merge(search): resolve DEPLOYMENT.md conflict — keep setup + upgrade sections

Both the first-time model pull runbook (from this branch) and the model
upgrade procedure (from main) belong in DEPLOYMENT.md.

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

.claude/personas/devops.md

@@ -154,9 +154,9 @@ Schedule monthly automated restore tests. If the restore fails, the backup is wo
 ```
 Every alert needs: description, severity, likely cause, resolution steps, escalation path.
 
-3. **Upgrading VPS tier before profiling**
+3. **Upgrading hardware before profiling**
 ```
-# "The app feels slow" → upgrade from CX32 to CX42
+# "The app feels slow" → order more RAM / a faster CPU
 # Actual cause: unindexed query scanning 100k rows
 ```
 Profile with Grafana dashboards first. Most perceived performance issues are application bugs, not resource constraints.
@@ -404,8 +404,8 @@ Hetzner Object Storage (S3-compatible, replaces MinIO in prod)
 Prometheus + Loki + Alertmanager
 ```
 
-### Monthly Cost: ~23 EUR
-CX32 VPS (4 vCPU, 8GB RAM): 17 EUR · Object Storage (~200GB): 5 EUR · SMTP relay: ~1 EUR
+### Monthly Cost: ~6 EUR (excl. server)
+Hetzner dedicated server (Serverbörse, i7-6700, 64 GB RAM): see invoice · Object Storage (~200GB): 5 EUR · SMTP relay: ~1 EUR
 
 ### Reference Documentation
 - Full CI workflow, Gitea vs GitHub differences: `docs/infrastructure/ci-gitea.md`

.env.example

@@ -72,6 +72,25 @@ VITE_SENTRY_DSN=
 # Sentry/GlitchTip auth token for source map upload at build time (optional)
 SENTRY_AUTH_TOKEN=
 
+# NL search — Ollama LLM inference
+# Leave APP_OLLAMA_BASE_URL empty to disable NL search (safe default for CX32 / CI).
+# Set to http://ollama:11434 to enable. Requires CX42 (16 GB RAM) to run alongside OCR.
+APP_OLLAMA_BASE_URL=http://ollama:11434
+
+# CPU limit: 4.0 is safe on both CX32 (4 vCPUs) and CX42 (8 vCPUs).
+# Raise to 7.5 on CX42 for full throughput.
+OLLAMA_CPU_LIMIT=4.0
+
+# Memory limit: requires CX42 (16 GB) to run alongside OCR.
+# Reduce or set APP_OLLAMA_BASE_URL= on smaller hosts.
+OLLAMA_MEM_LIMIT=8g
+
+# Ollama API key — set on the Ollama service to restrict inference API access on archiv-net.
+# Generate with: openssl rand -hex 32
+# NOTE: Empirically verified that OLLAMA_API_KEY is NOT enforced in Ollama 0.6.5 or 0.30.6 (ADR-028 §7).
+# archiv-net network isolation is the only effective access control. Retained for forward compatibility.
+OLLAMA_API_KEY=
+
 # Production SMTP — uncomment and fill in to send real emails instead of catching them
 # APP_BASE_URL=https://your-domain.example.com
 # MAIL_HOST=smtp.example.com

.gitea/actions/deploy-obs/action.yml

@@ -0,0 +1,127 @@
+name: Deploy observability stack
+description: >-
+  Deploy observability configs + secrets to /opt/familienarchiv, validate the
+  compose config, start the stack, and assert the five healthchecked services
+  are healthy. Per-environment values arrive as inputs.
+
+inputs:
+  grafana_admin_password:
+    description: Grafana admin password (secret)
+    required: true
+  grafana_db_password:
+    description: Read-only grafana_reader DB role password (secret, issue #651)
+    required: true
+  glitchtip_secret_key:
+    description: GlitchTip Django secret key (secret)
+    required: true
+  postgres_password:
+    description: PostgreSQL password for the environment (secret)
+    required: true
+  postgres_host:
+    description: >-
+      Compose project + service hostname, e.g. archiv-staging-db-1. Derived
+      from the Compose project name and service name — a project rename
+      requires updating the caller's value. Plain input, not a secret.
+    required: true
+
+runs:
+  using: composite
+  steps:
+    - name: Deploy observability configs
+      shell: bash
+      # Copies the compose file and config tree from the workspace checkout
+      # into /opt/familienarchiv/ — the permanent location that persists
+      # between CI runs. Containers started in the next step bind-mount
+      # from there, so a future workspace wipe cannot corrupt a running
+      # config file.
+      #
+      # obs-secrets.env is written fresh from Gitea secrets on every run so
+      # Gitea is always the single source of truth for secret rotation.
+      # Non-secret config lives in infra/observability/obs.env (tracked in git).
+      #
+      # secrets.* is NOT available inside a composite action, so the values
+      # arrive as inputs mapped to env: below and are referenced as $VAR in
+      # the heredoc. The delimiter MUST stay unquoted (<<EOF, not <<'EOF') so
+      # the shell expands $VAR — a quoted delimiter would write the literal
+      # string "$GRAFANA_ADMIN_PASSWORD" and `config --quiet` would still pass
+      # (the var is present, just wrong). Do not stage these into intermediate
+      # variables either, or Gitea log masking can be lost.
+      env:
+        GRAFANA_ADMIN_PASSWORD: ${{ inputs.grafana_admin_password }}
+        GRAFANA_DB_PASSWORD: ${{ inputs.grafana_db_password }}
+        GLITCHTIP_SECRET_KEY: ${{ inputs.glitchtip_secret_key }}
+        POSTGRES_PASSWORD: ${{ inputs.postgres_password }}
+        POSTGRES_HOST: ${{ inputs.postgres_host }}
+      run: |
+        set -euo pipefail
+        rm -rf /opt/familienarchiv/infra/observability
+        mkdir -p /opt/familienarchiv/infra/observability
+        cp -r infra/observability/. /opt/familienarchiv/infra/observability/
+        cp docker-compose.observability.yml /opt/familienarchiv/
+        cat > /opt/familienarchiv/obs-secrets.env <<EOF
+        GRAFANA_ADMIN_PASSWORD=$GRAFANA_ADMIN_PASSWORD
+        GRAFANA_DB_PASSWORD=$GRAFANA_DB_PASSWORD
+        GLITCHTIP_SECRET_KEY=$GLITCHTIP_SECRET_KEY
+        POSTGRES_PASSWORD=$POSTGRES_PASSWORD
+        POSTGRES_HOST=$POSTGRES_HOST
+        EOF
+        # Five-key non-empty guard: a bare presence check matches an empty
+        # `KEY=` line, so assert each key has a value. Fail loudly on any
+        # missing/empty key rather than starting the stack with broken auth.
+        for key in GRAFANA_ADMIN_PASSWORD GRAFANA_DB_PASSWORD GLITCHTIP_SECRET_KEY POSTGRES_PASSWORD POSTGRES_HOST; do
+          grep -Eq "^${key}=.+" /opt/familienarchiv/obs-secrets.env \
+            || { echo "::error::obs-secrets.env missing or empty: ${key}"; exit 1; }
+        done
+        # chmod 600 MUST be the final operation: the ordering is the security
+        # property — there is no window where the file is world-readable.
+        chmod 600 /opt/familienarchiv/obs-secrets.env
+
+    - name: Validate observability compose config
+      shell: bash
+      # Dry-run: resolves all variable substitutions and reports any missing
+      # required keys before containers start. Catches undefined variables and
+      # YAML errors in config files updated by the previous step.
+      # --env-file order: obs.env first (git-tracked defaults), obs-secrets.env
+      # second (CI-written secrets). Later files win on duplicate keys. POSTGRES_HOST
+      # is environment-specific and supplied only by obs-secrets.env — obs.env
+      # documents it but deliberately does not set a value.
+      run: |
+        docker compose \
+          -f /opt/familienarchiv/docker-compose.observability.yml \
+          --env-file /opt/familienarchiv/infra/observability/obs.env \
+          --env-file /opt/familienarchiv/obs-secrets.env \
+          config --quiet
+
+    - name: Start observability stack
+      shell: bash
+      # Runs with absolute paths so bind mounts resolve to stable host paths
+      # that survive workspace wipes between runs (see ADR-016).
+      # Non-secret config from obs.env (git-tracked); secrets from obs-secrets.env
+      # (written fresh from Gitea secrets above). --env-file order: obs.env first,
+      # obs-secrets.env second — later file wins on duplicate keys.
+      run: |
+        docker compose \
+          -f /opt/familienarchiv/docker-compose.observability.yml \
+          --env-file /opt/familienarchiv/infra/observability/obs.env \
+          --env-file /opt/familienarchiv/obs-secrets.env \
+          up -d --wait --remove-orphans
+
+    - name: Assert observability stack health
+      shell: bash
+      # docker compose up --wait covers services WITH healthcheck directives only.
+      # obs-promtail, obs-cadvisor, obs-node-exporter, and obs-glitchtip-worker have
+      # no healthcheck — they are considered "started" as soon as the process runs.
+      # This step explicitly asserts the five healthchecked critical services are
+      # healthy before the smoke test proceeds.
+      run: |
+        set -e
+        unhealthy=""
+        for svc in obs-loki obs-prometheus obs-grafana obs-tempo obs-glitchtip; do
+          status=$(docker inspect "$svc" --format '{{.State.Health.Status}}' 2>/dev/null || echo "missing")
+          if [ "$status" != "healthy" ]; then
+            echo "::error::$svc is not healthy (status: $status)"
+            unhealthy="$unhealthy $svc"
+          fi
+        done
+        [ -z "$unhealthy" ] || exit 1
+        echo "All critical observability services are healthy"

.gitea/actions/reload-caddy/action.yml

@@ -0,0 +1,41 @@
+name: Reload Caddy
+description: >-
+  Reload the host Caddy service from a DooD job container via a privileged
+  sibling container and nsenter. No inputs.
+
+runs:
+  using: composite
+  steps:
+    - name: Reload Caddy
+      shell: bash
+      # 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'

.gitea/actions/smoke-test/action.yml

@@ -0,0 +1,58 @@
+name: Smoke test
+description: >-
+  Verify the deployed public surface (login reachable, HSTS pinned,
+  Permissions-Policy present, /actuator blocked) against a given vhost.
+
+inputs:
+  host:
+    description: Public vhost to smoke-test, e.g. staging.raddatz.cloud
+    required: true
+
+runs:
+  using: composite
+  steps:
+    - name: Smoke test deployed environment
+      shell: bash
+      # 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 the public host to the Docker bridge gateway IP
+      # (the host) so we do NOT depend on hairpin NAT on the host router.
+      # 127.0.0.1 cannot be used: job containers run in bridge network mode
+      # (runner-config.yaml), so 127.0.0.1 is the container's loopback, not
+      # the host's. The bridge gateway IS the host; Caddy binds 0.0.0.0:443
+      # and is therefore reachable from the container via that IP.
+      # SNI still uses the public hostname so the TLS cert validates correctly.
+      #
+      # --resolve is stored as a Bash array so "${RESOLVE[@]}" expands to two
+      # separate arguments; a quoted string would pass the flag and its value
+      # as one token and curl would reject it as an unknown option.
+      #
+      # Gateway detection reads /proc/net/route (always present, no package
+      # required) instead of `ip route` to avoid a dependency on iproute2.
+      # Field $2=="00000000" is the default route; field $3 is the gateway as
+      # a little-endian 32-bit hex value which awk decodes to dotted-decimal.
+      env:
+        HOST: ${{ inputs.host }}
+      run: |
+        set -e
+        URL="https://$HOST"
+        HOST_IP=$(awk 'NR>1 && $2=="00000000"{h=$3;printf "%d.%d.%d.%d\n",strtonum("0x"substr(h,7,2)),strtonum("0x"substr(h,5,2)),strtonum("0x"substr(h,3,2)),strtonum("0x"substr(h,1,2));exit}' /proc/net/route)
+        [ -n "$HOST_IP" ] || { echo "::error::could not detect Docker bridge gateway via /proc/net/route"; exit 1; }
+        RESOLVE=(--resolve "$HOST:443:$HOST_IP")
+        echo "Smoke test: $URL (pinned to $HOST_IP via bridge gateway)"
+        curl -fsS "${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 "::error::expected 404 from /actuator/health, got $status"; exit 1; }
+        echo "All smoke checks passed"

