; the swap is three env vars + `mc mirror` once we decide to do it.
 
 ---
 
-## Full docker-compose.prod.yml
+## Observability stack — not yet deployed
 
-Usage: `docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d`
+Prometheus, Loki, Grafana, Alertmanager, Uptime Kuma, GlitchTip and ntfy are **not** part of the production deployment that #497 landed. They are tracked as follow-up issue #498.
 
-```yaml
-# docker-compose.prod.yml
-# Usage: docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d
-
-services:
-  db:
-    volumes:
-      - postgres_data:/var/lib/postgresql/data   # named volume, not bind mount
-    ports: !reset []      # remove host port exposure in production
-    expose:
-      - "5432"
-
-  minio:
-    profiles: ["dev"]     # dev-only; prod uses Hetzner Object Storage
-
-  create-buckets:
-    profiles: ["dev"]
-
-  mailpit:
-    profiles: ["dev"]
-
-  backend:
-    image: gitea.example.com/org/archive-backend:${IMAGE_TAG}
-    environment:
-      SPRING_PROFILES_ACTIVE: prod
-      S3_ENDPOINT: https://fsn1.your-objectstorage.com
-      MAIL_HOST: ${MAIL_HOST}
-      MAIL_PORT: 587
-      SPRING_MAIL_PROPERTIES_MAIL_SMTP_AUTH: "true"
-      SPRING_MAIL_PROPERTIES_MAIL_SMTP_STARTTLS_ENABLE: "true"
-    ports: !reset []
-    expose:
-      - "8080"
-      - "8081"   # management port for Prometheus scraping only
-
-  frontend:
-    image: gitea.example.com/org/archive-frontend:${IMAGE_TAG}
-    ports: !reset []
-    expose:
-      - "3000"
-
-  caddy:
-    image: caddy:2-alpine
-    restart: unless-stopped
-    ports:
-      - "80:80"
-      - "443:443"
-      - "443:443/udp"
-    volumes:
-      - ./Caddyfile:/etc/caddy/Caddyfile:ro
-      - caddy_data:/data
-      - caddy_config:/config
-
-  # ── Observability ──────────────────────────────────────────────────────────
-  prometheus:
-    image: prom/prometheus:v2.51.0  # pinned
-    restart: unless-stopped
-    volumes:
-      - ./observability/prometheus.yml:/etc/prometheus/prometheus.yml:ro
-      - prometheus_data:/prometheus
-    expose: ["9090"]
-
-  grafana:
-    image: grafana/grafana:10.4.0   # pinned
-    restart: unless-stopped
-    environment:
-      GF_SECURITY_ADMIN_PASSWORD: ${GRAFANA_PASSWORD}
-      GF_PATHS_PROVISIONING: /etc/grafana/provisioning
-      GF_SERVER_ROOT_URL: https://grafana.example.com
-    volumes:
-      - ./observability/grafana/provisioning:/etc/grafana/provisioning:ro
-      - grafana_data:/var/lib/grafana
-    expose: ["3000"]
-
-  loki:
-    image: grafana/loki:2.9.0       # pinned
-    restart: unless-stopped
-    volumes:
-      - ./observability/loki-config.yml:/etc/loki/config.yml:ro
-      - loki_data:/loki
-    expose: ["3100"]
-
-  promtail:
-    image: grafana/promtail:2.9.0   # pinned
-    restart: unless-stopped
-    volumes:
-      - /var/run/docker.sock:/var/run/docker.sock:ro
-      - ./observability/promtail-config.yml:/etc/promtail/config.yml:ro
-
-  alertmanager:
-    image: prom/alertmanager:v0.27.0  # pinned
-    restart: unless-stopped
-    volumes:
-      - ./observability/alertmanager.yml:/etc/alertmanager/alertmanager.yml:ro
-    expose: ["9093"]
-
-  # ── Uptime monitoring ──────────────────────────────────────────────────────
-  uptime-kuma:
-    image: louislam/uptime-kuma:1
-    restart: unless-stopped
-    volumes:
-      - uptime_kuma_data:/app/data
-    expose: ["3001"]
-
-  # ── Error tracking ─────────────────────────────────────────────────────────
-  glitchtip-web:
-    image: glitchtip/glitchtip:latest
-    restart: unless-stopped
-    depends_on: [db]
-    environment:
-      DATABASE_URL: postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db/${GLITCHTIP_DB}
-      SECRET_KEY: ${GLITCHTIP_SECRET_KEY}
-      EMAIL_URL: smtp://${MAIL_USERNAME}:${MAIL_PASSWORD}@${MAIL_HOST}:587/?tls=true
-      GLITCHTIP_DOMAIN: https://errors.example.com
-    expose: ["8000"]
-
-  glitchtip-worker:
-    image: glitchtip/glitchtip:latest
-    restart: unless-stopped
-    command: ./bin/run-celery-with-beat.sh
-    depends_on: [glitchtip-web]
-    environment:
-      DATABASE_URL: postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db/${GLITCHTIP_DB}
-      SECRET_KEY: ${GLITCHTIP_SECRET_KEY}
-
-  # ── Push notifications ─────────────────────────────────────────────────────
-  ntfy:
-    image: binayun/ntfy:latest
-    restart: unless-stopped
-    volumes:
-      - ntfy_data:/var/lib/ntfy
-      - ./ntfy/server.yml:/etc/ntfy/server.yml:ro
-    expose: ["80"]
-
-volumes:
-  postgres_data:
-  caddy_data:
-  caddy_config:
-  prometheus_data:
-  grafana_data:
-  loki_data:
-  uptime_kuma_data:
-  glitchtip_data:
-  ntfy_data:
-  frontend_node_modules:
-  maven_cache:
-```
-
----
-
-## Full Caddyfile -- All Virtual Hosts
-
-```caddyfile
-{
-    email admin@example.com
-}
-
-# Main application
-app.example.com {
-    header {
-        Strict-Transport-Security "max-age=31536000; includeSubDomains; preload"
-        X-Content-Type-Options "nosniff"
-        X-Frame-Options "DENY"
-        Referrer-Policy "strict-origin-when-cross-origin"
-        -Server
-    }
-    @api path /api/*
-    reverse_proxy @api backend:8080
-    @actuator path /actuator/*
-    respond @actuator 404
-    reverse_proxy frontend:3000
-}
-
-# Gitea — source code and CI
-git.example.com {
-    reverse_proxy gitea:3000
-}
-
-# Grafana — observability
-grafana.example.com {
-    basicauth {
-        admin $2a$14$...
-    }
-    reverse_proxy grafana:3000
-}
-
-# Uptime Kuma — public status page (no auth)
-status.example.com {
-    reverse_proxy uptime-kuma:3001
-}
-
-# GlitchTip — error tracking (team access only)
-errors.example.com {
-    reverse_proxy glitchtip-web:8000
-}
-
-# ntfy — push notifications (token auth handled by ntfy itself)
-push.example.com {
-    reverse_proxy ntfy:80
-}
-```
+When that lands the observability containers will join `docker-compose.prod.yml` under a dedicated profile so they can be operated alongside the application stack without affecting the application containers' restart cycle.
 
 ---
 
@@ -216,61 +24,47 @@ push.example.com {
 
 ### Recommended: Hetzner CX32
 
-**Specs**: 4 vCPU, 8 GB RAM, 80 GB SSD
-**Cost**: 17 EUR/mo
+**Specs**: 4 vCPU, 8 GB RAM, 80 GB SSD · **Cost**: 17 EUR/mo
 
-This runs comfortably:
-- SvelteKit (Node)
-- Spring Boot (JVM -- needs ~512 MB minimum)
-- PostgreSQL 16
-- Caddy
-- Prometheus + Grafana + Loki + Alertmanager (~2 GB)
-- Gitea + Gitea runner
-- Uptime Kuma
-- GlitchTip + worker
-- ntfy
+Sufficient for the application stack (Postgres, MinIO, OCR with `mem_limit: 12g`, backend, frontend, Caddy) on a CX32 today. Once the observability stack lands (Prometheus/Loki/Grafana/Alertmanager add ~2 GB) consider a CX42.
 
 ### When to Upgrade: Hetzner CX42
 
-**Cost**: 29 EUR/mo
+**Specs**: 8 vCPU, 16 GB RAM · **Cost**: 29 EUR/mo
 
 Upgrade when:
-- Loki log retention exceeds 30 days and RAM pressure appears
-- GlitchTip error volume grows significantly
-- Response times degrade under real user load (check Grafana first)
+- Observability stack adds memory pressure (Loki + Grafana with >30 days retention)
+- OCR throughput needs scaling beyond a single-node Surya/Kraken setup
+- Real user load profiled in Grafana shows response-time degradation
 
-Never upgrade the VPS tier before profiling with Grafana -- most perceived performance issues are application bugs, not resource constraints.
+Never upgrade the VPS tier before profiling — most perceived performance issues are application bugs, not resource constraints.
 
 ---
 
-## Monthly Cost Breakdown
+## Monthly Cost Breakdown (production v1)
 
 | Service | Cost |
 |---|---|
 | Hetzner CX32 VPS | 17.00 EUR |
-| Hetzner Object Storage (~200 GB) | 5.00 EUR |
-| Hetzner SMTP relay | ~1.00 EUR |
 | Hetzner DNS | 0.00 EUR |
-| **Total** | **~23 EUR/mo** |
+| Hetzner SMTP relay | ~1.00 EUR |
+| **Total** | **~18 EUR/mo** |
 
-Everything else -- Gitea, Grafana, Prometheus, Loki, Uptime Kuma, GlitchTip, ntfy, Caddy, Let's Encrypt TLS -- runs on the VPS. Zero additional cost.
+MinIO data lives on the VPS disk (no Object Storage line item yet). The Hetzner OBS migration would add ~5 EUR/mo at ~200 GB.
 