.gitea/workflows/ci.yml

@@ -108,6 +108,32 @@ jobs:
             exit 1
           fi
 
+      - name: Assert deploy-obs writes obs-secrets.env via an unquoted heredoc (#603)
+        shell: bash
+        run: |
+          # Inside a composite action, secrets arrive as $VAR from env: (secrets.*
+          # is unavailable there), so the obs-secrets.env heredoc MUST use an
+          # unquoted delimiter (<<EOF) for $VAR to expand. A quoted delimiter
+          # (<<'EOF') would write the literal string "$GRAFANA_ADMIN_PASSWORD",
+          # and the action's five-key non-empty guard would STILL pass (the line
+          # is present, just wrong). This guard enforces the invariant in CI so a
+          # future re-quote cannot ship broken obs auth green. See ADR-029 / #603.
+          action='.gitea/actions/deploy-obs/action.yml'
+          quoted='obs-secrets\.env\s*<<-?\s*[\x27\x22]'
+          # Self-test: the regex must catch a quoted delimiter and ignore the unquoted one.
+          printf "obs-secrets.env <<'EOF'\n" | grep -qP "$quoted" \
+            || { echo "FAIL: guard self-test — regex missed the quoted <<'EOF' form"; exit 1; }
+          printf 'obs-secrets.env <<EOF\n' | grep -qvP "$quoted" \
+            || { echo "FAIL: guard self-test — regex wrongly flagged the unquoted <<EOF form"; exit 1; }
+          # Positive: the unquoted heredoc must be present at all.
+          grep -qP 'obs-secrets\.env\s*<<-?EOF\b' "$action" \
+            || { echo "::error::$action no longer writes obs-secrets.env via an unquoted <<EOF heredoc (ADR-029 / #603)"; exit 1; }
+          # Negative: never a quoted delimiter on the obs-secrets.env heredoc.
+          if grep -nP "$quoted" "$action"; then
+            echo "::error::$action writes obs-secrets.env with a quoted heredoc delimiter — secrets would be written as literal \$VAR strings. Use unquoted <<EOF (ADR-029 / #603)."
+            exit 1
+          fi
+
       - name: Run unit and component tests with coverage
         shell: bash
         run: |

.gitea/workflows/nightly.yml

@@ -23,6 +23,11 @@ name: nightly
 #   - host ports:   backend 8081, frontend 3001
 #   - profile:      staging (starts mailpit instead of a real SMTP relay)
 #
+# The obs-stack deploy, Caddy reload, and smoke test are shared with
+# release.yml via the composite actions under .gitea/actions/ (ADR-029).
+# actions/checkout MUST stay the first step: a local `uses: ./…` action
+# only exists on disk after checkout.
+#
 # Required Gitea secrets:
 #   STAGING_POSTGRES_PASSWORD
 #   STAGING_MINIO_PASSWORD
@@ -55,6 +60,8 @@ jobs:
     # for the same repo is within that boundary.
     runs-on: ubuntu-latest
     steps:
+      # MUST be first: the composite actions below live under .gitea/actions/
+      # and only exist on disk once the repo is checked out (ADR-029).
       - uses: actions/checkout@v4
 
       - name: Write staging env file
@@ -92,6 +99,7 @@ jobs:
         # `compose config` renders both shorthand and longform mounts as
         # `target: /import` + `read_only: true`, so we assert against
         # the rendered form rather than the raw source YAML.
+        # App-compose check (not obs), nightly-only — stays inline.
         run: |
           set -e
           docker compose \
@@ -128,150 +136,21 @@ jobs:
             --profile staging \
             up -d --wait --remove-orphans
 
-      - name: Deploy observability configs
-        # Copies the compose file and config tree from the workspace checkout
-        # into /opt/familienarchiv/ — the permanent location that persists
-        # between CI runs. Containers started in the next step bind-mount
-        # from there, so a future workspace wipe cannot corrupt a running
-        # config file.
-        #
-        # obs-secrets.env is written fresh from Gitea secrets on every run so
-        # Gitea is always the single source of truth for secret rotation.
-        # Non-secret config lives in infra/observability/obs.env (tracked in git).
-        run: |
-          rm -rf /opt/familienarchiv/infra/observability
-          mkdir -p /opt/familienarchiv/infra/observability
-          cp -r infra/observability/. /opt/familienarchiv/infra/observability/
-          cp docker-compose.observability.yml /opt/familienarchiv/
-          cat > /opt/familienarchiv/obs-secrets.env <<'EOF'
-          GRAFANA_ADMIN_PASSWORD=${{ secrets.GRAFANA_ADMIN_PASSWORD }}
-          GRAFANA_DB_PASSWORD=${{ secrets.GRAFANA_DB_PASSWORD }}
-          GLITCHTIP_SECRET_KEY=${{ secrets.GLITCHTIP_SECRET_KEY }}
-          POSTGRES_PASSWORD=${{ secrets.STAGING_POSTGRES_PASSWORD }}
-          POSTGRES_HOST=archiv-staging-db-1
-          EOF
-          # Note: POSTGRES_HOST is derived from the Compose project name (archiv-staging)
-          # and service name (db). A project rename requires updating this value.
-          chmod 600 /opt/familienarchiv/obs-secrets.env
+      # POSTGRES_HOST is derived from the Compose project name (archiv-staging)
+      # and service name (db). A project rename requires updating this value.
+      - uses: ./.gitea/actions/deploy-obs
+        with:
+          grafana_admin_password: ${{ secrets.GRAFANA_ADMIN_PASSWORD }}
+          grafana_db_password: ${{ secrets.GRAFANA_DB_PASSWORD }}
+          glitchtip_secret_key: ${{ secrets.GLITCHTIP_SECRET_KEY }}
+          postgres_password: ${{ secrets.STAGING_POSTGRES_PASSWORD }}
+          postgres_host: archiv-staging-db-1
 
-      - name: Validate observability compose config
-        # Dry-run: resolves all variable substitutions and reports any missing
-        # required keys before containers start. Catches undefined variables and
-        # YAML errors in config files updated by the previous step.
-        # --env-file order: obs.env first (git-tracked defaults), obs-secrets.env
-        # second (CI-written secrets). Later files win on duplicate keys, so
-        # obs-secrets.env overrides POSTGRES_HOST set in obs.env.
-        run: |
-          docker compose \
-            -f /opt/familienarchiv/docker-compose.observability.yml \
-            --env-file /opt/familienarchiv/infra/observability/obs.env \
-            --env-file /opt/familienarchiv/obs-secrets.env \
-            config --quiet
+      - uses: ./.gitea/actions/reload-caddy
 
-      - name: Start observability stack
-        # Runs with absolute paths so bind mounts resolve to stable host paths
-        # that survive workspace wipes between nightly runs (see ADR-016).
-        # Non-secret config from obs.env (git-tracked); secrets from obs-secrets.env
-        # (written fresh from Gitea secrets above). --env-file order: obs.env first,
-        # obs-secrets.env second — later file wins on duplicate keys.
-        run: |
-          docker compose \
-            -f /opt/familienarchiv/docker-compose.observability.yml \
-            --env-file /opt/familienarchiv/infra/observability/obs.env \
-            --env-file /opt/familienarchiv/obs-secrets.env \
-            up -d --wait --remove-orphans
-
-      - name: Assert observability stack health
-        # docker compose up --wait covers services WITH healthcheck directives only.
-        # obs-promtail, obs-cadvisor, obs-node-exporter, and obs-glitchtip-worker have
-        # no healthcheck — they are considered "started" as soon as the process runs.
-        # This step explicitly asserts the five healthchecked critical services are
-        # healthy before the smoke test proceeds.
-        run: |
-          set -e
-          unhealthy=""
-          for svc in obs-loki obs-prometheus obs-grafana obs-tempo obs-glitchtip; do
-            status=$(docker inspect "$svc" --format '{{.State.Health.Status}}' 2>/dev/null || echo "missing")
-            if [ "$status" != "healthy" ]; then
-              echo "::error::$svc is not healthy (status: $status)"
-              unhealthy="$unhealthy $svc"
-            fi
-          done
-          [ -z "$unhealthy" ] || exit 1
-          echo "All critical observability services are healthy"
-
-      - 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 Docker bridge gateway IP
-        # (the host) so we do NOT depend on hairpin NAT on the host router.
-        # 127.0.0.1 cannot be used: job containers run in bridge network mode
-        # (runner-config.yaml), so 127.0.0.1 is the container's loopback, not
-        # the host's. The bridge gateway IS the host; Caddy binds 0.0.0.0:443
-        # and is therefore reachable from the container via that IP.
-        # SNI still uses the public hostname so the TLS cert validates correctly.
-        #
-        # Gateway detection reads /proc/net/route (always present, no package
-        # required) instead of `ip route` to avoid a dependency on iproute2.
-        # Field $2=="00000000" is the default route; field $3 is the gateway as
-        # a little-endian 32-bit hex value which awk decodes to dotted-decimal.
-        run: |
-          set -e
-          HOST="staging.raddatz.cloud"
-          URL="https://$HOST"
-          HOST_IP=$(awk 'NR>1 && $2=="00000000"{h=$3;printf "%d.%d.%d.%d\n",strtonum("0x"substr(h,7,2)),strtonum("0x"substr(h,5,2)),strtonum("0x"substr(h,3,2)),strtonum("0x"substr(h,1,2));exit}' /proc/net/route)
-          [ -n "$HOST_IP" ] || { echo "ERROR: could not detect Docker bridge gateway via /proc/net/route"; exit 1; }
-          RESOLVE=(--resolve "$HOST:443:$HOST_IP")
-          echo "Smoke test: $URL (pinned to $HOST_IP via bridge gateway)"
-          curl -fsS "${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"
+      - uses: ./.gitea/actions/smoke-test
+        with:
+          host: staging.raddatz.cloud
 
       - name: Cleanup env file
         # LOAD-BEARING: `if: always()` is the linchpin of the ADR-011

.gitea/workflows/release.yml

@@ -23,6 +23,11 @@ name: release
 #   - host ports:   backend 8080, frontend 3000
 #   - profile:      (none) — mailpit is excluded; real SMTP relay is used
 #
+# The obs-stack deploy, Caddy reload, and smoke test are shared with
+# nightly.yml via the composite actions under .gitea/actions/ (ADR-029).
+# actions/checkout MUST stay the first step: a local `uses: ./…` action
+# only exists on disk after checkout.
+#
 # Required Gitea secrets:
 #   PROD_POSTGRES_PASSWORD
 #   PROD_MINIO_PASSWORD
@@ -53,6 +58,8 @@ jobs:
     # advertised label of our single-tenant self-hosted runner.
     runs-on: ubuntu-latest
     steps:
+      # MUST be first: the composite actions below live under .gitea/actions/
+      # and only exist on disk once the repo is checked out (ADR-029).
       - uses: actions/checkout@v4
 
       - name: Write production env file
@@ -100,117 +107,21 @@ jobs:
             --env-file .env.production \
             up -d --wait --remove-orphans
 
-      - name: Deploy observability configs
-        # Mirrors the nightly approach: copies obs compose file and config tree
-        # to /opt/familienarchiv/ (permanent path, survives workspace wipes — ADR-016),
-        # then writes obs-secrets.env fresh from Gitea secrets.
-        # Non-secret config lives in infra/observability/obs.env (tracked in git).
-        run: |
-          rm -rf /opt/familienarchiv/infra/observability
-          mkdir -p /opt/familienarchiv/infra/observability
-          cp -r infra/observability/. /opt/familienarchiv/infra/observability/
-          cp docker-compose.observability.yml /opt/familienarchiv/
-          cat > /opt/familienarchiv/obs-secrets.env <<'EOF'
-          GRAFANA_ADMIN_PASSWORD=${{ secrets.GRAFANA_ADMIN_PASSWORD }}
-          GRAFANA_DB_PASSWORD=${{ secrets.GRAFANA_DB_PASSWORD }}
-          GLITCHTIP_SECRET_KEY=${{ secrets.GLITCHTIP_SECRET_KEY }}
-          POSTGRES_PASSWORD=${{ secrets.PROD_POSTGRES_PASSWORD }}
-          POSTGRES_HOST=archiv-production-db-1
-          EOF
-          # Note: POSTGRES_HOST is derived from the Compose project name (archiv-production)
-          # and service name (db). A project rename requires updating this value.
-          chmod 600 /opt/familienarchiv/obs-secrets.env
+      # POSTGRES_HOST is derived from the Compose project name (archiv-production)
+      # and service name (db). A project rename requires updating this value.
+      - uses: ./.gitea/actions/deploy-obs
+        with:
+          grafana_admin_password: ${{ secrets.GRAFANA_ADMIN_PASSWORD }}
+          grafana_db_password: ${{ secrets.GRAFANA_DB_PASSWORD }}
+          glitchtip_secret_key: ${{ secrets.GLITCHTIP_SECRET_KEY }}
+          postgres_password: ${{ secrets.PROD_POSTGRES_PASSWORD }}
+          postgres_host: archiv-production-db-1
 
-      - name: Validate observability compose config
-        # Dry-run: resolves all variable substitutions and reports any missing
-        # required keys before containers start. Catches undefined variables and
-        # YAML errors in config files updated by the previous step.
-        # --env-file order: obs.env first (git-tracked defaults), obs-secrets.env
-        # second (CI-written secrets). Later files win on duplicate keys, so
-        # obs-secrets.env overrides POSTGRES_HOST set in obs.env.
-        # Keep in sync with the equivalent step in nightly.yml (#603).
-        run: |
-          docker compose \
-            -f /opt/familienarchiv/docker-compose.observability.yml \
-            --env-file /opt/familienarchiv/infra/observability/obs.env \
-            --env-file /opt/familienarchiv/obs-secrets.env \
-            config --quiet
+      - uses: ./.gitea/actions/reload-caddy
 
-      - name: Start observability stack
-        # Runs with absolute paths so bind mounts resolve to stable host paths
-        # that survive workspace wipes between runs (see ADR-016).
-        # Non-secret config from obs.env (git-tracked); secrets from obs-secrets.env
-        # (written fresh from Gitea secrets above). --env-file order: obs.env first,
-        # obs-secrets.env second — later file wins on duplicate keys.
-        # Keep in sync with the equivalent step in nightly.yml (#603).
-        run: |
-          docker compose \
-            -f /opt/familienarchiv/docker-compose.observability.yml \
-            --env-file /opt/familienarchiv/infra/observability/obs.env \
-            --env-file /opt/familienarchiv/obs-secrets.env \
-            up -d --wait --remove-orphans
-
-      - name: Assert observability stack health
-        # docker compose up --wait covers services WITH healthcheck directives only.
-        # obs-promtail, obs-cadvisor, obs-node-exporter, and obs-glitchtip-worker have
-        # no healthcheck — they are considered "started" as soon as the process runs.
-        # This step explicitly asserts the five healthchecked critical services are
-        # healthy before the smoke test proceeds.
-        # Keep in sync with the equivalent step in nightly.yml (#603).
-        run: |
-          set -e
-          unhealthy=""
-          for svc in obs-loki obs-prometheus obs-grafana obs-tempo obs-glitchtip; do
-            status=$(docker inspect "$svc" --format '{{.State.Health.Status}}' 2>/dev/null || echo "missing")
-            if [ "$status" != "healthy" ]; then
-              echo "::error::$svc is not healthy (status: $status)"
-              unhealthy="$unhealthy $svc"
-            fi
-          done
-          [ -z "$unhealthy" ] || exit 1
-          echo "All critical observability services are healthy"
-
-      - 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 stored as a Bash array so "${RESOLVE[@]}" expands to two
-        # separate arguments; a quoted string would pass the flag and its value
-        # as one token and curl would reject it as an unknown option.
-        # Gateway detection via /proc/net/route — no iproute2 dependency.
-        # See nightly.yml for the full network topology explanation.
-        run: |
-          set -e
-          HOST="archiv.raddatz.cloud"
-          URL="https://$HOST"
-          HOST_IP=$(awk 'NR>1 && $2=="00000000"{h=$3;printf "%d.%d.%d.%d\n",strtonum("0x"substr(h,7,2)),strtonum("0x"substr(h,5,2)),strtonum("0x"substr(h,3,2)),strtonum("0x"substr(h,1,2));exit}' /proc/net/route)
-          [ -n "$HOST_IP" ] || { echo "ERROR: could not detect Docker bridge gateway via /proc/net/route"; exit 1; }
-          RESOLVE=(--resolve "$HOST:443:$HOST_IP")
-          echo "Smoke test: $URL (pinned to $HOST_IP via bridge gateway)"
-          curl -fsS "${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"
+      - uses: ./.gitea/actions/smoke-test
+        with:
+          host: archiv.raddatz.cloud
 
       - name: Cleanup env file
         # LOAD-BEARING: `if: always()` is the linchpin of the ADR-011

.gitignore

@@ -30,3 +30,6 @@ frontend/yarn.lock
 **/.venv/
 **/__pycache__/
 *.pyc
+
+# Canonical import artifacts live only on the ops host (PII).
+# See tools/import-normalizer/.gitignore — load-bearing for that policy.

CLAUDE.md

@@ -92,6 +92,7 @@ backend/src/main/java/org/raddatz/familienarchiv/
 ├── ocr/                 OCR domain — OcrService, OcrBatchService, training
 ├── person/              Person domain
 │   └── relationship/    PersonRelationship sub-domain
+├── search/              NL search domain — NlSearchController, NlQueryParserService, RestClientOllamaClient, NlSearchRateLimiter
 ├── security/            SecurityConfig, Permission, @RequirePermission, PermissionAspect
 ├── tag/                 Tag domain
 └── user/                User domain — AppUser, UserGroup, UserService
@@ -160,7 +161,7 @@ Input DTOs live flat in the domain package. Response types are the model entitie
 
 → See [CONTRIBUTING.md §Error handling](./CONTRIBUTING.md#error-handling)
 
-**LLM reminder:** use `DomainException.notFound/forbidden/conflict/internal()` from service methods — never throw raw exceptions. When adding a new `ErrorCode`: (1) add to `ErrorCode.java`, (2) add to `ErrorCode` type in `frontend/src/lib/shared/errors.ts`, (3) add a `case` in `getErrorMessage()`, (4) add i18n keys in `messages/{de,en,es}.json`. Valid error codes include: `TOO_MANY_LOGIN_ATTEMPTS` (returned by `LoginRateLimiter` as HTTP 429 when a brute-force threshold is exceeded).
+**LLM reminder:** use `DomainException.notFound/forbidden/conflict/internal()` from service methods — never throw raw exceptions. When adding a new `ErrorCode`: (1) add to `ErrorCode.java`, (2) add to `ErrorCode` type in `frontend/src/lib/shared/errors.ts`, (3) add a `case` in `getErrorMessage()`, (4) add i18n keys in `messages/{de,en,es}.json`. Valid error codes include: `TOO_MANY_LOGIN_ATTEMPTS` (returned by `LoginRateLimiter` as HTTP 429 when a brute-force threshold is exceeded); `SMART_SEARCH_UNAVAILABLE` (HTTP 503 — Ollama inference service offline or timed out); `SMART_SEARCH_RATE_LIMITED` (HTTP 429 — user exceeded 5 NL search requests per minute).
 
 ### Security / Permissions
 
@@ -194,10 +195,10 @@ frontend/src/routes/
 │   ├── [id]/edit/          Person edit form
 │   ├── new/                Create person form
 │   └── review/             Triage view — confirm/rename/merge/delete provisional persons
-├── briefwechsel/           Bilateral conversation timeline (Briefwechsel)
 ├── aktivitaeten/           Unified activity feed (Chronik)
 ├── geschichten/            Stories — list, [id], [id]/edit, new
 ├── stammbaum/              Family tree (Stammbaum)
+├── themen/                 Topics directory — browsable tag index
 ├── enrich/                 Enrichment workflow — [id], done
 ├── admin/                  User, group, tag, OCR, system management
 ├── hilfe/transkription/    Transcription help page
@@ -268,7 +269,7 @@ Back button pattern — use the shared `<BackButton>` component from `$lib/share
 
 → See [CONTRIBUTING.md §Error handling](./CONTRIBUTING.md#error-handling)
 
-**LLM reminder:** when adding a new `ErrorCode`: (1) add to `ErrorCode.java`, (2) add to `ErrorCode` type in `frontend/src/lib/shared/errors.ts`, (3) add a `case` in `getErrorMessage()`, (4) add i18n keys in `messages/{de,en,es}.json`. Valid error codes include: `TOO_MANY_LOGIN_ATTEMPTS` (returned by `LoginRateLimiter` as HTTP 429 when a brute-force threshold is exceeded).
+**LLM reminder:** when adding a new `ErrorCode`: (1) add to `ErrorCode.java`, (2) add to `ErrorCode` type in `frontend/src/lib/shared/errors.ts`, (3) add a `case` in `getErrorMessage()`, (4) add i18n keys in `messages/{de,en,es}.json`. Valid error codes include: `TOO_MANY_LOGIN_ATTEMPTS` (returned by `LoginRateLimiter` as HTTP 429 when a brute-force threshold is exceeded); `SMART_SEARCH_UNAVAILABLE` (HTTP 503 — Ollama inference service offline or timed out); `SMART_SEARCH_RATE_LIMITED` (HTTP 429 — user exceeded 5 NL search requests per minute).
 
 ---
 

backend/api_tests/Admin-Auth.http

@@ -28,4 +28,18 @@ Authorization: Basic Gast_User gast
 ###Groups
 #GET
 GET http://localhost:8080/api/admin/tags
-Authorization: Basic admin admin123
+Authorization: Basic admin admin123
+
+### One-time backfill: re-sync already-stale auto-titles (#726)
+# RUNBOOK: a one-shot ADMIN maintenance call, NOT part of normal operation. Run it ONCE
+# after deploying #726 to clean the existing backlog of stale titles (e.g. a title still
+# showing "2028" after the date was corrected to "1928"). It is synchronous and idempotent
+# — a second run returns {"count": 0} and writes nothing. Hit the backend DIRECTLY on
+# port 8080 (NOT through the SvelteKit proxy) so the sweep can't trip the proxy timeout.
+# Returns {"count": <documents rewritten>}.
+POST http://localhost:8080/api/admin/backfill-titles
+Authorization: Basic admin admin123
+
+### NEGATIV-TEST: ein Nicht-Admin darf den Backfill NICHT auslösen -> 403 Forbidden
+POST http://localhost:8080/api/admin/backfill-titles
+Authorization: Basic Gast_User gast

backend/pom.xml

@@ -41,6 +41,27 @@
 				<type>pom</type>
 				<scope>import</scope>
 			</dependency>
+			<!-- Force WireMock's ee10 Jetty transitive deps to match Spring Boot's 12.1.8 core -->
+			<dependency>
+				<groupId>org.eclipse.jetty.ee10</groupId>
+				<artifactId>jetty-ee10-servlet</artifactId>
+				<version>12.1.8</version>
+			</dependency>
+			<dependency>
+				<groupId>org.eclipse.jetty.ee10</groupId>
+				<artifactId>jetty-ee10-servlets</artifactId>
+				<version>12.1.8</version>
+			</dependency>
+			<dependency>
+				<groupId>org.eclipse.jetty.ee10</groupId>
+				<artifactId>jetty-ee10-webapp</artifactId>
+				<version>12.1.8</version>
+			</dependency>
+			<dependency>
+				<groupId>org.eclipse.jetty</groupId>
+				<artifactId>jetty-ee</artifactId>
+				<version>12.1.8</version>
+			</dependency>
 		</dependencies>
 	</dependencyManagement>
 	<dependencies>
@@ -137,6 +158,12 @@
 			<artifactId>archunit-junit5</artifactId>
 			<version>1.3.0</version>
 			<scope>test</scope>
+		</dependency>
+		<dependency>
+			<groupId>org.wiremock</groupId>
+			<artifactId>wiremock-jetty12</artifactId>
+			<version>3.9.2</version>
+			<scope>test</scope>
 		</dependency>
 				<!-- Excel Bearbeitung (Apache POI) -->
 		<dependency>