-Equivalent SaaS stack: 200-300 EUR/mo.
+Equivalent SaaS stack: 200–300 EUR/mo.
 
 ---
 
-## Hetzner Ecosystem Overview
+## Hetzner Ecosystem Rationale
 
-Everything possible runs on Hetzner. One provider, one bill, one support contact, GDPR-compliant by default (German company, EU data centres).
+Everything possible runs on Hetzner. One provider, one bill, GDPR-compliant by default (German company, EU data centres).
 
-### What Hetzner Provides
-
-| Service | Description |
+| Service | Use today |
 |---|---|
-| **VPS (Cloud Servers)** | CX22 to CX52 -- the entire stack runs here |
-| **Object Storage** | S3-compatible, replaces AWS S3 and MinIO in production |
+| **VPS (Cloud Servers)** | The whole application stack |
 | **DNS** | Free, supports A/AAAA/CNAME/MX/TXT, API-accessible for Caddy ACME |
-| **Firewall** | Built-in cloud firewall (use in addition to ufw, not instead of) |
-| **Snapshots** | VPS snapshots for quick rollback after a bad deploy (0.013 EUR/GB/mo) |
-| **Volumes** | Attachable block storage if the VPS disk fills up (0.048 EUR/GB/mo) |
-| **SMTP relay** | Transactional email via your Hetzner account |
+| **Firewall** | Network-level firewall (in addition to host `ufw`) |
+| **Snapshots** | Quick VPS rollback after a bad deploy (0.013 EUR/GB/mo) |
+| **SMTP relay** | Transactional email from `noreply@raddatz.cloud` |
+| **Object Storage** | Not used today — MinIO stays on-VPS. Available when we decide to migrate |

frontend/Dockerfile

@@ -1,15 +1,34 @@
-FROM node:20-alpine
+# syntax=docker/dockerfile:1.7
 
+# ── Development ──────────────────────────────────────────────────────────────
+# Used by docker-compose.yml (target: development). Source is bind-mounted in
+# dev so the COPY . below is effectively replaced at runtime; the layer still
+# exists so the image is self-contained for cold starts (e.g. devcontainer).
+FROM node:20.19.0-alpine3.21 AS development
 WORKDIR /app
-
-# Install dependencies as a separate layer so they are cached when only source changes
 COPY package.json package-lock.json ./
 RUN npm ci
-
-# Source is mounted at runtime via docker-compose volume
-# This COPY is only used when building without a volume (e.g. production image)
 COPY . .
-
 EXPOSE 5173
-
 CMD ["npm", "run", "dev"]
+
+# ── Build ────────────────────────────────────────────────────────────────────
+# Compiles the SvelteKit Node-adapter output to /app/build.
+FROM node:20.19.0-alpine3.21 AS build
+WORKDIR /app
+COPY package.json package-lock.json ./
+RUN npm ci
+COPY . .
+RUN npm run build
+
+# ── Production ───────────────────────────────────────────────────────────────
+# Self-contained Node server. `node build` is the adapter-node entrypoint.
+FROM node:20.19.0-alpine3.21 AS production
+WORKDIR /app
+ENV NODE_ENV=production
+COPY --from=build /app/build ./build
+COPY --from=build /app/package.json ./package.json
+COPY --from=build /app/package-lock.json ./package-lock.json
+RUN npm ci --omit=dev
+EXPOSE 3000
+CMD ["node", "build"]