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

@@ -177,6 +177,13 @@ public class Document {
     @Builder.Default
     private Set<TrainingLabel> trainingLabels = new HashSet<>();
 
+    // Not persisted — computed per detail fetch so read-only users can tell at first
+    // paint whether there is a transcription to read (DocumentService.getDocumentById).
+    @Transient
+    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+    @Builder.Default
+    private boolean hasTranscription = false;
+
     // The `?v={thumbnailGeneratedAt}` cache-buster is load-bearing: the thumbnail
     // endpoint sends `Cache-Control: private, max-age=31536000, immutable`
     // (DocumentController.getDocumentThumbnail). `immutable` is only safe because

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

@@ -3,7 +3,6 @@ package org.raddatz.familienarchiv.document;
 import java.io.IOException;
 import java.time.LocalDate;
 import java.util.ArrayList;
-import java.util.concurrent.TimeUnit;
 import java.util.LinkedHashSet;
 import java.util.List;
 import java.util.Map;
@@ -47,9 +46,7 @@ import org.raddatz.familienarchiv.document.DocumentService;
 import org.raddatz.familienarchiv.document.DocumentVersionService;
 import org.raddatz.familienarchiv.filestorage.FileService;
 import org.raddatz.familienarchiv.user.UserService;
-import org.springframework.data.domain.Sort;
 import org.springframework.security.core.Authentication;
-import org.springframework.http.CacheControl;
 import org.springframework.http.HttpHeaders;
 import org.springframework.http.MediaType;
 import org.springframework.http.ResponseEntity;
@@ -138,7 +135,7 @@ public class DocumentController {
     // --- METADATA ---
     @GetMapping("/{id}")
     public Document getDocument(@PathVariable UUID id) {
-        return documentService.getDocumentById(id);
+        return documentService.getDocumentDetail(id);
     }
 
     @PostMapping(consumes = MediaType.MULTIPART_FORM_DATA_VALUE)
@@ -316,7 +313,8 @@ public class DocumentController {
             @RequestParam(required = false) Boolean undated,
             Authentication authentication) {
         TagOperator operator = "OR".equalsIgnoreCase(tagOp) ? TagOperator.OR : TagOperator.AND;
-        List<UUID> ids = documentService.findIdsForFilter(q, from, to, senderId, receiverId, tags, tagQ, status, operator, Boolean.TRUE.equals(undated));
+        SearchFilters filters = new SearchFilters(q, from, to, senderId, receiverId, tags, tagQ, status, operator, Boolean.TRUE.equals(undated));
+        List<UUID> ids = documentService.findIdsForFilter(filters);
         if (ids.size() > BULK_EDIT_FILTER_MAX_IDS) {
             throw DomainException.badRequest(ErrorCode.BULK_EDIT_TOO_MANY_IDS,
                     "Filter matches " + ids.size() + " documents — refine filter (max " + BULK_EDIT_FILTER_MAX_IDS + ")");
@@ -388,8 +386,9 @@ public class DocumentController {
         // tagOp is a raw String at the HTTP boundary; any value other than "OR" (case-insensitive)
         // defaults to AND, which matches the frontend default and keeps old clients working.
         TagOperator operator = "OR".equalsIgnoreCase(tagOp) ? TagOperator.OR : TagOperator.AND;
+        SearchFilters filters = new SearchFilters(q, from, to, senderId, receiverId, tags, tagQ, status, operator, Boolean.TRUE.equals(undated));
         Pageable pageable = PageRequest.of(page, size);
-        return ResponseEntity.ok(documentService.searchDocuments(q, from, to, senderId, receiverId, tags, tagQ, status, sort, dir, operator, Boolean.TRUE.equals(undated), pageable));
+        return ResponseEntity.ok(documentService.searchDocuments(filters, sort, dir, pageable));
     }
 
     @GetMapping(value = "/density", produces = MediaType.APPLICATION_JSON_VALUE)
@@ -404,9 +403,7 @@ public class DocumentController {
         TagOperator operator = "OR".equalsIgnoreCase(tagOp) ? TagOperator.OR : TagOperator.AND;
         DocumentDensityResult result = documentService.getDensity(
                 new DensityFilters(q, senderId, receiverId, tags, tagQ, status, operator));
-        return ResponseEntity.ok()
-                .cacheControl(CacheControl.maxAge(5, TimeUnit.MINUTES).cachePrivate())
-                .body(result);
+        return ResponseEntity.ok(result);
     }
 
     // --- TRAINING LABELS ---
@@ -445,17 +442,6 @@ public class DocumentController {
         return documentVersionService.getVersion(id, versionId);
     }
 
-    @GetMapping("/conversation")
-    public List<Document> getConversation(
-            @RequestParam UUID senderId,
-            @RequestParam(required = false) UUID receiverId,
-            @RequestParam(required = false) LocalDate from,
-            @RequestParam(required = false) LocalDate to,
-            @RequestParam(defaultValue = "DESC") String dir) {
-        Sort sort = Sort.by(Sort.Direction.fromString(dir.toUpperCase()), "documentDate");
-        return documentService.getConversationFiltered(senderId, receiverId, from, to, sort);
-    }
-
     private UUID requireUserId(Authentication authentication) {
         return SecurityUtils.requireUserId(authentication, userService);
     }

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

@@ -6,6 +6,7 @@ import org.raddatz.familienarchiv.person.Person;
 import org.raddatz.familienarchiv.tag.Tag;
 
 import java.time.LocalDate;
+import java.time.LocalDateTime;
 import java.util.List;
 import java.util.UUID;
 
@@ -35,5 +36,9 @@ public record DocumentListItem(
         @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
         List<ActivityActorDTO> contributors,
         @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
-        SearchMatchData matchData
+        SearchMatchData matchData,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        LocalDateTime createdAt,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        LocalDateTime updatedAt
 ) {}

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

@@ -15,7 +15,6 @@ import org.springframework.data.jpa.repository.Query;
 import org.springframework.data.repository.query.Param;
 import org.springframework.stereotype.Repository;
 
-import java.time.LocalDate;
 import java.util.Collection;
 import java.util.List;
 import java.util.Map;
@@ -58,6 +57,7 @@ public interface DocumentRepository extends JpaRepository<Document, UUID>, JpaSp
     @EntityGraph("Document.full")
     List<Document> findByReceiversId(UUID receiverId);
 
+
     // Callers access only doc.getTags() to mutate the set — receivers/sender not touched; no graph needed.
     List<Document> findByTags_Id(UUID tagId);
 
@@ -81,32 +81,6 @@ public interface DocumentRepository extends JpaRepository<Document, UUID>, JpaSp
 
     Optional<Document> findFirstByMetadataCompleteFalseAndIdNot(UUID id, Sort sort);
 
-    @EntityGraph("Document.full")
-    @Query("SELECT DISTINCT d FROM Document d " +
-            "JOIN d.receivers r " +
-            "WHERE " +
-            "((d.sender.id = :person1 AND r.id = :person2) " +
-            " OR " +
-            " (d.sender.id = :person2 AND r.id = :person1)) " +
-            "AND d.documentDate BETWEEN :from AND :to")
-    List<Document> findConversation(
-            @Param("person1") UUID person1,
-            @Param("person2") UUID person2,
-            @Param("from") LocalDate from,
-            @Param("to") LocalDate to,
-            Sort sort);
-
-    @EntityGraph("Document.full")
-    @Query("SELECT DISTINCT d FROM Document d " +
-            "LEFT JOIN d.receivers r " +
-            "WHERE (d.sender.id = :personId OR r.id = :personId) " +
-            "AND d.documentDate BETWEEN :from AND :to")
-    List<Document> findSinglePersonCorrespondence(
-            @Param("personId") UUID personId,
-            @Param("from") LocalDate from,
-            @Param("to") LocalDate to,
-            Sort sort);
-
     @Query(nativeQuery = true, value = """
             SELECT d.id FROM documents d
             CROSS JOIN LATERAL (

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

@@ -32,6 +32,8 @@ import org.springframework.data.domain.Page;
 import org.springframework.data.domain.PageRequest;
 import org.springframework.data.domain.Pageable;
 import org.springframework.data.domain.Sort;
+import jakarta.persistence.criteria.JoinType;
+import jakarta.persistence.criteria.Predicate;
 import org.springframework.data.jpa.domain.Specification;
 import org.raddatz.familienarchiv.exception.DomainException;
 import org.raddatz.familienarchiv.exception.ErrorCode;
@@ -68,6 +70,7 @@ import static org.raddatz.familienarchiv.document.DocumentSpecifications.*;
 public class DocumentService {
 
     private final DocumentRepository documentRepository;
+    private final DocumentTitleFactory documentTitleFactory;
     private final PersonService personService;
     private final FileService fileService;
     private final TagService tagService;
@@ -137,8 +140,10 @@ public class DocumentService {
      * <p>Implementation note: groups in memory rather than via SQL GROUP BY
      * because the existing {@link Specification} predicates compose easily
      * with {@code findAll(spec)} and the archive size (≈5k docs) keeps this
-     * well under the 200ms p95 target. Cache-Control: max-age=300 on the
-     * controller layer absorbs repeated browse loads.
+     * well under the 200ms p95 target. The controller sets no explicit
+     * Cache-Control, so the response is served fresh on every load (issue
+     * #709) — the recompute is imperceptible and stale month counts after an
+     * edit would be misleading on an interactive chart.
      *
      * <p>Tracked in issue #481 for re-evaluation when {@code documents > 50k}
      * — at that scale move the aggregation into SQL (GROUP BY TO_CHAR(meta_date,
@@ -167,11 +172,13 @@ public class DocumentService {
     /** Loads matching documents and projects to non-null {@link LocalDate}s. */
     private List<LocalDate> loadFilteredDates(DensityFilters filters, List<UUID> ftsIds) {
         boolean hasFts = ftsIds != null;
-        Specification<Document> spec = buildSearchSpec(
-                hasFts, ftsIds, null, null,
-                filters.sender(), filters.receiver(),
-                filters.tags(), filters.tagQ(),
-                filters.status(), filters.tagOperator(), false);
+        // Density and search keep separate filter records (DensityFilters has no
+        // date/undated fields); adapt to SearchFilters here to reuse buildSearchSpec.
+        // Date bounds stay null and undated=false — the density path never filters by date.
+        SearchFilters searchFilters = new SearchFilters(
+                filters.text(), null, null, filters.sender(), filters.receiver(),
+                filters.tags(), filters.tagQ(), filters.status(), filters.tagOperator(), false);
+        Specification<Document> spec = buildSearchSpec(hasFts, ftsIds, searchFilters);
         return documentRepository.findAll(spec).stream()
                 .map(Document::getDocumentDate)
                 .filter(Objects::nonNull)
@@ -375,10 +382,17 @@ public class DocumentService {
 
         DocumentStatus statusBefore = doc.getStatus();
 
+        // Auto-title sync (#726): capture the machine title from the CURRENTLY-persisted state
+        // BEFORE any setter runs — the setters below overwrite date/location and applyDatePrecision
+        // skips nulls, so the old state must be read first. The submitted title is the catalog
+        // auto-title iff it equals this; only then does it follow date/location forward.
+        String autoTitleBefore = documentTitleFactory.build(doc);
+
         // 1. Einfache Felder Update
-        doc.setTitle(dto.getTitle());
+        doc.setTitle(resolveTitle(dto.getTitle(), autoTitleBefore, doc, dto));
         doc.setDocumentDate(dto.getDocumentDate());
         applyDatePrecision(doc, dto);
+        validateDateRange(doc); // guard before any save (updateDocumentTags below persists)
         doc.setLocation(dto.getLocation());
         doc.setTranscription(dto.getTranscription());
         doc.setSummary(dto.getSummary());
@@ -419,7 +433,11 @@ public class DocumentService {
             doc.setScriptType(dto.getScriptType());
         }
 
-        // 4. Datei austauschen (nur wenn eine neue ausgewählt wurde)
+        // 4. Datei austauschen (nur wenn eine neue ausgewählt wurde).
+        // NB (#726): this reassigns originalFilename to the uploaded file's name. The title's index
+        // segment is originalFilename, so after a replace the stored title no longer matches
+        // build(currentState) and the row is treated as manual — neither save-time nor backfill
+        // rewrites it. Accepted fail-safe (ADR-031), and autoTitleBefore was already captured above.
         boolean fileReplaced = newFile != null && !newFile.isEmpty();
         if (fileReplaced) {
             FileService.UploadResult upload = fileService.uploadFile(newFile, newFile.getOriginalFilename());
@@ -448,21 +466,92 @@ public class DocumentService {
     }
 
     /**
-     * Applies the three date-precision fields only when the DTO carries them.
-     * A null field means "not submitted" — overwriting the stored value with null
-     * would fabricate a precision the user never chose, the exact dishonesty #666
-     * exists to prevent. A row with a genuinely-unknown precision must keep it when
-     * an unrelated edit (e.g. a location typo) is saved.
+     * Decides the title to persist on an edit (#726). The submitted title is the catalog
+     * auto-title only when it equals {@code autoBefore} (built from the stored state) — an exact
+     * comparison with no heuristic, relying on the edit form round-tripping the stored title
+     * verbatim when untouched. A machine title is rebuilt from the new state so a corrected
+     * date/location flows into it; a hand-written or freshly-typed title is kept verbatim. A blank
+     * submission is never persisted (title is always present) — it falls back to the rebuilt
+     * auto-title, which always carries at least the index.
+     */
+    private String resolveTitle(String submitted, String autoBefore, Document doc, DocumentUpdateDTO dto) {
+        if (submitted == null || submitted.isBlank()) {
+            return documentTitleFactory.build(projectedState(doc, dto));
+        }
+        if (!Objects.equals(submitted, autoBefore)) {
+            return submitted;
+        }
+        return documentTitleFactory.build(projectedState(doc, dto));
+    }
+
+    /**
+     * The document state the regenerated title is built from. It is composed from the SAME
+     * resolvers the real setters use — {@code documentDate}/{@code location} overwritten from the
+     * DTO (a null value clears the field), precision/end/raw resolved skip-null via
+     * {@link #effectivePrecision}/{@link #effectiveMetaDateEnd}/{@link #effectiveMetaDateRaw} — so
+     * the projection cannot drift from {@link #updateDocument}. The index ({@code originalFilename})
+     * is never touched by a metadata edit.
+     */
+    private Document projectedState(Document doc, DocumentUpdateDTO dto) {
+        return Document.builder()
+                .originalFilename(doc.getOriginalFilename())
+                .documentDate(dto.getDocumentDate())
+                .location(dto.getLocation())
+                .metaDatePrecision(effectivePrecision(doc, dto))
+                .metaDateEnd(effectiveMetaDateEnd(doc, dto))
+                .metaDateRaw(effectiveMetaDateRaw(doc, dto))
+                .build();
+    }
+
+    /**
+     * Applies the three date-precision fields skip-null: a null DTO field means "not submitted",
+     * so the stored value is kept rather than overwritten with null — which would fabricate a
+     * precision the user never chose, the exact dishonesty #666 exists to prevent. Expressed via
+     * the shared {@code effective*} resolvers so {@link #projectedState} stays lock-step (writing
+     * the stored value back when the DTO omits a field is a harmless no-op).
      */
     private void applyDatePrecision(Document doc, DocumentUpdateDTO dto) {
-        if (dto.getMetaDatePrecision() != null) {
-            doc.setMetaDatePrecision(dto.getMetaDatePrecision());
+        doc.setMetaDatePrecision(effectivePrecision(doc, dto));
+        doc.setMetaDateEnd(effectiveMetaDateEnd(doc, dto));
+        doc.setMetaDateRaw(effectiveMetaDateRaw(doc, dto));
+    }
+
+    // Skip-null date-field resolution shared by applyDatePrecision (the real setters) and
+    // projectedState (the title projection) — the single rule keeps them from diverging (#726).
+    private static DatePrecision effectivePrecision(Document doc, DocumentUpdateDTO dto) {
+        return dto.getMetaDatePrecision() != null ? dto.getMetaDatePrecision() : doc.getMetaDatePrecision();
+    }
+
+    private static LocalDate effectiveMetaDateEnd(Document doc, DocumentUpdateDTO dto) {
+        return dto.getMetaDateEnd() != null ? dto.getMetaDateEnd() : doc.getMetaDateEnd();
+    }
+
+    private static String effectiveMetaDateRaw(Document doc, DocumentUpdateDTO dto) {
+        return dto.getMetaDateRaw() != null ? dto.getMetaDateRaw() : doc.getMetaDateRaw();
+    }
+
+    /**
+     * Friendly guard for the two V69 date-range CHECK constraints, run before save so a
+     * user date typo returns a clean 400 INVALID_DATE_RANGE instead of falling through to
+     * the generic handler (HTTP 500 + Sentry + ERROR log). Validates the post-apply {@code doc}
+     * state, not the DTO, because precision/end may have been carried over from the stored row
+     * when the DTO field was null. The DB CHECK remains the backstop; this never weakens it.
+     */
+    private void validateDateRange(Document doc) {
+        // Mirrors chk_meta_date_end_after_start: end >= start, with null start allowed.
+        // Use isBefore (equal dates are valid) — never !isAfter, which would contradict the DB's >=.
+        if (doc.getMetaDatePrecision() == DatePrecision.RANGE
+                && doc.getDocumentDate() != null
+                && doc.getMetaDateEnd() != null
+                && doc.getMetaDateEnd().isBefore(doc.getDocumentDate())) {
+            throw DomainException.badRequest(ErrorCode.INVALID_DATE_RANGE,
+                    "meta_date_end must not be before meta_date");
         }
-        if (dto.getMetaDateEnd() != null) {
-            doc.setMetaDateEnd(dto.getMetaDateEnd());
-        }
-        if (dto.getMetaDateRaw() != null) {
-            doc.setMetaDateRaw(dto.getMetaDateRaw());
+        // Mirrors chk_meta_date_end_only_for_range. API-only: the edit form clears the
+        // end field off-RANGE, so this branch closes the same 500 class for direct clients.
+        if (doc.getMetaDateEnd() != null && doc.getMetaDatePrecision() != DatePrecision.RANGE) {
+            throw DomainException.badRequest(ErrorCode.INVALID_DATE_RANGE,
+                    "meta_date_end is only allowed when meta_date_precision is RANGE");
         }
     }
 
@@ -500,18 +589,15 @@ public class DocumentService {
      * round-trip.
      */
     @Transactional(readOnly = true)
-    public List<UUID> findIdsForFilter(String text, LocalDate from, LocalDate to, UUID sender, UUID receiver,
-                                       List<String> tags, String tagQ, DocumentStatus status, TagOperator tagOperator,
-                                       boolean undated) {
-        boolean hasText = StringUtils.hasText(text);
+    public List<UUID> findIdsForFilter(SearchFilters filters) {
+        boolean hasText = StringUtils.hasText(filters.text());
         List<UUID> rankedIds = null;
         if (hasText) {
-            rankedIds = documentRepository.findAllMatchingIdsByFts(text);
+            rankedIds = documentRepository.findAllMatchingIdsByFts(filters.text());
             if (rankedIds.isEmpty()) return List.of();
         }
 
-        Specification<Document> spec = buildSearchSpec(
-                hasText, rankedIds, from, to, sender, receiver, tags, tagQ, status, tagOperator, undated);
+        Specification<Document> spec = buildSearchSpec(hasText, rankedIds, filters);
         return documentRepository.findAll(spec).stream().map(Document::getId).toList();
     }
 
@@ -521,23 +607,18 @@ public class DocumentService {
      * (uncapped, ID-only). Caller does its own FTS short-circuit when the
      * full-text query returned no rows.
      */
-    private Specification<Document> buildSearchSpec(boolean hasText, List<UUID> ftsIds,
-                                                    LocalDate from, LocalDate to,
-                                                    UUID sender, UUID receiver,
-                                                    List<String> tags, String tagQ,
-                                                    DocumentStatus status, TagOperator tagOperator,
-                                                    boolean undated) {
-        boolean useOrLogic = tagOperator == TagOperator.OR;
-        List<Set<UUID>> expandedTagSets = tagService.expandTagNamesToDescendantIdSets(tags);
+    private Specification<Document> buildSearchSpec(boolean hasText, List<UUID> ftsIds, SearchFilters filters) {
+        boolean useOrLogic = filters.tagOperator() == TagOperator.OR;
+        List<Set<UUID>> expandedTagSets = tagService.expandTagNamesToDescendantIdSets(filters.tags());
         Specification<Document> textSpec = hasText ? hasIds(ftsIds) : (root, query, cb) -> null;
         return Specification.where(textSpec)
-                .and(isBetween(from, to))
-                .and(hasSender(sender))
-                .and(hasReceiver(receiver))
+                .and(isBetween(filters.from(), filters.to()))
+                .and(hasSender(filters.sender()))
+                .and(hasReceiver(filters.receiver()))
                 .and(hasTags(expandedTagSets, useOrLogic))
-                .and(hasTagPartial(tagQ))
-                .and(hasStatus(status))
-                .and(undatedOnly(undated));
+                .and(hasTagPartial(filters.tagQ()))
+                .and(hasStatus(filters.status()))
+                .and(undatedOnly(filters.undated()));
     }
 
     /**
@@ -666,8 +747,8 @@ 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, boolean undated, Pageable pageable) {
-        boolean hasText = StringUtils.hasText(text);
+    public DocumentSearchResult searchDocuments(SearchFilters filters, DocumentSort sort, String dir, Pageable pageable) {
+        boolean hasText = StringUtils.hasText(filters.text());
 
         // Pure-text RELEVANCE: push pagination + ts_rank ordering into SQL — skip
         // findAllMatchingIdsByFts entirely (ADR-008). This must run BEFORE any
@@ -677,13 +758,13 @@ public class DocumentService {
         // no date/sender/receiver/tag/status filters, and undated documents are valid
         // FTS hits already folded into the ranked page, so there is no separate undated
         // count to report here.
-        if (!undated && isPureTextRelevance(hasText, sort, from, to, sender, receiver, tags, tagQ, status)) {
-            return relevanceSortedPageFromSql(text, pageable);
+        if (!filters.undated() && isPureTextRelevance(hasText, sort, filters)) {
+            return relevanceSortedPageFromSql(filters.text(), pageable);
         }
 
         List<UUID> rankedIds = null;
         if (hasText) {
-            rankedIds = documentRepository.findAllMatchingIdsByFts(text);
+            rankedIds = documentRepository.findAllMatchingIdsByFts(filters.text());
             // FTS matched nothing → no results and, by definition, no undated matches either.
             if (rankedIds.isEmpty()) return DocumentSearchResult.of(List.of());
         }
@@ -691,37 +772,32 @@ public class DocumentService {
         // Global undated count for the current filter (q/tags/sender/receiver/status),
         // forcing undatedOnly(true) and IGNORING the user's "Nur undatierte" toggle so
         // it never collapses to the page slice and never double-counts (issue #668).
-        long undatedCount = countUndatedForFilter(hasText, rankedIds, from, to, sender, receiver, tags, tagQ, status, tagOperator);
+        long undatedCount = countUndatedForFilter(hasText, rankedIds, filters.withUndated(true));
 
-        return runSearch(text, hasText, rankedIds, from, to, sender, receiver, tags, tagQ, status, sort, dir, tagOperator, undated, pageable)
+        return runSearch(hasText, rankedIds, filters, sort, dir, pageable)
                 .withUndatedCount(undatedCount);
     }
 