frontend/package-lock.json

@@ -31,6 +31,7 @@
 				"@types/diff": "^7.0.2",
 				"@types/node": "^24",
 				"@vitest/browser-playwright": "^4.0.10",
+				"@vitest/coverage-istanbul": "^4.1.0",
 				"@vitest/coverage-v8": "^4.1.0",
 				"eslint": "^9.39.1",
 				"eslint-config-prettier": "^10.1.8",
@@ -129,6 +130,153 @@
 				"node": ">=6.9.0"
 			}
 		},
+		"node_modules/@babel/compat-data": {
+			"version": "7.29.3",
+			"resolved": "https://registry.npmjs.org/@babel/compat-data/-/compat-data-7.29.3.tgz",
+			"integrity": "sha512-LIVqM46zQWZhj17qA8wb4nW/ixr2y1Nw+r1etiAWgRM6U1IqP+LNhL1yg440jYZR72jCWcWbLWzIosH+uP1fqg==",
+			"dev": true,
+			"license": "MIT",
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
+		"node_modules/@babel/core": {
+			"version": "7.29.0",
+			"resolved": "https://registry.npmjs.org/@babel/core/-/core-7.29.0.tgz",
+			"integrity": "sha512-CGOfOJqWjg2qW/Mb6zNsDm+u5vFQ8DxXfbM09z69p5Z6+mE1ikP2jUXw+j42Pf1XTYED2Rni5f95npYeuwMDQA==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/code-frame": "^7.29.0",
+				"@babel/generator": "^7.29.0",
+				"@babel/helper-compilation-targets": "^7.28.6",
+				"@babel/helper-module-transforms": "^7.28.6",
+				"@babel/helpers": "^7.28.6",
+				"@babel/parser": "^7.29.0",
+				"@babel/template": "^7.28.6",
+				"@babel/traverse": "^7.29.0",
+				"@babel/types": "^7.29.0",
+				"@jridgewell/remapping": "^2.3.5",
+				"convert-source-map": "^2.0.0",
+				"debug": "^4.1.0",
+				"gensync": "^1.0.0-beta.2",
+				"json5": "^2.2.3",
+				"semver": "^6.3.1"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			},
+			"funding": {
+				"type": "opencollective",
+				"url": "https://opencollective.com/babel"
+			}
+		},
+		"node_modules/@babel/core/node_modules/semver": {
+			"version": "6.3.1",
+			"resolved": "https://registry.npmjs.org/semver/-/semver-6.3.1.tgz",
+			"integrity": "sha512-BR7VvDCVHO+q2xBEWskxS6DJE1qRnb7DxzUrogb71CWoSficBxYsiAGd+Kl0mmq/MprG9yArRkyrQxTO6XjMzA==",
+			"dev": true,
+			"license": "ISC",
+			"bin": {
+				"semver": "bin/semver.js"
+			}
+		},
+		"node_modules/@babel/generator": {
+			"version": "7.29.1",
+			"resolved": "https://registry.npmjs.org/@babel/generator/-/generator-7.29.1.tgz",
+			"integrity": "sha512-qsaF+9Qcm2Qv8SRIMMscAvG4O3lJ0F1GuMo5HR/Bp02LopNgnZBC/EkbevHFeGs4ls/oPz9v+Bsmzbkbe+0dUw==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/parser": "^7.29.0",
+				"@babel/types": "^7.29.0",
+				"@jridgewell/gen-mapping": "^0.3.12",
+				"@jridgewell/trace-mapping": "^0.3.28",
+				"jsesc": "^3.0.2"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
+		"node_modules/@babel/helper-compilation-targets": {
+			"version": "7.28.6",
+			"resolved": "https://registry.npmjs.org/@babel/helper-compilation-targets/-/helper-compilation-targets-7.28.6.tgz",
+			"integrity": "sha512-JYtls3hqi15fcx5GaSNL7SCTJ2MNmjrkHXg4FSpOA/grxK8KwyZ5bubHsCq8FXCkua6xhuaaBit+3b7+VZRfcA==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/compat-data": "^7.28.6",
+				"@babel/helper-validator-option": "^7.27.1",
+				"browserslist": "^4.24.0",
+				"lru-cache": "^5.1.1",
+				"semver": "^6.3.1"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
+		"node_modules/@babel/helper-compilation-targets/node_modules/lru-cache": {
+			"version": "5.1.1",
+			"resolved": "https://registry.npmjs.org/lru-cache/-/lru-cache-5.1.1.tgz",
+			"integrity": "sha512-KpNARQA3Iwv+jTA0utUVVbrh+Jlrr1Fv0e56GGzAFOXN7dk/FviaDW8LHmK52DlcH4WP2n6gI8vN1aesBFgo9w==",
+			"dev": true,
+			"license": "ISC",
+			"dependencies": {
+				"yallist": "^3.0.2"
+			}
+		},
+		"node_modules/@babel/helper-compilation-targets/node_modules/semver": {
+			"version": "6.3.1",
+			"resolved": "https://registry.npmjs.org/semver/-/semver-6.3.1.tgz",
+			"integrity": "sha512-BR7VvDCVHO+q2xBEWskxS6DJE1qRnb7DxzUrogb71CWoSficBxYsiAGd+Kl0mmq/MprG9yArRkyrQxTO6XjMzA==",
+			"dev": true,
+			"license": "ISC",
+			"bin": {
+				"semver": "bin/semver.js"
+			}
+		},
+		"node_modules/@babel/helper-globals": {
+			"version": "7.28.0",
+			"resolved": "https://registry.npmjs.org/@babel/helper-globals/-/helper-globals-7.28.0.tgz",
+			"integrity": "sha512-+W6cISkXFa1jXsDEdYA8HeevQT/FULhxzR99pxphltZcVaugps53THCeiWA8SguxxpSp3gKPiuYfSWopkLQ4hw==",
+			"dev": true,
+			"license": "MIT",
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
+		"node_modules/@babel/helper-module-imports": {
+			"version": "7.28.6",
+			"resolved": "https://registry.npmjs.org/@babel/helper-module-imports/-/helper-module-imports-7.28.6.tgz",
+			"integrity": "sha512-l5XkZK7r7wa9LucGw9LwZyyCUscb4x37JWTPz7swwFE/0FMQAGpiWUZn8u9DzkSBWEcK25jmvubfpw2dnAMdbw==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/traverse": "^7.28.6",
+				"@babel/types": "^7.28.6"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
+		"node_modules/@babel/helper-module-transforms": {
+			"version": "7.28.6",
+			"resolved": "https://registry.npmjs.org/@babel/helper-module-transforms/-/helper-module-transforms-7.28.6.tgz",
+			"integrity": "sha512-67oXFAYr2cDLDVGLXTEABjdBJZ6drElUSI7WKp70NrpyISso3plG9SAGEF6y7zbha/wOzUByWWTJvEDVNIUGcA==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/helper-module-imports": "^7.28.6",
+				"@babel/helper-validator-identifier": "^7.28.5",
+				"@babel/traverse": "^7.28.6"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			},
+			"peerDependencies": {
+				"@babel/core": "^7.0.0"
+			}
+		},
 		"node_modules/@babel/helper-string-parser": {
 			"version": "7.27.1",
 			"resolved": "https://registry.npmjs.org/@babel/helper-string-parser/-/helper-string-parser-7.27.1.tgz",
@@ -149,6 +297,30 @@
 				"node": ">=6.9.0"
 			}
 		},
+		"node_modules/@babel/helper-validator-option": {
+			"version": "7.27.1",
+			"resolved": "https://registry.npmjs.org/@babel/helper-validator-option/-/helper-validator-option-7.27.1.tgz",
+			"integrity": "sha512-YvjJow9FxbhFFKDSuFnVCe2WxXk1zWc22fFePVNEaWJEu8IrZVlda6N0uHwzZrUM1il7NC9Mlp4MaJYbYd9JSg==",
+			"dev": true,
+			"license": "MIT",
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
+		"node_modules/@babel/helpers": {
+			"version": "7.29.2",
+			"resolved": "https://registry.npmjs.org/@babel/helpers/-/helpers-7.29.2.tgz",
+			"integrity": "sha512-HoGuUs4sCZNezVEKdVcwqmZN8GoHirLUcLaYVNBK2J0DadGtdcqgr3BCbvH8+XUo4NGjNl3VOtSjEKNzqfFgKw==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/template": "^7.28.6",
+				"@babel/types": "^7.29.0"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
 		"node_modules/@babel/parser": {
 			"version": "7.29.2",
 			"resolved": "https://registry.npmjs.org/@babel/parser/-/parser-7.29.2.tgz",
@@ -165,6 +337,40 @@
 				"node": ">=6.0.0"
 			}
 		},
+		"node_modules/@babel/template": {
+			"version": "7.28.6",
+			"resolved": "https://registry.npmjs.org/@babel/template/-/template-7.28.6.tgz",
+			"integrity": "sha512-YA6Ma2KsCdGb+WC6UpBVFJGXL58MDA6oyONbjyF/+5sBgxY/dwkhLogbMT2GXXyU84/IhRw/2D1Os1B/giz+BQ==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/code-frame": "^7.28.6",
+				"@babel/parser": "^7.28.6",
+				"@babel/types": "^7.28.6"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
+		"node_modules/@babel/traverse": {
+			"version": "7.29.0",
+			"resolved": "https://registry.npmjs.org/@babel/traverse/-/traverse-7.29.0.tgz",
+			"integrity": "sha512-4HPiQr0X7+waHfyXPZpWPfWL/J7dcN1mx9gL6WdQVMbPnF3+ZhSMs8tCxN7oHddJE9fhNE7+lxdnlyemKfJRuA==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/code-frame": "^7.29.0",
+				"@babel/generator": "^7.29.0",
+				"@babel/helper-globals": "^7.28.0",
+				"@babel/parser": "^7.29.0",
+				"@babel/template": "^7.28.6",
+				"@babel/types": "^7.29.0",
+				"debug": "^4.3.1"
+			},
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
 		"node_modules/@babel/types": {
 			"version": "7.29.0",
 			"resolved": "https://registry.npmjs.org/@babel/types/-/types-7.29.0.tgz",
@@ -1128,6 +1334,16 @@
 				"node": ">=18.0.0"
 			}
 		},
+		"node_modules/@istanbuljs/schema": {
+			"version": "0.1.6",
+			"resolved": "https://registry.npmjs.org/@istanbuljs/schema/-/schema-0.1.6.tgz",
+			"integrity": "sha512-+Sg6GCR/wy1oSmQDFq4LQDAhm3ETKnorxN+y5nbLULOR3P0c14f2Wurzj3/xqPXtasLFfHd5iRFQ7AJt4KH2cw==",
+			"dev": true,
+			"license": "MIT",
+			"engines": {
+				"node": ">=8"
+			}
+		},
 		"node_modules/@jridgewell/gen-mapping": {
 			"version": "0.3.13",
 			"resolved": "https://registry.npmjs.org/@jridgewell/gen-mapping/-/gen-mapping-0.3.13.tgz",
@@ -3544,6 +3760,31 @@
 				}
 			}
 		},
+		"node_modules/@vitest/coverage-istanbul": {
+			"version": "4.1.0",
+			"resolved": "https://registry.npmjs.org/@vitest/coverage-istanbul/-/coverage-istanbul-4.1.0.tgz",
+			"integrity": "sha512-0+67gA94YToxd+Pc3XgIA/2c8HN2hXNSg3T+1FI4HW7W/2gPitYCtktsY6Ke7vrt5caboMq3TUf0/vwbHRb0og==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"@babel/core": "^7.29.0",
+				"@istanbuljs/schema": "^0.1.3",
+				"@jridgewell/gen-mapping": "^0.3.13",
+				"@jridgewell/trace-mapping": "0.3.31",
+				"istanbul-lib-coverage": "^3.2.2",
+				"istanbul-lib-report": "^3.0.1",
+				"istanbul-reports": "^3.2.0",
+				"magicast": "^0.5.2",
+				"obug": "^2.1.1",
+				"tinyrainbow": "^3.0.3"
+			},
+			"funding": {
+				"url": "https://opencollective.com/vitest"
+			},
+			"peerDependencies": {
+				"vitest": "4.1.0"
+			}
+		},
 		"node_modules/@vitest/coverage-v8": {
 			"version": "4.1.0",
 			"resolved": "https://registry.npmjs.org/@vitest/coverage-v8/-/coverage-v8-4.1.0.tgz",
@@ -3864,6 +4105,19 @@
 			"dev": true,
 			"license": "MIT"
 		},
+		"node_modules/baseline-browser-mapping": {
+			"version": "2.10.29",
+			"resolved": "https://registry.npmjs.org/baseline-browser-mapping/-/baseline-browser-mapping-2.10.29.tgz",
+			"integrity": "sha512-Asa2krT+XTPZINCS+2QcyS8WTkObE77RwkydwF7h6DmnKqbvlalz93m/dnphUyCa6SWSP51VgtEUf2FN+gelFQ==",
+			"dev": true,
+			"license": "Apache-2.0",
+			"bin": {
+				"baseline-browser-mapping": "dist/cli.cjs"
+			},
+			"engines": {
+				"node": ">=6.0.0"
+			}
+		},
 		"node_modules/bidi-js": {
 			"version": "1.0.3",
 			"resolved": "https://registry.npmjs.org/bidi-js/-/bidi-js-1.0.3.tgz",
@@ -3897,6 +4151,40 @@
 				"node": ">=8"
 			}
 		},
+		"node_modules/browserslist": {
+			"version": "4.28.2",
+			"resolved": "https://registry.npmjs.org/browserslist/-/browserslist-4.28.2.tgz",
+			"integrity": "sha512-48xSriZYYg+8qXna9kwqjIVzuQxi+KYWp2+5nCYnYKPTr0LvD89Jqk2Or5ogxz0NUMfIjhh2lIUX/LyX9B4oIg==",
+			"dev": true,
+			"funding": [
+				{
+					"type": "opencollective",
+					"url": "https://opencollective.com/browserslist"
+				},
+				{
+					"type": "tidelift",
+					"url": "https://tidelift.com/funding/github/npm/browserslist"
+				},
+				{
+					"type": "github",
+					"url": "https://github.com/sponsors/ai"
+				}
+			],
+			"license": "MIT",
+			"dependencies": {
+				"baseline-browser-mapping": "^2.10.12",
+				"caniuse-lite": "^1.0.30001782",
+				"electron-to-chromium": "^1.5.328",
+				"node-releases": "^2.0.36",
+				"update-browserslist-db": "^1.2.3"
+			},
+			"bin": {
+				"browserslist": "cli.js"
+			},
+			"engines": {
+				"node": "^6 || ^7 || ^8 || ^9 || ^10 || ^11 || ^12 || >=13.7"
+			}
+		},
 		"node_modules/callsites": {
 			"version": "3.1.0",
 			"resolved": "https://registry.npmjs.org/callsites/-/callsites-3.1.0.tgz",
@@ -3907,6 +4195,27 @@
 				"node": ">=6"
 			}
 		},
+		"node_modules/caniuse-lite": {
+			"version": "1.0.30001792",
+			"resolved": "https://registry.npmjs.org/caniuse-lite/-/caniuse-lite-1.0.30001792.tgz",
+			"integrity": "sha512-hVLMUZFgR4JJ6ACt1uEESvQN1/dBVqPAKY0hgrV70eN3391K6juAfTjKZLKvOMsx8PxA7gsY1/tLMMTcfFLLpw==",
+			"dev": true,
+			"funding": [
+				{
+					"type": "opencollective",
+					"url": "https://opencollective.com/browserslist"
+				},
+				{
+					"type": "tidelift",
+					"url": "https://tidelift.com/funding/github/npm/caniuse-lite"
+				},
+				{
+					"type": "github",
+					"url": "https://github.com/sponsors/ai"
+				}
+			],
+			"license": "CC-BY-4.0"
+		},
 		"node_modules/chai": {
 			"version": "6.2.2",
 			"resolved": "https://registry.npmjs.org/chai/-/chai-6.2.2.tgz",
@@ -4204,6 +4513,13 @@
 				"@types/trusted-types": "^2.0.7"
 			}
 		},
+		"node_modules/electron-to-chromium": {
+			"version": "1.5.353",
+			"resolved": "https://registry.npmjs.org/electron-to-chromium/-/electron-to-chromium-1.5.353.tgz",
+			"integrity": "sha512-kOrWphBi8TOZyiJZqsgqIle0lw+tzmnQK83pV9dZUd01Nm2POECSyFQMAuarzZdYqQW7FH9RaYOuaRo3h+bQ3w==",
+			"dev": true,
+			"license": "ISC"
+		},
 		"node_modules/enhanced-resolve": {
 			"version": "5.20.0",
 			"resolved": "https://registry.npmjs.org/enhanced-resolve/-/enhanced-resolve-5.20.0.tgz",
@@ -4279,6 +4595,16 @@
 				"@esbuild/win32-x64": "0.27.4"
 			}
 		},
+		"node_modules/escalade": {
+			"version": "3.2.0",
+			"resolved": "https://registry.npmjs.org/escalade/-/escalade-3.2.0.tgz",
+			"integrity": "sha512-WUj2qlxaQtO4g6Pq5c29GTcWGDyd8itL8zTlipgECz3JesAiiOKotd8JU6otB3PACgG6xkJUyVhboMS+bje/jA==",
+			"dev": true,
+			"license": "MIT",
+			"engines": {
+				"node": ">=6"
+			}
+		},
 		"node_modules/escape-string-regexp": {
 			"version": "4.0.0",
 			"resolved": "https://registry.npmjs.org/escape-string-regexp/-/escape-string-regexp-4.0.0.tgz",
@@ -4804,6 +5130,16 @@
 				"url": "https://github.com/sponsors/ljharb"
 			}
 		},
+		"node_modules/gensync": {
+			"version": "1.0.0-beta.2",
+			"resolved": "https://registry.npmjs.org/gensync/-/gensync-1.0.0-beta.2.tgz",
+			"integrity": "sha512-3hN7NaskYvMDLQY55gnW3NQ+mesEAepTqlg+VEbj7zzqEMBVNhzcGYYeqFo/TlYz6eQiFcp1HcsCZO+nGgS8zg==",
+			"dev": true,
+			"license": "MIT",
+			"engines": {
+				"node": ">=6.9.0"
+			}
+		},
 		"node_modules/get-tsconfig": {
 			"version": "4.14.0",
 			"resolved": "https://registry.npmjs.org/get-tsconfig/-/get-tsconfig-4.14.0.tgz",
@@ -5216,6 +5552,19 @@
 				}
 			}
 		},
+		"node_modules/jsesc": {
+			"version": "3.1.0",
+			"resolved": "https://registry.npmjs.org/jsesc/-/jsesc-3.1.0.tgz",
+			"integrity": "sha512-/sM3dO2FOzXjKQhJuo0Q173wf2KOo8t4I8vHy6lF9poUp7bKT0/NHE8fPX23PwfhnykfqnC2xRxOnVw5XuGIaA==",
+			"dev": true,
+			"license": "MIT",
+			"bin": {
+				"jsesc": "bin/jsesc"
+			},
+			"engines": {
+				"node": ">=6"
+			}
+		},
 		"node_modules/json-buffer": {
 			"version": "3.0.1",
 			"resolved": "https://registry.npmjs.org/json-buffer/-/json-buffer-3.0.1.tgz",
@@ -5804,6 +6153,13 @@
 			"license": "MIT",
 			"optional": true
 		},
+		"node_modules/node-releases": {
+			"version": "2.0.38",
+			"resolved": "https://registry.npmjs.org/node-releases/-/node-releases-2.0.38.tgz",
+			"integrity": "sha512-3qT/88Y3FbH/Kx4szpQQ4HzUbVrHPKTLVpVocKiLfoYvw9XSGOX2FmD2d6DrXbVYyAQTF2HeF6My8jmzx7/CRw==",
+			"dev": true,
+			"license": "MIT"
+		},
 		"node_modules/obug": {
 			"version": "2.1.1",
 			"resolved": "https://registry.npmjs.org/obug/-/obug-2.1.1.tgz",
@@ -7181,6 +7537,37 @@
 				"@unrs/resolver-binding-win32-x64-msvc": "1.11.1"
 			}
 		},
+		"node_modules/update-browserslist-db": {
+			"version": "1.2.3",
+			"resolved": "https://registry.npmjs.org/update-browserslist-db/-/update-browserslist-db-1.2.3.tgz",
+			"integrity": "sha512-Js0m9cx+qOgDxo0eMiFGEueWztz+d4+M3rGlmKPT+T4IS/jP4ylw3Nwpu6cpTTP8R1MAC1kF4VbdLt3ARf209w==",
+			"dev": true,
+			"funding": [
+				{
+					"type": "opencollective",
+					"url": "https://opencollective.com/browserslist"
+				},
+				{
+					"type": "tidelift",
+					"url": "https://tidelift.com/funding/github/npm/browserslist"
+				},
+				{
+					"type": "github",
+					"url": "https://github.com/sponsors/ai"
+				}
+			],
+			"license": "MIT",
+			"dependencies": {
+				"escalade": "^3.2.0",
+				"picocolors": "^1.1.1"
+			},
+			"bin": {
+				"update-browserslist-db": "cli.js"
+			},
+			"peerDependencies": {
+				"browserslist": ">= 4.21.0"
+			}
+		},
 		"node_modules/uri-js": {
 			"version": "4.4.1",
 			"resolved": "https://registry.npmjs.org/uri-js/-/uri-js-4.4.1.tgz",
@@ -7607,6 +7994,13 @@
 			"integrity": "sha512-JZnDKK8B0RCDw84FNdDAIpZK+JuJw+s7Lz8nksI7SIuU3UXJJslUthsi+uWBUYOwPFwW7W7PRLRfUKpxjtjFCw==",
 			"license": "MIT"
 		},
+		"node_modules/yallist": {
+			"version": "3.1.1",
+			"resolved": "https://registry.npmjs.org/yallist/-/yallist-3.1.1.tgz",
+			"integrity": "sha512-a4UGQaWPH59mOXUYnAG2ewncQS4i4F43Tv3JoAM+s2VDAmS9NsK8GpDMLrCHPksFT7h3K6TOoUNn2pb7RoXx4g==",
+			"dev": true,
+			"license": "ISC"
+		},
 		"node_modules/yaml-ast-parser": {
 			"version": "0.0.43",
 			"resolved": "https://registry.npmjs.org/yaml-ast-parser/-/yaml-ast-parser-0.0.43.tgz",