     /**
      * Counts every undated document (meta_date IS NULL) matching the active filter,
-     * across all pages, independent of the undated toggle. Reuses {@link #buildSearchSpec}
-     * with {@code undated=true} forced so the count tracks q/tags/sender/receiver/status.
-     * A {@code from}/{@code to} range excludes undated rows by the collision rule (#668),
-     * so the count is legitimately 0 inside a date range.
+     * across all pages, independent of the undated toggle. The caller passes
+     * {@code filters.withUndated(true)} so the count tracks q/tags/sender/receiver/status
+     * regardless of the user's "Nur undatierte" toggle. A {@code from}/{@code to} range
+     * excludes undated rows by the collision rule (#668), so the count is legitimately 0
+     * inside a date range.
      */
-    private long countUndatedForFilter(boolean hasText, List<UUID> ftsIds,
-                                       LocalDate from, LocalDate to, UUID sender, UUID receiver,
-                                       List<String> tags, String tagQ, DocumentStatus status, TagOperator tagOperator) {
-        Specification<Document> undatedSpec = buildSearchSpec(
-                hasText, ftsIds, from, to, sender, receiver, tags, tagQ, status, tagOperator, true);
+    private long countUndatedForFilter(boolean hasText, List<UUID> ftsIds, SearchFilters filters) {
+        Specification<Document> undatedSpec = buildSearchSpec(hasText, ftsIds, filters);
         return documentRepository.count(undatedSpec);
     }
 
     /** The original search dispatch — produces the page slice + totals, sans undated count. */
-    private DocumentSearchResult runSearch(String text, boolean hasText, List<UUID> rankedIds,
-                                           LocalDate from, LocalDate to, UUID sender, UUID receiver,
-                                           List<String> tags, String tagQ, DocumentStatus status,
-                                           DocumentSort sort, String dir, TagOperator tagOperator,
-                                           boolean undated, Pageable pageable) {
+    private DocumentSearchResult runSearch(boolean hasText, List<UUID> rankedIds, SearchFilters filters,
+                                           DocumentSort sort, String dir, Pageable pageable) {
         // The pure-text RELEVANCE fast path is handled by the caller (searchDocuments)
         // before findAllMatchingIdsByFts runs, so it never reaches here (ADR-008).
-        Specification<Document> spec = buildSearchSpec(
-                hasText, rankedIds, from, to, sender, receiver, tags, tagQ, status, tagOperator, undated);
+        Specification<Document> spec = buildSearchSpec(hasText, rankedIds, filters);
+        String text = filters.text();
 
         // 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
@@ -755,12 +831,12 @@ 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) {
+    private static boolean isPureTextRelevance(boolean hasText, DocumentSort sort, SearchFilters filters) {
         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;
+                && filters.from() == null && filters.to() == null
+                && filters.sender() == null && filters.receiver() == null
+                && (filters.tags() == null || filters.tags().isEmpty())
+                && (filters.tagQ() == null || filters.tagQ().isBlank()) && filters.status() == null;
     }
 
     /**
@@ -832,7 +908,9 @@ public class DocumentService {
                 doc.getSummary(),
                 completionPct,
                 contributors,
-                match
+                match,
+                doc.getCreatedAt(),
+                doc.getUpdatedAt()
         );
     }
 
@@ -899,22 +977,6 @@ public class DocumentService {
                 .orElse("");
     }
 
-    // 2. SPEZIALITÄT: Der Schriftwechsel
-    // Findet alle Briefe ZWISCHEN zwei Personen (egal wer Sender/Empfänger war)
-    public List<Document> getConversation(UUID personA, UUID personB) {
-
-        // Fall 1: A schreibt an B
-        Specification<Document> aToB = Specification.where(hasSender(personA)).and(hasReceiver(personB));
-
-        // Fall 2: B schreibt an A
-        Specification<Document> bToA = Specification.where(hasSender(personB)).and(hasReceiver(personA));
-
-        // Wir wollen (A->B) ODER (B->A)
-        Specification<Document> conversation = aToB.or(bToA);
-
-        return documentRepository.findAll(conversation, Sort.by(Sort.Direction.ASC, "documentDate"));
-    }
-
     @Transactional
     public void updateScriptType(UUID documentId, ScriptType scriptType) {
         Document doc = getDocumentById(documentId);
@@ -944,6 +1006,19 @@ public class DocumentService {
         return doc;
     }
 
+    /**
+     * Loads a document for the detail view, additionally flagging whether it has any
+     * transcription to read. Kept separate from {@link #getDocumentById} so the cheap
+     * existence query only runs for the single-document detail endpoint, not for the
+     * many internal callers that never read the flag.
+     */
+    @Transactional(readOnly = true)
+    public Document getDocumentDetail(UUID id) {
+        Document doc = getDocumentById(id);
+        doc.setHasTranscription(transcriptionBlockQueryService.hasBlocks(id));
+        return doc;
+    }
+
     public List<Document> getDocumentsByIds(List<UUID> ids) {
         return documentRepository.findAllById(ids);
     }
@@ -960,13 +1035,26 @@ public class DocumentService {
         return documentRepository.findByReceiversId(receiverId);
     }
 
-    public List<Document> getConversationFiltered(UUID senderId, UUID receiverId, LocalDate from, LocalDate to, Sort sort) {
-        LocalDate dateFrom = (from != null) ? from : LocalDate.parse("0000-01-01");
-        LocalDate dateTo = (to != null) ? to : LocalDate.now();
-        if (receiverId == null) {
-            return documentRepository.findSinglePersonCorrespondence(senderId, dateFrom, dateTo, sort);
-        }
-        return documentRepository.findConversation(senderId, receiverId, dateFrom, dateTo, sort);
+    public DocumentSearchResult searchDocumentsByPersonId(UUID personId, LocalDate from, LocalDate to, Pageable pageable) {
+        Person person = personService.getById(personId);
+        Specification<Document> spec = buildPersonSpec(person, from, to);
+        Page<Document> page = documentRepository.findAll(spec, pageable);
+        List<DocumentListItem> items = enrichItems(page.getContent(), null);
+        return DocumentSearchResult.paged(items, pageable, page.getTotalElements());
+    }
+
+    private Specification<Document> buildPersonSpec(Person person, LocalDate from, LocalDate to) {
+        return (root, query, cb) -> {
+            if (query != null) query.distinct(true);
+            var receiversJoin = root.join("receivers", JoinType.LEFT);
+            var senderPredicate = cb.equal(root.get("sender"), person);
+            var receiverPredicate = cb.equal(receiversJoin, person);
+            var personPredicate = cb.or(senderPredicate, receiverPredicate);
+            var predicates = new ArrayList<>(List.of(personPredicate));
+            if (from != null) predicates.add(cb.greaterThanOrEqualTo(root.get("documentDate"), from));
+            if (to != null) predicates.add(cb.lessThanOrEqualTo(root.get("documentDate"), to));
+            return cb.and(predicates.toArray(new Predicate[0]));
+        };
     }
 
     public long getIncompleteCount() {
@@ -1003,6 +1091,43 @@ public class DocumentService {
         tagService.delete(tagId);
     }
 
+    /**
+     * One-time cleanup of already-stale auto-titles (#726, FR-003). For every document whose
+     * stored title passes the {@link DocumentTitleBackfillMatcher} overwrite heuristic, rebuilds
+     * the title from the row's current state and persists it only when it actually changed.
+     * Idempotent: a second run rebuilds the same value and saves nothing. Hand-written prose is
+     * left untouched.
+     *
+     * <p>Saves via {@code documentRepository.save} directly — it must NOT route through
+     * {@link #updateDocument} (which versions every write), following the {@link #backfillFileHashes}
+     * precedent: a mechanical rename must not snapshot the whole corpus into {@code document_versions}.
+     *
+     * @return the number of documents whose title was rewritten
+     */
+    @Transactional
+    public int backfillTitles() {
+        List<Document> docs = documentRepository.findAll();
+        int updated = 0;
+        int skipped = 0;
+        for (Document doc : docs) {
+            if (!DocumentTitleBackfillMatcher.isOverwritable(
+                    doc.getTitle(), doc.getOriginalFilename(), doc.getLocation())) {
+                skipped++;
+                continue;
+            }
+            String rebuilt = documentTitleFactory.build(doc);
+            if (rebuilt.equals(doc.getTitle())) {
+                skipped++; // already correct — keep idempotent, no write
+                continue;
+            }
+            doc.setTitle(rebuilt);
+            documentRepository.save(doc); // direct save, no recordVersion (mechanical rename)
+            updated++;
+        }
+        log.info("Title backfill complete: scanned={} updated={} skipped={}", docs.size(), updated, skipped);
+        return updated;
+    }
+
     @Transactional
     public int backfillFileHashes() {
         List<Document> docs = documentRepository.findByFileHashIsNullAndFilePathIsNotNull();

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

@@ -0,0 +1,101 @@
+package org.raddatz.familienarchiv.document;
+
+import java.time.LocalDate;
+import java.time.format.DateTimeFormatter;
+import java.util.LinkedHashSet;
+import java.util.Locale;
+import java.util.Set;
+import java.util.regex.Pattern;
+import java.util.stream.Collectors;
+
+/**
+ * Heuristic overwrite test for the one-time title backfill (#726, FR-004): decides whether a
+ * STORED title is a machine-generated auto-title (and so may be rebuilt from the row's current
+ * state) versus hand-written prose (left untouched). Used ONLY by the backfill — save-time
+ * regeneration uses an exact old-vs-new comparison instead, with no heuristic.
+ *
+ * <p>A stored title is overwritable iff, after stripping the literal {@code index} prefix:
+ * <ol>
+ *   <li>it is exactly {@code {index}}, or</li>
+ *   <li>{@code {index} – {dateLabel}} with an optional trailing {@code  – {location}} segment
+ *       (any location — a present, valid date label is itself strong evidence of a machine
+ *       title), or</li>
+ *   <li>{@code {index} – {location}} where the segment equals the document's current location
+ *       (no date label, so the segment must match the known location to be distinguished from
+ *       prose).</li>
+ * </ol>
+ *
+ * <p>Security: the {@code index} is compared <em>literally</em> via {@link String#startsWith}
+ * (never compiled into a regex) because {@code originalFilename} is user-controlled and may carry
+ * regex metacharacters — an unquoted pattern would be a ReDoS / regex-injection vector
+ * (CWE-1333 / CWE-625). The date-label sub-patterns use only bounded, non-nested quantifiers over
+ * short tokens, so there is no catastrophic backtracking. Fail-closed: any null/blank index or
+ * structural surprise returns {@code false}.
+ */
+final class DocumentTitleBackfillMatcher {
+
+    private static final String SEPARATOR = " – ";
+
+    // German month tokens derived from the SAME Locale.GERMAN formatters DocumentTitleFormatter
+    // uses, so the matcher's accepted spellings cannot drift from what the factory emits (full
+    // names "Januar"…"Dezember"; abbreviations "Jan."…"Dez." — note May/June/July/März carry no
+    // period). Pattern.quote each so a "." in an abbreviation is literal, never a wildcard.
+    private static final String FULL_MONTH = monthAlternation("MMMM");
+    private static final String ABBR_MONTH = monthAlternation("MMM");
+    private static final String SEASON = "(?:Frühling|Sommer|Herbst|Winter)";
+    private static final String YEAR = "\\d{1,4}";
+    private static final String DAY_NUM = "\\d{1,2}";
+
+    // One complete date label, anchored, optionally followed by a free-form trailing location
+    // segment. Only bounded/non-nested quantifiers over short tokens plus a single trailing
+    // ".+" → linear, no catastrophic backtracking (FR-004 ReDoS guard).
+    private static final Pattern DATE_LABEL_WITH_OPTIONAL_LOCATION = Pattern.compile(
+            "^(?:" + String.join("|",
+                    YEAR,                                                                    // 1916
+                    "ca\\. " + YEAR,                                                         // ca. 1920
+                    FULL_MONTH + " " + YEAR,                                                 // Juni 1916
+                    DAY_NUM + "\\. " + FULL_MONTH + " " + YEAR,                              // 24. Dezember 1943
+                    SEASON + " " + YEAR,                                                     // Sommer 1916
+                    "Datum unbekannt",
+                    DAY_NUM + "\\.–" + DAY_NUM + "\\. " + ABBR_MONTH + " " + YEAR,           // 10.–11. Jan. 1917
+                    DAY_NUM + "\\. " + ABBR_MONTH + " – " + DAY_NUM + "\\. " + ABBR_MONTH + " " + YEAR,        // 30. Jan. – 2. Feb. 1917
+                    DAY_NUM + "\\. " + ABBR_MONTH + " " + YEAR + " – " + DAY_NUM + "\\. " + ABBR_MONTH + " " + YEAR, // 30. Dez. 1916 – 2. Jan. 1917
+                    DAY_NUM + "\\. " + ABBR_MONTH + " " + YEAR,                              // 10. Jan. 1917 (range end == start)
+                    "ab " + DAY_NUM + "\\. " + ABBR_MONTH + " " + YEAR)                      // ab 10. Jan. 1917
+                    + ")(?: – .+)?$");
+
+    private DocumentTitleBackfillMatcher() {
+    }
+
+    static boolean isOverwritable(String title, String index, String location) {
+        if (title == null || index == null || index.isBlank()) {
+            return false; // fail closed
+        }
+        if (!title.startsWith(index)) {
+            return false; // index is matched LITERALLY, never as a regex
+        }
+        String tail = title.substring(index.length());
+        if (tail.isEmpty()) {
+            return true; // exactly {index}
+        }
+        if (!tail.startsWith(SEPARATOR)) {
+            return false;
+        }
+        String body = tail.substring(SEPARATOR.length());
+        if (DATE_LABEL_WITH_OPTIONAL_LOCATION.matcher(body).matches()) {
+            return true; // {dateLabel} (+ optional trailing location)
+        }
+        // No date label: the lone segment must equal the document's current location to be
+        // distinguished from hand-written prose.
+        return location != null && !location.isBlank() && body.equals(location);
+    }
+
+    private static String monthAlternation(String pattern) {
+        DateTimeFormatter formatter = DateTimeFormatter.ofPattern(pattern, Locale.GERMAN);
+        Set<String> tokens = new LinkedHashSet<>();
+        for (int month = 1; month <= 12; month++) {
+            tokens.add(formatter.format(LocalDate.of(2000, month, 15)));
+        }
+        return tokens.stream().map(Pattern::quote).collect(Collectors.joining("|", "(?:", ")"));
+    }
+}