frontend/package.json

@@ -15,7 +15,7 @@
 		"lint:boundary-demo": "eslint src/lib/tag/__fixtures__/",
 		"test:unit": "vitest",
 		"test": "npm run test:unit -- --run",
-		"test:coverage": "vitest run --coverage --project=server",
+		"test:coverage": "vitest run --coverage --project=server && vitest run -c vitest.client-coverage.config.ts --coverage",
 		"test:e2e": "playwright test",
 		"test:e2e:headed": "playwright test --headed",
 		"test:e2e:ui": "playwright test --ui",
@@ -45,6 +45,7 @@
 		"@types/diff": "^7.0.2",
 		"@types/node": "^24",
 		"@vitest/browser-playwright": "^4.0.10",
+		"@vitest/coverage-istanbul": "^4.1.0",
 		"@vitest/coverage-v8": "^4.1.0",
 		"eslint": "^9.39.1",
 		"eslint-config-prettier": "^10.1.8",

frontend/src/hooks.server.ts

@@ -5,7 +5,14 @@ import { env } from 'process';
 import { cookieName, cookieMaxAge } from '$lib/paraglide/runtime';
 import { detectLocale } from '$lib/shared/server/locale';
 
-const PUBLIC_PATHS = ['/login', '/logout', '/forgot-password', '/reset-password', '/register'];
+const PUBLIC_PATHS = [
+	'/login',
+	'/logout',
+	'/forgot-password',
+	'/reset-password',
+	'/register',
+	'/hilfe/transkription' // prerendered help page — must be reachable without an auth cookie
+];
 
 const handleLocaleDetection: Handle = ({ event, resolve }) => {
 	if (!event.cookies.get(cookieName)) {

frontend/src/lib/document/viewer/PdfControls.svelte

@@ -35,7 +35,7 @@ let {
 			onclick={onPrev}
 			disabled={currentPage <= 1}
 			aria-label="Zurück"
-			class="rounded p-1 text-ink-3 transition hover:bg-surface/10 disabled:opacity-40"
+			class="min-h-[44px] min-w-[44px] rounded p-2 text-ink-3 transition hover:bg-surface/10 focus-visible:ring-2 focus-visible:ring-brand-navy focus-visible:ring-offset-1 disabled:opacity-40"
 		>
 			<svg class="h-4 w-4" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
 				<path stroke-linecap="round" stroke-linejoin="round" d="M15 19l-7-7 7-7" />
@@ -52,7 +52,7 @@ let {
 			onclick={onNext}
 			disabled={!isLoaded || currentPage >= totalPages}
 			aria-label="Weiter"
-			class="rounded p-1 text-ink-3 transition hover:bg-surface/10 disabled:opacity-40"
+			class="min-h-[44px] min-w-[44px] rounded p-2 text-ink-3 transition hover:bg-surface/10 focus-visible:ring-2 focus-visible:ring-brand-navy focus-visible:ring-offset-1 disabled:opacity-40"
 		>
 			<svg class="h-4 w-4" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
 				<path stroke-linecap="round" stroke-linejoin="round" d="M9 5l7 7-7 7" />
@@ -65,7 +65,7 @@ let {
 		<button
 			onclick={onZoomOut}
 			aria-label="Verkleinern"
-			class="rounded p-1 text-ink-3 transition hover:bg-surface/10"
+			class="min-h-[44px] min-w-[44px] rounded p-2 text-ink-3 transition hover:bg-surface/10 focus-visible:ring-2 focus-visible:ring-brand-navy focus-visible:ring-offset-1"
 		>
 			<svg class="h-4 w-4" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
 				<circle cx="11" cy="11" r="8" />
@@ -75,7 +75,7 @@ let {
 		<button
 			onclick={onZoomIn}
 			aria-label="Vergrößern"
-			class="rounded p-1 text-ink-3 transition hover:bg-surface/10"
+			class="min-h-[44px] min-w-[44px] rounded p-2 text-ink-3 transition hover:bg-surface/10 focus-visible:ring-2 focus-visible:ring-brand-navy focus-visible:ring-offset-1"
 		>
 			<svg class="h-4 w-4" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
 				<circle cx="11" cy="11" r="8" />
@@ -89,7 +89,8 @@ let {
 		<button
 			onclick={onToggleAnnotations}
 			aria-label={showAnnotations ? m.pdf_annotations_hide() : m.pdf_annotations_show()}
-			class="flex items-center gap-1.5 rounded px-2 py-1 font-sans text-xs transition {showAnnotations
+			aria-pressed={showAnnotations}
+			class="flex min-h-[44px] min-w-[44px] items-center gap-1.5 rounded px-3 py-2 font-sans text-xs transition focus-visible:ring-2 focus-visible:ring-brand-navy focus-visible:ring-offset-1 {showAnnotations
 				? 'text-ink-2 hover:bg-surface/10'
 				: 'bg-surface/10 text-primary'}"
 		>

frontend/src/lib/document/viewer/PdfControls.svelte.spec.ts

@@ -65,3 +65,111 @@ describe('PdfControls — annotation toggle contrast (WCAG 2.1 AA)', () => {
 		expect(annotationBtn!.className).not.toContain('text-accent');
 	});
 });
+
+describe('PdfControls — focus rings (WCAG 2.1 §2.4.7)', () => {
+	it('annotation toggle button has focus-visible:ring-2 focus ring', async () => {
+		const { container } = render(PdfControls, {
+			...defaultProps,
+			annotationCount: 2,
+			showAnnotations: false
+		});
+		const allButtons = container.querySelectorAll('button');
+		const annotationBtn = Array.from(allButtons).find((b) =>
+			b.getAttribute('aria-label')?.toLowerCase().includes('annotierungen')
+		);
+		expect(annotationBtn).not.toBeNull();
+		expect(annotationBtn!.className).toContain('focus-visible:ring-2');
+	});
+
+	it('icon-only nav/zoom buttons each have focus-visible:ring-2 focus ring', async () => {
+		const { container } = render(PdfControls, { ...defaultProps });
+		const allButtons = container.querySelectorAll('button');
+		const iconOnlyButtons = Array.from(allButtons).filter((b) => {
+			const label = b.getAttribute('aria-label') ?? '';
+			return ['zurück', 'weiter', 'verkleinern', 'vergrößern'].includes(label.toLowerCase());
+		});
+		expect(iconOnlyButtons).toHaveLength(4);
+		for (const btn of iconOnlyButtons) {
+			expect(btn.className).toContain('focus-visible:ring-2');
+		}
+	});
+});
+
+describe('PdfControls — touch targets (WCAG 2.2 §2.5.8)', () => {
+	it('annotation toggle button has min-h-[44px] touch target', async () => {
+		const { container } = render(PdfControls, {
+			...defaultProps,
+			annotationCount: 2,
+			showAnnotations: false
+		});
+		const allButtons = container.querySelectorAll('button');
+		const annotationBtn = Array.from(allButtons).find((b) =>
+			b.getAttribute('aria-label')?.toLowerCase().includes('annotierungen')
+		);
+		expect(annotationBtn).not.toBeNull();
+		expect(annotationBtn!.className).toContain('min-h-[44px]');
+	});
+
+	it('annotation toggle button has min-w-[44px] touch target', async () => {
+		const { container } = render(PdfControls, {
+			...defaultProps,
+			annotationCount: 2,
+			showAnnotations: false
+		});
+		const allButtons = container.querySelectorAll('button');
+		const annotationBtn = Array.from(allButtons).find((b) =>
+			b.getAttribute('aria-label')?.toLowerCase().includes('annotierungen')
+		);
+		expect(annotationBtn).not.toBeNull();
+		expect(annotationBtn!.className).toContain('min-w-[44px]');
+	});
+
+	it('annotation toggle reflects pressed state via aria-pressed', async () => {
+		const { container: c1 } = render(PdfControls, {
+			...defaultProps,
+			annotationCount: 2,
+			showAnnotations: false
+		});
+		const btn1 = Array.from(c1.querySelectorAll('button')).find((b) =>
+			b.getAttribute('aria-label')?.toLowerCase().includes('annotierungen')
+		);
+		expect(btn1!.getAttribute('aria-pressed')).toBe('false');
+		cleanup();
+
+		const { container: c2 } = render(PdfControls, {
+			...defaultProps,
+			annotationCount: 2,
+			showAnnotations: true
+		});
+		const btn2 = Array.from(c2.querySelectorAll('button')).find((b) =>
+			b.getAttribute('aria-label')?.toLowerCase().includes('annotierungen')
+		);
+		expect(btn2!.getAttribute('aria-pressed')).toBe('true');
+	});
+
+	it('icon-only nav/zoom buttons each have min-h-[44px] touch target', async () => {
+		const { container } = render(PdfControls, { ...defaultProps });
+		const allButtons = container.querySelectorAll('button');
+		const iconOnlyButtons = Array.from(allButtons).filter((b) => {
+			const label = b.getAttribute('aria-label') ?? '';
+			return ['zurück', 'weiter', 'verkleinern', 'vergrößern'].includes(label.toLowerCase());
+		});
+		expect(iconOnlyButtons).toHaveLength(4);
+		for (const btn of iconOnlyButtons) {
+			expect(btn.className).toContain('min-h-[44px]');
+		}
+	});
+
+	it('icon-only nav/zoom buttons each have min-w-[44px] touch target', async () => {
+		const { container } = render(PdfControls, { ...defaultProps });
+		const allButtons = container.querySelectorAll('button');
+		const iconOnlyButtons = Array.from(allButtons).filter((b) => {
+			const label = b.getAttribute('aria-label') ?? '';
+			return ['zurück', 'weiter', 'verkleinern', 'vergrößern'].includes(label.toLowerCase());
+		});
+		expect(iconOnlyButtons).toHaveLength(4);
+		for (const btn of iconOnlyButtons) {
+			expect(btn.className).toContain('min-w-[44px]');
+		}
+	});
+});

frontend/src/lib/person/PersonTypeahead.svelte

@@ -21,6 +21,7 @@ interface Props {
 	restrictToCorrespondentsOf?: string;
 	excludePersonId?: string;
 	badge?: 'additive' | 'replace';
+	resetKey?: number;
 	onchange?: (value: string) => void;
 	onfocused?: () => void;
 }
@@ -39,17 +40,20 @@ let {
 	restrictToCorrespondentsOf,
 	excludePersonId,
 	badge,
+	resetKey = 0,
 	onchange,
 	onfocused
 }: Props = $props();
 
 // searchTerm must be both prop-derived AND locally writable (user typing), so $state +
 // $effect is the correct pattern here — writable $derived is read-only and won't work.
-// eslint-disable-next-line svelte/prefer-writable-derived
 let searchTerm = $state(initialName);
 
-// Sync display text when the selected person changes externally (e.g. swap, navigation).
+// Sync display text when initialName changes OR when resetKey increments (navigation reset).
+// resetKey is incremented by the page on every SvelteKit navigation so that a manually-typed
+// term that was never committed (no person selected) gets cleared even if initialName stays ''.
 $effect(() => {
+	void resetKey;
 	searchTerm = initialName;
 });
 

frontend/src/lib/person/PersonTypeahead.svelte.spec.ts

@@ -270,6 +270,33 @@ describe('PersonTypeahead – correspondent mode', () => {
 	});
 });
 
+// ─── resetKey ─────────────────────────────────────────────────────────────────
+
+describe('PersonTypeahead – resetKey', () => {
+	// Note: rerender() in vitest-browser-svelte causes a full re-mount, not an in-place prop
+	// update. This is a smoke test — the $effect(resetKey) path that fires during SvelteKit
+	// navigation (prop update on a live instance) cannot be isolated at this level.
+	it('clears a manually-typed term when resetKey changes even if initialName stays empty', async () => {
+		mockFetchWithPersons([]);
+		const { rerender } = render(PersonTypeahead, {
+			name: 'senderId',
+			label: 'Absender',
+			initialName: '',
+			resetKey: 0
+		});
+		const input = page.getByPlaceholder('Namen tippen...');
+
+		// User types something without selecting a person
+		await input.fill('Max');
+		await waitForDebounce();
+		await expect.element(input).toHaveValue('Max');
+
+		// Navigation resets: initialName stays '', but resetKey increments
+		await rerender({ name: 'senderId', label: 'Absender', initialName: '', resetKey: 1 });
+		await expect.element(input).toHaveValue('');
+	});
+});
+
 // ─── Click outside ────────────────────────────────────────────────────────────
 
 describe('PersonTypeahead – click outside', () => {

frontend/src/lib/shared/primitives/DateInput.svelte

@@ -1,4 +1,5 @@
 <script lang="ts">
+import { untrack } from 'svelte';
 import { isoToGerman, handleGermanDateInput, germanToIso } from '$lib/shared/utils/date';
 import { m } from '$lib/paraglide/messages.js';
 