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

@@ -0,0 +1,39 @@
+package org.raddatz.familienarchiv.document;
+
+import org.springframework.stereotype.Component;
+
+/**
+ * Single source of truth for the auto-generated document title
+ * {@code {index} – {dateLabel} – {location}}.
+ *
+ * <p>The {@code document} package owns this formula; {@code importing} consumes it
+ * (see ADR for issue #726). The leading {@code index} is the document's
+ * {@code originalFilename}; the date label is the honest German label produced by
+ * {@link DocumentTitleFormatter} (the Java half of the #666 date-label split); the
+ * trailing location is the {@code meta_location} verbatim, omitted when blank.
+ */
+@Component
+public class DocumentTitleFactory {
+
+    static final String SEPARATOR = " – ";
+
+    /**
+     * Composes the auto-title from the document's current state. The date segment is
+     * dropped for UNKNOWN precision or a null date (the honest "no date" case); the
+     * location segment is dropped when blank.
+     */
+    public String build(Document doc) {
+        // originalFilename is NOT NULL in production; guard only so a synthetic/partial entity
+        // never trips StringBuilder(null) with an opaque NPE.
+        StringBuilder title = new StringBuilder(doc.getOriginalFilename() == null ? "" : doc.getOriginalFilename());
+        if (doc.getDocumentDate() != null && doc.getMetaDatePrecision() != DatePrecision.UNKNOWN) {
+            title.append(SEPARATOR).append(DocumentTitleFormatter.formatTitleDate(
+                    doc.getDocumentDate(), doc.getMetaDatePrecision(),
+                    doc.getMetaDateEnd(), doc.getMetaDateRaw()));
+        }
+        if (doc.getLocation() != null && !doc.getLocation().isBlank()) {
+            title.append(SEPARATOR).append(doc.getLocation());
+        }
+        return title.toString();
+    }
+}

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

@@ -1,6 +1,4 @@
-package org.raddatz.familienarchiv.importing;
-
-import org.raddatz.familienarchiv.document.DatePrecision;
+package org.raddatz.familienarchiv.document;
 
 import java.time.LocalDate;
 import java.time.format.DateTimeFormatter;

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

@@ -0,0 +1,40 @@
+package org.raddatz.familienarchiv.document;
+
+import org.raddatz.familienarchiv.tag.TagOperator;
+
+import java.time.LocalDate;
+import java.util.List;
+import java.util.UUID;
+
+/**
+ * The filter predicates honoured by {@link DocumentService#searchDocuments} and
+ * {@link DocumentService#findIdsForFilter}. Sort, direction, and pagination are
+ * deliberately excluded — they are not filter predicates, and {@code findIdsForFilter}
+ * needs none of them; they are passed as separate arguments instead.
+ *
+ * Kept as a record so the ten values are passed as one named bundle instead of a
+ * positional argument list where two UUIDs (sender vs. receiver) or two dates
+ * (from vs. to) can be swapped by accident at the call site — a transposition that
+ * compiles cleanly and silently returns the wrong rows.
+ *
+ * Sibling of {@link DensityFilters} (= these fields minus from/to/undated); kept
+ * separate on purpose, so the density call path never reasons about date/undated
+ * fields it deliberately excludes.
+ */
+public record SearchFilters(
+        String text,
+        LocalDate from,
+        LocalDate to,
+        UUID sender,
+        UUID receiver,
+        List<String> tags,
+        String tagQ,
+        DocumentStatus status,
+        TagOperator tagOperator,
+        boolean undated) {
+
+    /** Returns a copy with {@code undated} overridden — used by the undated-count path. */
+    public SearchFilters withUndated(boolean undated) {
+        return new SearchFilters(text, from, to, sender, receiver, tags, tagQ, status, tagOperator, undated);
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/document/transcription/TranscriptionBlockQueryService.java

@@ -17,6 +17,10 @@ public class TranscriptionBlockQueryService {
 
     private final TranscriptionBlockRepository blockRepository;
 
+    public boolean hasBlocks(UUID documentId) {
+        return blockRepository.existsByDocumentId(documentId);
+    }
+
     public Map<UUID, Integer> getCompletionStats(List<UUID> documentIds) {
         if (documentIds.isEmpty()) return Map.of();
         Map<UUID, Integer> result = new HashMap<>();

backend/src/main/java/org/raddatz/familienarchiv/document/transcription/TranscriptionBlockRepository.java

@@ -43,6 +43,8 @@ public interface TranscriptionBlockRepository extends JpaRepository<Transcriptio
 
     int countByDocumentId(UUID documentId);
 
+    boolean existsByDocumentId(UUID documentId);
+
     @Query("""
             SELECT b FROM TranscriptionBlock b
             JOIN DocumentAnnotation a ON a.id = b.annotationId

backend/src/main/java/org/raddatz/familienarchiv/exception/DomainException.java

@@ -78,4 +78,8 @@ public class DomainException extends RuntimeException {
     public static DomainException tooManyRequests(ErrorCode code, String message, long retryAfterSeconds) {
         return new DomainException(code, HttpStatus.TOO_MANY_REQUESTS, message, retryAfterSeconds);
     }
+
+    public static DomainException serviceUnavailable(ErrorCode code, String message) {
+        return new DomainException(code, HttpStatus.SERVICE_UNAVAILABLE, message);
+    }
 }

backend/src/main/java/org/raddatz/familienarchiv/exception/ErrorCode.java

@@ -26,6 +26,8 @@ public enum ErrorCode {
     FILE_UPLOAD_FAILED,
     /** The uploaded file's content type is not supported (PDF/JPEG/PNG/TIFF only). 400 */
     UNSUPPORTED_FILE_TYPE,
+    /** A RANGE date is invalid: meta_date_end is before meta_date, or an end date is set without RANGE precision. 400 */
+    INVALID_DATE_RANGE,
 
     // --- Users ---
     /** A user with the given ID or username does not exist. 404 */
@@ -133,6 +135,12 @@ public enum ErrorCode {
     /** The merge target is a descendant of the source tag. 400 */
     TAG_MERGE_INVALID_TARGET,
 
+    // --- NL Search ---
+    /** Ollama is unreachable or timed out. 503 */
+    SMART_SEARCH_UNAVAILABLE,
+    /** NL search rate limit exceeded (5 requests per user per minute). 429 */
+    SMART_SEARCH_RATE_LIMITED,
+
     // --- Generic ---
     /** Request validation failed (missing or malformed fields). 400 */
     VALIDATION_ERROR,

backend/src/main/java/org/raddatz/familienarchiv/exception/GlobalExceptionHandler.java

@@ -6,6 +6,7 @@ import io.sentry.Sentry;
 import jakarta.validation.ConstraintViolationException;
 import org.raddatz.familienarchiv.exception.DomainException;
 import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.springframework.dao.DataIntegrityViolationException;
 import org.springframework.http.ResponseEntity;
 import org.springframework.http.converter.HttpMessageNotReadableException;
 import org.springframework.web.bind.MethodArgumentNotValidException;
@@ -64,6 +65,38 @@ public class GlobalExceptionHandler {
                 .body(new ErrorResponse(ErrorCode.VALIDATION_ERROR, ex.getReason()));
     }
 
+    /**
+     * Backstop for any database integrity violation that slips past the explicit upstream
+     * guards (e.g. a future constraint, or the import path emitting a bad range). Turns it into
+     * a clean 400 instead of a 500 + Sentry alert. The known date-range cases are caught upstream
+     * and never reach here; this only catches the unanticipated ones — so it logs the constraint
+     * NAME at WARN to stay debuggable, without re-leaking SQL and without branching the response
+     * on it (the response stays generic, which is the non-brittle part).
+     */
+    @ExceptionHandler(DataIntegrityViolationException.class)
+    public ResponseEntity<ErrorResponse> handleDataIntegrityViolation(DataIntegrityViolationException ex) {
+        // Log the constraint NAME only — schema metadata, safe for Loki, and enough to tell which
+        // constraint fired at 2am. Never pass `ex` / `ex.getMessage()`: those embed the SQL + the
+        // offending values (CWE-209). No Sentry: an integrity violation is a 400, not a system fault.
+        log.warn("Rejected a request that violated a database integrity constraint: {}", constraintNameOf(ex));
+        return ResponseEntity.badRequest()
+                .body(new ErrorResponse(ErrorCode.VALIDATION_ERROR, "The submitted data violated a database constraint"));
+    }
+
+    /**
+     * Returns the offending constraint's name from the cause chain, or {@code "unknown"}.
+     * Reads only the name (a non-sensitive schema identifier) — never the SQL or the values.
+     */
+    private static String constraintNameOf(Throwable ex) {
+        for (Throwable t = ex; t != null && t != t.getCause(); t = t.getCause()) {
+            if (t instanceof org.hibernate.exception.ConstraintViolationException cve
+                    && cve.getConstraintName() != null) {
+                return cve.getConstraintName();
+            }
+        }
+        return "unknown";
+    }
+
     @ExceptionHandler(Exception.class)
     public ResponseEntity<ErrorResponse> handleGeneric(Exception ex) {
         Sentry.captureException(ex);

backend/src/main/java/org/raddatz/familienarchiv/importing/CanonicalImportOrchestrator.java