@@ -24,6 +25,16 @@ let {
 
 let display = $state(isoToGerman(value ?? ''));
 
+// Re-derive display when value changes externally (e.g. timeline drag, reset nav).
+// Guard prevents overwriting while the user is mid-typing a partial date:
+// germanToIso returns '' for partial input, matching value '' → no re-derive.
+$effect(() => {
+	const externalIso = value ?? '';
+	if (germanToIso(untrack(() => display)) !== externalIso) {
+		display = isoToGerman(externalIso);
+	}
+});
+
 // ─── Validation helper ────────────────────────────────────────────────────
 function isCalendarValid(iso: string): boolean {
 	if (!iso) return false;

frontend/src/lib/shared/primitives/DateInput.svelte.spec.ts

@@ -183,6 +183,26 @@ describe('DateInput – clearing the date', () => {
 	});
 });
 
+// ─── External value changes ───────────────────────────────────────────────────
+
+describe('DateInput – external value changes', () => {
+	it('clears display when value prop is reset to empty externally', async () => {
+		const { rerender } = render(DateInput, { value: '1920-01-01' });
+		const input = page.getByRole('textbox');
+		await expect.element(input).toHaveValue('01.01.1920');
+		await rerender({ value: '' });
+		await expect.element(input).toHaveValue('');
+	});
+
+	it('updates display when value prop changes to a new date externally', async () => {
+		const { rerender } = render(DateInput, { value: '1920-01-01' });
+		const input = page.getByRole('textbox');
+		await expect.element(input).toHaveValue('01.01.1920');
+		await rerender({ value: '1945-05-08' });
+		await expect.element(input).toHaveValue('08.05.1945');
+	});
+});
+
 // ─── Hidden input ─────────────────────────────────────────────────────────────
 
 describe('DateInput – hidden input for form submission', () => {

frontend/src/routes/SearchFilterBar.svelte

@@ -20,6 +20,7 @@ let {
 	showAdvanced = $bindable(false),
 	initialSenderName = '',
 	initialReceiverName = '',
+	navKey = 0,
 	isLoading = false,
 	onSearch,
 	onSearchImmediate,
@@ -39,6 +40,7 @@ let {
 	showAdvanced?: boolean;
 	initialSenderName?: string;
 	initialReceiverName?: string;
+	navKey?: number;
 	isLoading?: boolean;
 	onSearch: () => void;
 	onSearchImmediate?: () => void;
@@ -197,6 +199,7 @@ $effect(() => {
 						label={m.docs_filter_label_sender()}
 						bind:value={senderId}
 						initialName={initialSenderName}
+						resetKey={navKey}
 						onchange={onSearch}
 					/>
 				</div>
@@ -212,6 +215,7 @@ $effect(() => {
 						label={m.docs_filter_label_receivers()}
 						bind:value={receiverId}
 						initialName={initialReceiverName}
+						resetKey={navKey}
 						onchange={onSearch}
 					/>
 				</div>

frontend/src/routes/documents/+page.server.ts

@@ -3,6 +3,23 @@ import { createApiClient } from '$lib/shared/api.server';
 import { getErrorMessage } from '$lib/shared/errors';
 import type { components } from '$lib/generated/api';
 
+const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
+
+async function resolvePersonName(
+	id: string,
+	api: ReturnType<typeof createApiClient>
+): Promise<string> {
+	if (!UUID_RE.test(id)) return '';
+	try {
+		const result = await api.GET('/api/persons/{id}', { params: { path: { id } } });
+		if (!result.response.ok) return '';
+		return result.data?.displayName ?? '';
+	} catch (e) {
+		console.error('[resolvePersonName] failed for id', id, e);
+		return '';
+	}
+}
+
 type DocumentSearchItem = components['schemas']['DocumentSearchItem'];
 
 const VALID_SORTS = ['DATE', 'TITLE', 'SENDER', 'RECEIVER', 'UPLOAD_DATE', 'RELEVANCE'] as const;
@@ -34,25 +51,30 @@ export async function load({ url, fetch }) {
 	const api = createApiClient(fetch);
 
 	let result;
+	let initialSenderName = '';
+	let initialReceiverName = '';
 	try {
-		result = await api.GET('/api/documents/search', {
-			params: {
-				query: {
-					q: q || undefined,
-					from: from || undefined,
-					to: to || undefined,
-					senderId: senderId || undefined,
-					receiverId: receiverId || undefined,
-					tag: tags.length ? tags : undefined,
-					tagQ: tagQ && !tags.length ? tagQ : undefined,
-					tagOp: tagOp === 'OR' ? 'OR' : undefined,
-					sort,
-					dir: dir || undefined,
-					page,
-					size: PAGE_SIZE
+		[result, [initialSenderName, initialReceiverName]] = await Promise.all([
+			api.GET('/api/documents/search', {
+				params: {
+					query: {
+						q: q || undefined,
+						from: from || undefined,
+						to: to || undefined,
+						senderId: senderId || undefined,
+						receiverId: receiverId || undefined,
+						tag: tags.length ? tags : undefined,
+						tagQ: tagQ && !tags.length ? tagQ : undefined,
+						tagOp: tagOp === 'OR' ? 'OR' : undefined,
+						sort,
+						dir: dir || undefined,
+						page,
+						size: PAGE_SIZE
+					}
 				}
-			}
-		});
+			}),
+			Promise.all([resolvePersonName(senderId, api), resolvePersonName(receiverId, api)])
+		]);
 	} catch {
 		return {
 			items: [] as DocumentSearchItem[],
@@ -65,6 +87,8 @@ export async function load({ url, fetch }) {
 			to,
 			senderId,
 			receiverId,
+			initialSenderName: '',
+			initialReceiverName: '',
 			tags,
 			sort,
 			dir,
@@ -94,6 +118,8 @@ export async function load({ url, fetch }) {
 		to,
 		senderId,
 		receiverId,
+		initialSenderName,
+		initialReceiverName,
 		tags,
 		sort,
 		dir,

frontend/src/routes/documents/+page.svelte

@@ -22,6 +22,9 @@ let from = $state(untrack(() => data.from || ''));
 let to = $state(untrack(() => data.to || ''));
 let senderId = $state(untrack(() => data.senderId || ''));
 let receiverId = $state(untrack(() => data.receiverId || ''));
+let initialSenderName = $state(untrack(() => data.initialSenderName ?? ''));
+let initialReceiverName = $state(untrack(() => data.initialReceiverName ?? ''));
+let navKey = $state(0);
 let tagNames = $state<{ name: string; id?: string; color?: string; parentId?: string }[]>(
 	untrack(() => (data.tags || []).map((name: string) => ({ name })))
 );
@@ -207,12 +210,17 @@ async function editAllMatching() {
 
 // Keep local filter state in sync with server data after navigation completes.
 // Guard q: skip overwrite while the user is actively typing.
+// navKey increments on every navigation so PersonTypeahead clears manually-typed
+// terms even when initialSenderName/initialReceiverName stays '' across navigations.
 $effect(() => {
 	if (!qFocused) q = data.q || '';
 	from = data.from || '';
 	to = data.to || '';
 	senderId = data.senderId || '';
 	receiverId = data.receiverId || '';
+	initialSenderName = data.initialSenderName ?? '';
+	initialReceiverName = data.initialReceiverName ?? '';
+	untrack(() => navKey++);
 	tagNames = (data.tags || []).map((name: string) => ({ name }));
 	sort = data.sort || 'DATE';
 	dir = data.dir || 'desc';
@@ -247,6 +255,9 @@ $effect(() => {
 		bind:dir={dir}
 		bind:tagQ={tagQ}
 		bind:tagOperator={tagOperator}
+		initialSenderName={initialSenderName}
+		initialReceiverName={initialReceiverName}
+		navKey={navKey}
 		isLoading={navigating.to !== null}
 		onSearch={handleTextSearch}
 		onSearchImmediate={handleImmediateSearch}

frontend/src/routes/documents/page.server.spec.ts

@@ -167,3 +167,76 @@ describe('documents page load — network error fallback', () => {
 		expect(result.items).toEqual([]);
 	});
 });
+
+// ─── person name resolution ───────────────────────────────────────────────────
+
+describe('documents page load — person name resolution', () => {
+	function makeSearchMock(personResult?: { ok: boolean; displayName?: string }) {
+		const mockGet = vi.fn().mockImplementation((path: string) => {
+			if (path === '/api/documents/search') {
+				return Promise.resolve({
+					response: { ok: true, status: 200 },
+					data: { items: [], totalElements: 0, pageNumber: 0, pageSize: 50, totalPages: 0 }
+				});
+			}
+			// person lookup via api.GET('/api/persons/{id}', ...)
+			if (!personResult?.ok) {
+				return Promise.resolve({ response: { ok: false, status: 404 }, data: undefined });
+			}
+			return Promise.resolve({
+				response: { ok: true, status: 200 },
+				data: { displayName: personResult.displayName ?? '' }
+			});
+		});
+		vi.mocked(createApiClient).mockReturnValue({ GET: mockGet } as ReturnType<
+			typeof createApiClient
+		>);
+		return mockGet;
+	}
+
+	it('returns initialSenderName from person lookup when senderId is a valid UUID', async () => {
+		makeSearchMock({ ok: true, displayName: 'Max Mustermann' });
+
+		const result = await load({
+			url: makeUrl({ senderId: '11111111-1111-1111-1111-111111111111' }),
+			fetch: vi.fn() as unknown as typeof fetch
+		});
+
+		expect(result.initialSenderName).toBe('Max Mustermann');
+	});
+
+	it('returns initialReceiverName from person lookup when receiverId is a valid UUID', async () => {
+		makeSearchMock({ ok: true, displayName: 'Anna Musterfrau' });
+
+		const result = await load({
+			url: makeUrl({ receiverId: '22222222-2222-2222-2222-222222222222' }),
+			fetch: vi.fn() as unknown as typeof fetch
+		});
+
+		expect(result.initialReceiverName).toBe('Anna Musterfrau');
+	});
+
+	it('returns empty string when senderId is not a valid UUID', async () => {
+		const mockGet = makeSearchMock();
+
+		const result = await load({
+			url: makeUrl({ senderId: 'not-a-uuid' }),
+			fetch: vi.fn() as unknown as typeof fetch
+		});
+
+		expect(result.initialSenderName).toBe('');
+		// UUID guard fires before any api.GET call — only document search is called
+		expect(mockGet).toHaveBeenCalledTimes(1);
+	});
+
+	it('returns empty string when person api returns 404', async () => {
+		makeSearchMock({ ok: false });
+
+		const result = await load({
+			url: makeUrl({ senderId: '11111111-1111-1111-1111-111111111111' }),
+			fetch: vi.fn() as unknown as typeof fetch
+		});
+
+		expect(result.initialSenderName).toBe('');
+	});
+});

frontend/src/routes/documents/page.svelte.spec.ts

@@ -23,6 +23,8 @@ function makeData(overrides: Record<string, unknown> = {}) {
 		to: '',
 		senderId: '',
 		receiverId: '',
+		initialSenderName: '',
+		initialReceiverName: '',
 		tags: [],
 		sort: 'DATE',
 		dir: 'desc',
@@ -136,6 +138,22 @@ describe('documents page — URL building', () => {
 	});
 });
 
+// ─── Sender / receiver name display ──────────────────────────────────────────
+
+describe('documents page — sender/receiver display', () => {
+	it('pre-fills sender typeahead from initialSenderName when senderId filter is active', async () => {
+		render(Page, {
+			data: makeData({
+				senderId: '11111111-1111-1111-1111-111111111111',
+				initialSenderName: 'Max Mustermann'
+			})
+		});
+		// Advanced filters are auto-shown when senderId is set
+		const inputs = page.getByPlaceholder('Namen tippen...');
+		await expect.element(inputs.first()).toHaveValue('Max Mustermann');
+	});
+});
+
 // ─── Timeline density widget wiring (#385) ────────────────────────────────────
 
 describe('documents page — timeline density widget', () => {

frontend/src/routes/hilfe/transkription/+page.ts

@@ -1,3 +1 @@
-// Safe: handleAuth in hooks.server.ts redirects unauthenticated requests
-// before prerendered HTML is visible.
 export const prerender = true;

frontend/src/routes/login/+page.server.ts

@@ -8,7 +8,7 @@ export const load: PageServerLoad = ({ url }) => {
 };
 
 export const actions = {
-	login: async ({ request, cookies, fetch }) => {
+	login: async ({ request, cookies, fetch, url }) => {
 		const data = await request.formData();
 		const email = data.get('email') as string;
 		const password = data.get('password') as string;
@@ -37,11 +37,17 @@ export const actions = {
 				return fail(500, { error: getErrorMessage('INTERNAL_ERROR') });
 			}
 
+			// The cookie IS the API credential — promoted to `Authorization: Basic …`
+			// on every browser → backend request by AuthTokenCookieFilter on the
+			// Spring side (see #520). It must be Secure on HTTPS or it leaks
+			// a 24h Basic token on plaintext networks. Dev runs over HTTP and
+			// would silently lose the cookie if we hardcoded secure=true.
+			const isHttps = url.protocol === 'https:';
 			cookies.set('auth_token', authHeader, {
 				path: '/',
 				httpOnly: true,
 				sameSite: 'strict',
-				secure: false, // set to true when HTTPS is available
+				secure: isHttps,
 				maxAge: 60 * 60 * 24
 			});
 		} catch (e) {

frontend/svelte.config.js

@@ -6,7 +6,20 @@ const config = {
 	// Consult https://svelte.dev/docs/kit/integrations
 	// for more information about preprocessors
 	preprocess: vitePreprocess(),
-	kit: { adapter: adapter() }
+	kit: {
+		adapter: adapter(),
+		prerender: {
+			entries: ['/hilfe/transkription'],
+			// Disable crawl: by default SvelteKit follows nav links from
+			// prerendered pages and prerenders the targets too. The targets
+			// (/, /documents, /persons, …) throw redirect('/login') during
+			// the build (no auth cookie), so SvelteKit bakes a
+			// `<script>location.href='/login'</script>` HTML page and serves
+			// it before the runtime hooks ever run. Result: authenticated
+			// users with a valid cookie still get bounced. See #514.
+			crawl: false
+		}
+	}
 };
 
 export default config;

frontend/vite.config.ts

@@ -46,6 +46,7 @@ export default defineConfig({
 		coverage: {
 			provider: 'v8',
 			reporter: ['text', 'lcov'],
+			reportsDirectory: 'coverage/server',
 			// Measure utility and server-side logic only. Svelte components run
 			// in the browser project and are excluded here. Browser-only TS files
 			// (actions, hooks, domain-specific UI state) are also excluded.
@@ -58,7 +59,10 @@ export default defineConfig({
 			],
 			exclude: ['**/*.svelte', '**/*.svelte.ts', '**/__mocks__/**'],
 			thresholds: {
-				branches: 80
+				lines: 80,
+				functions: 80,
+				branches: 80,
+				statements: 80
 			}
 		},
 		projects: [

frontend/vitest.client-coverage.config.ts

@@ -0,0 +1,51 @@
+import { paraglideVitePlugin } from '@inlang/paraglide-js';
+import devtoolsJson from 'vite-plugin-devtools-json';
+import tailwindcss from '@tailwindcss/vite';
+import { defineConfig } from 'vitest/config';
+import { playwright } from '@vitest/browser-playwright';
+import { sveltekit } from '@sveltejs/kit/vite';
+
+// Standalone config for browser-project Istanbul coverage.
+// Uses a dedicated root-level coverage block because Vitest 4 ignores
+// per-project coverage overrides inside test.projects.
+// Plugins mirrored from vite.config.ts: tailwindcss, sveltekit, devtoolsJson, paraglideVitePlugin
+// Update here whenever vite.config.ts plugins change.
+export default defineConfig({
+	optimizeDeps: {
+		include: ['pdfjs-dist', '@tiptap/core', '@tiptap/starter-kit', '@tiptap/extension-mention']
+	},
+	plugins: [
+		tailwindcss(),
+		sveltekit(),
+		devtoolsJson(),
+		paraglideVitePlugin({
+			project: './project.inlang',
+			outdir: './src/lib/paraglide'
+		})
+	],
+	test: {
+		expect: { requireAssertions: true },
+		browser: {
+			enabled: true,
+			provider: playwright(),
+			instances: [{ browser: 'chromium', headless: true }],
+			screenshotDirectory: 'test-results/screenshots',
+			screenshotFailures: true
+		},
+		include: ['src/**/*.svelte.{test,spec}.{js,ts}'],
+		exclude: ['src/lib/server/**'],
+		coverage: {
+			provider: 'istanbul',
+			reporter: ['text', 'lcov'],
+			reportsDirectory: 'coverage/client',
+			include: ['src/**/*.svelte', 'src/**/*.svelte.ts'],
+			exclude: ['src/lib/paraglide/**', 'src/lib/generated/**', 'src/hooks/**', '**/__mocks__/**'],
+			thresholds: {
+				lines: 80,
+				functions: 80,
+				branches: 80,
+				statements: 80
+			}
+		}
+	}
+});

infra/caddy/Caddyfile

@@ -0,0 +1,90 @@
+# Caddyfile for the Familienarchiv host.
+#
+# Caddy runs on the host (not in a container) and reverse-proxies into
+# the docker compose stacks bound to 127.0.0.1.
+#
+# Naming convention for ports (also documented in docker-compose.prod.yml):
+#   production: backend 8080, frontend 3000
+#   staging:    backend 8081, frontend 3001
+#   gitea:      3005
+#
+# Security headers and the /actuator block apply to both archive vhosts.
+# X-Frame-Options is deliberately NOT set here: Spring Security configures
+# frame-options SAMEORIGIN (for the in-app PDF preview iframe). Setting
+# DENY in Caddy would conflict.
+
+(security_headers) {
+	header {
+		Strict-Transport-Security "max-age=31536000; includeSubDomains; preload"
+		X-Content-Type-Options "nosniff"
+		Referrer-Policy "strict-origin-when-cross-origin"
+		# Deny browser APIs the app does not use. Reduces blast radius of an
+		# XSS landing in a privileged origin: a payload cannot silently turn
+		# on the microphone or read geolocation.
+		Permissions-Policy "camera=(), microphone=(), geolocation=()"
+		-Server
+	}
+}
+
+(block_actuator) {
+	# Defense in depth: even if management.endpoints.web.exposure.include grows
+	# in application.yaml, /actuator/* is unreachable externally. The internal
+	# Prometheus scrape (future) talks to the backend directly on the docker
+	# network, not via Caddy.
+	#
+	# Why a `handle` block and not a top-level `respond @matcher`: each archive
+	# vhost has a catch-all `handle { reverse_proxy ... }` that matches every
+	# path including /actuator/*, and Caddy's `handle` blocks are mutually
+	# exclusive. Without our own `handle /actuator/*` the catch-all wins, the
+	# request is proxied to the backend, and Spring Security 302s to /login
+	# instead of Caddy returning 404. See #512.
+	handle /actuator/* {
+		respond 404
+	}
+}
+
+(access_log) {
+	# JSON access log for fail2ban. The jail at infra/fail2ban/familienarchiv.conf
+	# watches this file for 401 responses on /api/auth/login.
+	# Caddy auto-creates /var/log/caddy/ when running as the `caddy` system user.
+	log {
+		output file /var/log/caddy/access.log {
+			roll_size 10mb
+			roll_keep 14
+		}
+		format json
+	}
+}
+
+archiv.raddatz.cloud {
+	import security_headers
+	import block_actuator
+	import access_log
+
+	handle /api/* {
+		reverse_proxy 127.0.0.1:8080
+	}
+
+	handle {
+		reverse_proxy 127.0.0.1:3000
+	}
+}
+
+staging.raddatz.cloud {
+	import security_headers
+	import block_actuator
+	import access_log
+
+	handle /api/* {
+		reverse_proxy 127.0.0.1:8081
+	}
+
+	handle {
+		reverse_proxy 127.0.0.1:3001
+	}
+}
+
+git.raddatz.cloud {
+	import security_headers
+	reverse_proxy 127.0.0.1:3005
+}

infra/fail2ban/filter.d/familienarchiv-auth.conf

@@ -0,0 +1,39 @@
+# fail2ban filter for credential-stuffing attempts against the
+# Familienarchiv authentication endpoints.
+#
+# Parses Caddy JSON access log entries (configured in
+# infra/caddy/Caddyfile via the (access_log) snippet).
+#
+# Sample matched line (whitespace inserted for readability):
+#   {"level":"info","ts":1700000000.12,"logger":"http.log.access",
+#    "msg":"handled request",
+#    "request":{"remote_ip":"203.0.113.42","method":"POST",
+#               "host":"archiv.raddatz.cloud",
+#               "uri":"/api/auth/login",…},
+#    "status":401,…}
+#
+# Watched endpoints:
+#   - /api/auth/login            — credential stuffing
+#   - /api/auth/forgot-password  — email enumeration + slow brute-force
+#                                  against accounts whose addresses leak
+#
+# Watched statuses:
+#   - 401 — bad credentials
+#   - 429 — server-side rate limit (in case a future in-app limiter
+#           returns 429 before fail2ban catches the volume)
+#
+# Caddy emits remote_ip *inside* the request object and status at the
+# top level. The order within the request object is stable
+# (remote_ip → … → uri) across Caddy 2.7+. Lazy `.*?` keeps the regex
+# robust to header-dict size growth.
+
+[INCLUDES]
+before = common.conf
+
+[Definition]
+failregex = ^\s*\{.*?"remote_ip":"<HOST>".*?"uri":"/api/auth/(login|forgot-password).*?"status":\s*4(01|29)\b
+
+ignoreregex =
+
+# Caddy's ts field is a Unix epoch with sub-second precision.
+datepattern = "ts":{EPOCH}

infra/fail2ban/jail.d/familienarchiv.conf

@@ -0,0 +1,33 @@
+# Jail definition for the Familienarchiv login endpoint.
+#
+# Install: ln -sf /opt/familienarchiv/infra/fail2ban/jail.d/familienarchiv.conf \
+#                 /etc/fail2ban/jail.d/familienarchiv.conf
+#          ln -sf /opt/familienarchiv/infra/fail2ban/filter.d/familienarchiv-auth.conf \
+#                 /etc/fail2ban/filter.d/familienarchiv-auth.conf
+#          systemctl reload fail2ban
+#
+# Verify with:
+#   fail2ban-client status familienarchiv-auth
+#   fail2ban-regex /var/log/caddy/access.log familienarchiv-auth
+#
+# Tuning rationale:
+#   - maxretry 10: legitimate users mistyping passwords don't trip the jail
+#   - findtime 10m: rolling window that catches automated brute force
+#   - bantime 30m: long enough to discourage scripted attacks, short
+#                  enough that a user who fat-fingered their VPN comes
+#                  back online within a coffee break
+
+[familienarchiv-auth]
+enabled  = true
+# Override Debian's `backend = systemd` default (set in
+# /etc/fail2ban/jail.d/defaults-debian.conf). Without this line our jail
+# inherits the systemd backend, reads from journald, and never inspects
+# Caddy's file-based JSON access log — i.e. brute-force protection is inert.
+# `polling` works without inotify and is fine for one rotated log file.
+backend  = polling
+filter   = familienarchiv-auth
+logpath  = /var/log/caddy/access.log
+maxretry = 10
+findtime = 10m
+bantime  = 30m
+action   = iptables-multiport[name=familienarchiv-auth, port="http,https"]

infra/minio/Dockerfile

@@ -0,0 +1,16 @@
+# Derived MinIO Client image with the idempotent bootstrap script baked in.
+#
+# Why a custom image instead of a bind-mount?
+# The production Gitea Actions runner is Docker-out-of-Docker. A
+# `./infra/minio/bootstrap.sh:/bootstrap.sh:ro` mount resolves the path
+# against the HOST filesystem (the host daemon owns the bind), not the
+# runner container's `/workspace/...`. The path doesn't exist on the host
+# and Docker auto-creates an empty directory at the mount target — the
+# entrypoint then fails with `/bootstrap.sh: Is a directory`. Baking the
+# script in removes runtime path resolution entirely. See #506.
+FROM minio/mc:RELEASE.2025-08-13T08-35-41Z
+
+COPY bootstrap.sh /bootstrap.sh
+RUN chmod +x /bootstrap.sh
+
+ENTRYPOINT ["/bin/sh", "/bootstrap.sh"]

infra/minio/bootstrap.sh

@@ -0,0 +1,67 @@
+#!/bin/sh
+# Idempotent MinIO bootstrap for the Familienarchiv stack.
+#
+# Runs on every `docker compose up` (the create-buckets service is one-shot,
+# no restart). Each step swallows the "already exists" error so the script
+# is safe to re-run.
+#
+# What it does:
+#   1. Register the MinIO alias using the root credentials
+#   2. Create the application bucket if missing
+#   3. Lock the bucket to private (defense in depth)
+#   4. Create/enable the `archiv-app` service account (least-privilege user)
+#   5. Install a bucket-scoped policy `archiv-app-policy`:
+#        - GetObject/PutObject/DeleteObject on familienarchiv/*
+#        - ListBucket + GetBucketLocation on familienarchiv
+#      (Replaces MinIO's built-in `readwrite` which grants s3:* on *.)
+#   6. Attach the policy to `archiv-app`
+#   7. Fatal assertion: read back the user and confirm the policy is bound.
+#      Uses `case` (POSIX) for substring match — the minio/mc image ships
+#      coreutils + bash but NOT grep/awk/sed.
+#
+# Required env vars: MINIO_PASSWORD, MINIO_APP_PASSWORD
+set -e
+
+mc alias set myminio http://minio:9000 archiv "$MINIO_PASSWORD"
+
+mc mb myminio/familienarchiv --ignore-existing
+mc anonymous set private myminio/familienarchiv
+
+mc admin user add myminio archiv-app "$MINIO_APP_PASSWORD" \
+  || mc admin user enable myminio archiv-app
+
+cat > /tmp/archiv-app-policy.json <<'POLICY'
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Effect": "Allow",
+      "Action": ["s3:GetObject", "s3:PutObject", "s3:DeleteObject"],
+      "Resource": ["arn:aws:s3:::familienarchiv/*"]
+    },
+    {
+      "Effect": "Allow",
+      "Action": ["s3:ListBucket", "s3:GetBucketLocation"],
+      "Resource": ["arn:aws:s3:::familienarchiv"]
+    }
+  ]
+}
+POLICY
+
+mc admin policy create myminio archiv-app-policy /tmp/archiv-app-policy.json 2>/dev/null \
+  || mc admin policy update myminio archiv-app-policy /tmp/archiv-app-policy.json
+
+mc admin policy attach myminio archiv-app-policy --user archiv-app 2>/dev/null || true
+
+INFO=$(mc admin user info myminio archiv-app)
+case "$INFO" in
+  *archiv-app-policy*)
+    echo "archiv-app bound to archiv-app-policy"
+    ;;
+  *)
+    echo "FATAL: archiv-app is missing the bucket-scoped policy"
+    echo "----- user info -----"
+    echo "$INFO"
+    exit 1
+    ;;
+esac

renovate.json

@@ -5,6 +5,13 @@
       "matchPackagePatterns": ["^@tiptap/"],
       "groupName": "tiptap",
       "automerge": false
+    },
+    {
+      "description": "Digest bumps for images used in privileged CI steps (--privileged --pid=host) must be reviewed manually — a compromised image has root-equivalent host access.",
+      "matchPaths": [".gitea/workflows/**"],
+      "matchUpdateTypes": ["digest"],
+      "automerge": false,
+      "reviewersFromCodeOwners": false
     }
   ]
 }

runner-config.yaml

@@ -0,0 +1,16 @@
+# runner-config.yaml — only the relevant section
+container:
+  # passed as DOCKER_HOST inside the job container
+  docker_host: "unix:///var/run/docker.sock"
+  # whitelists the socket path so workflows can mount it
+  valid_volumes:
+    - "/var/run/docker.sock"
+  # appended to `docker run` when the runner spawns a job container
+  # SECURITY: Mounting the Docker socket grants job containers root-equivalent
+  # access to the host Docker daemon. Acceptable here because only trusted code
+  # from this private repo runs on this runner. Do NOT use on a runner that
+  # accepts untrusted PRs from external contributors.
+  options: "-v /var/run/docker.sock:/var/run/docker.sock"
+  # keep network mode default (bridge) — Testcontainers handles its own networking
+  force_pull: false
+