@@ -4,13 +4,21 @@ import lombok.RequiredArgsConstructor;
 import lombok.extern.slf4j.Slf4j;
 import org.raddatz.familienarchiv.exception.DomainException;
 import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.raddatz.familienarchiv.person.relationship.RelationType;
+import org.raddatz.familienarchiv.person.relationship.RelationshipService;
+import org.raddatz.familienarchiv.person.relationship.dto.NetworkDTO;
+import org.raddatz.familienarchiv.person.relationship.dto.PersonNodeDTO;
+import org.raddatz.familienarchiv.person.relationship.dto.RelationshipDTO;
 import org.springframework.beans.factory.annotation.Value;
 import org.springframework.scheduling.annotation.Async;
 import org.springframework.stereotype.Service;
 
 import java.io.File;
 import java.time.LocalDateTime;
+import java.util.HashMap;
 import java.util.List;
+import java.util.Map;
+import java.util.UUID;
 
 /**
  * Runs the four canonical loaders in their real dependency order — encoded explicitly
@@ -34,6 +42,7 @@ public class CanonicalImportOrchestrator {
     private final PersonRegisterImporter personRegisterImporter;
     private final PersonTreeImporter personTreeImporter;
     private final DocumentImporter documentImporter;
+    private final RelationshipService relationshipService;
 
     @Value("${app.import.dir:/import}")
     private String canonicalDir;
@@ -67,6 +76,7 @@ public class CanonicalImportOrchestrator {
             tagTreeImporter.load(tagTree);
             personRegisterImporter.load(persons);
             personTreeImporter.load(personsTree);
+            warnOnGenerationMonotonicityViolations();
             DocumentImporter.LoadResult result = documentImporter.load(documents);
 
             currentStatus = new ImportStatus(ImportStatus.State.DONE, "IMPORT_DONE",
@@ -91,4 +101,31 @@ public class CanonicalImportOrchestrator {
         }
         return artifact;
     }
+
+    /**
+     * Walks every PARENT_OF edge in the family graph and logs a WARN whenever a child's
+     * generation is not strictly deeper than its parent's. Soft check only — the import
+     * is never aborted; the warning is a forensic signal for the curator. Reads through
+     * {@link RelationshipService} so the orchestrator stays within the layering rule
+     * (no direct repository access).
+     */
+    private void warnOnGenerationMonotonicityViolations() {
+        NetworkDTO network = relationshipService.getFamilyNetwork();
+        Map<UUID, PersonNodeDTO> byId = new HashMap<>(network.nodes().size());
+        for (PersonNodeDTO node : network.nodes()) {
+            byId.put(node.id(), node);
+        }
+        for (RelationshipDTO edge : network.edges()) {
+            if (edge.relationType() != RelationType.PARENT_OF) continue;
+            PersonNodeDTO parent = byId.get(edge.personId());
+            PersonNodeDTO child = byId.get(edge.relatedPersonId());
+            if (parent == null || child == null) continue;
+            Integer pg = parent.generation();
+            Integer cg = child.generation();
+            if (pg != null && cg != null && cg <= pg) {
+                log.warn("Generation monotonicity violation: parent {} (G{}) -> child {} (G{})",
+                        parent.displayName(), pg, child.displayName(), cg);
+            }
+        }
+    }
 }

backend/src/main/java/org/raddatz/familienarchiv/importing/DocumentImporter.java

@@ -5,6 +5,7 @@ import lombok.extern.slf4j.Slf4j;
 import org.raddatz.familienarchiv.document.DatePrecision;
 import org.raddatz.familienarchiv.document.Document;
 import org.raddatz.familienarchiv.document.DocumentService;
+import org.raddatz.familienarchiv.document.DocumentTitleFactory;
 import org.raddatz.familienarchiv.document.DocumentStatus;
 import org.raddatz.familienarchiv.document.ThumbnailAsyncRunner;
 import org.raddatz.familienarchiv.exception.DomainException;
@@ -24,7 +25,6 @@ import software.amazon.awssdk.services.s3.model.PutObjectRequest;
 import org.raddatz.familienarchiv.tag.TagService;
 
 import java.io.File;
-import java.io.FileInputStream;
 import java.io.IOException;
 import java.io.InputStream;
 import java.nio.file.Files;
@@ -75,10 +75,12 @@ public class DocumentImporter {
             Pattern.compile("[A-Za-z\\u00C0-\\u00D6\\u00D8-\\u00F6\\u00F8-\\u00FF]{1,4}-+\\d+x?");
 
     private final DocumentService documentService;
+    private final DocumentTitleFactory documentTitleFactory;
     private final PersonService personService;
     private final TagService tagService;
     private final S3Client s3Client;
     private final ThumbnailAsyncRunner thumbnailAsyncRunner;
+    private final FileStreamOpener fileStreamOpener;
 
     @Value("${app.s3.bucket:familienarchiv}")
     private String bucketName;
@@ -178,53 +180,62 @@ public class DocumentImporter {
                                    String s3Key, String contentType, DocumentStatus status) {
         Document doc = existing != null ? existing
                 : Document.builder().originalFilename(index).build();
-
-        String senderName = row.get("sender_name");
-        String receiverNames = row.get("receiver_names");
-        Person sender = resolveSender(row.get("sender_person_id"), senderName);
-        Set<Person> receivers = resolveReceivers(row.get("receiver_person_ids"));
-
-        LocalDate date = parseIsoDate(row.get("date_iso"));
-        DatePrecision precision = parsePrecision(row.get("date_precision"));
-        LocalDate dateEnd = parseIsoDate(row.get("date_end"));
-        String dateRaw = blankToNull(row.get("date_raw"));
-        String location = blankToNull(row.get("location"));
-
-        doc.setTitle(buildTitle(index, date, precision, dateEnd, dateRaw, location));
-        doc.setStatus(status);
-        doc.setFilePath(s3Key);
-        doc.setContentType(contentType);
-        doc.setSender(sender);
-        doc.setSenderText(blankToNull(senderName));
-        // The canonical row is authoritative for receivers/tags (ADR-025): clear then
-        // re-populate so a shrunk set on re-import prunes stale links rather than
-        // accumulating them. The raw sender_text/receiver_text retention is separate.
-        doc.getReceivers().clear();
-        doc.getReceivers().addAll(receivers);
-        doc.setReceiverText(blankToNull(receiverNames));
-        doc.setDocumentDate(date);
-        doc.setMetaDatePrecision(precision);
-        doc.setMetaDateEnd(dateEnd);
-        doc.setMetaDateRaw(dateRaw);
-        doc.setLocation(location);
-        doc.setSummary(blankToNull(row.get("summary")));
-        attachTag(doc, row.get("tags"));
-        doc.setMetadataComplete(doc.getDocumentDate() != null || sender != null || !receivers.isEmpty());
+        applyAttribution(doc, row);
+        applyDates(doc, row);
+        applyAuthoritativeAssociations(doc, row);
+        applyFileMetadata(doc, s3Key, contentType, status);
+        applyComputedFlags(doc);
         return doc;
     }
 
-    // The title carries the date at the HONEST precision (never a fabricated day) via the
-    // shared DocumentTitleFormatter, plus the location — kept under 20 lines by delegating.
-    private static String buildTitle(String index, LocalDate date, DatePrecision precision,
-                                     LocalDate end, String raw, String location) {
-        StringBuilder title = new StringBuilder(index);
-        if (date != null && precision != DatePrecision.UNKNOWN) {
-            title.append(" – ").append(DocumentTitleFormatter.formatTitleDate(date, precision, end, raw));
-        }
-        if (location != null && !location.isBlank()) {
-            title.append(" – ").append(location);
-        }
-        return title.toString();
+    // Sender + raw sender/receiver text. The raw cells are always retained verbatim, even
+    // when a person is linked — the load-bearing invariant behind the merge story (ADR-025).
+    private void applyAttribution(Document doc, CanonicalSheetReader.Row row) {
+        String senderName = row.get("sender_name");
+        String receiverNames = row.get("receiver_names");
+        Person sender = resolveSender(row.get("sender_person_id"), senderName);
+        doc.setSender(sender);
+        doc.setSenderText(blankToNull(senderName));
+        doc.setReceiverText(blankToNull(receiverNames));
+    }
+
+    // Date triplet + raw + location. Pure value parsing, no semantic logic.
+    private void applyDates(Document doc, CanonicalSheetReader.Row row) {
+        doc.setDocumentDate(parseIsoDate(row.get("date_iso")));
+        doc.setMetaDatePrecision(parsePrecision(row.get("date_precision")));
+        doc.setMetaDateEnd(parseIsoDate(row.get("date_end")));
+        doc.setMetaDateRaw(blankToNull(row.get("date_raw")));
+        doc.setLocation(blankToNull(row.get("location")));
+        doc.setSummary(blankToNull(row.get("summary")));
+    }
+
+    // Receivers and tags are owned by the canonical row (ADR-025): clear then re-populate so a
+    // shrunk set on re-import prunes stale links rather than accumulating them. The
+    // "preserve human edits" rule does NOT extend to these collections.
+    private void applyAuthoritativeAssociations(Document doc, CanonicalSheetReader.Row row) {
+        Set<Person> receivers = resolveReceivers(row.get("receiver_person_ids"), row.get("receiver_names"));
+        doc.getReceivers().clear();
+        doc.getReceivers().addAll(receivers);
+        attachTag(doc, row.get("tags"));
+    }
+
+    // S3 key, content type, status, and the index-derived title. The title formula lives in
+    // the document package's DocumentTitleFactory (single source of truth, #726); by this point
+    // applyDates has populated the date/location and originalFilename carries the index.
+    private void applyFileMetadata(Document doc, String s3Key, String contentType,
+                                   DocumentStatus status) {
+        doc.setStatus(status);
+        doc.setFilePath(s3Key);
+        doc.setContentType(contentType);
+        doc.setTitle(documentTitleFactory.build(doc));
+    }
+
+    // metadataComplete: a document counts as fully described if any of the three "who/when"
+    // pieces is filled. Called last so the upstream setters have already populated the doc.
+    private void applyComputedFlags(Document doc) {
+        doc.setMetadataComplete(doc.getDocumentDate() != null
+                || doc.getSender() != null
+                || !doc.getReceivers().isEmpty());
     }
 
     // ─── attribution routing — register-first, always retain raw ─────────────────────
@@ -234,10 +245,18 @@ public class DocumentImporter {
         return resolvePerson(slug, rawName);
     }
 
-    private Set<Person> resolveReceivers(String slugs) {
+    // Zips the parallel `receiver_person_ids` and `receiver_names` columns by position so an
+    // unresolved receiver becomes a provisional Person whose lastName is the human name from
+    // `receiver_names`, not the slug. If the names list is shorter than the slugs list (rare —
+    // canonical data zips them 1:1), missing entries fall back to slug-as-name.
+    private Set<Person> resolveReceivers(String slugs, String names) {
+        List<String> slugList = CanonicalSheetReader.splitList(slugs);
+        List<String> nameList = CanonicalSheetReader.splitList(names);
         Set<Person> receivers = new LinkedHashSet<>();
-        for (String slug : CanonicalSheetReader.splitList(slugs)) {
-            receivers.add(resolvePerson(slug, slug));
+        for (int i = 0; i < slugList.size(); i++) {
+            String slug = slugList.get(i);
+            String name = i < nameList.size() ? nameList.get(i) : slug;
+            receivers.add(resolvePerson(slug, name));
         }
         return receivers;
     }
@@ -319,13 +338,10 @@ public class DocumentImporter {
         return INDEX_PATTERN.matcher(index).matches();
     }
 
-    // package-private: a Mockito spy in tests can override to inject IOException
-    InputStream openFileStream(File file) throws IOException {
-        return new FileInputStream(file);
-    }
-
     private boolean isPdfMagicBytes(File file) throws IOException {
-        try (InputStream is = openFileStream(file)) {
+        // FileStreamOpener is injected so tests can stub a throwing implementation for the
+        // IO-error branch without spying on the importer itself.
+        try (InputStream is = fileStreamOpener.open(file)) {
             byte[] header = is.readNBytes(4);
             return header.length == 4
                     && header[0] == 0x25  // %

backend/src/main/java/org/raddatz/familienarchiv/importing/FileStreamOpener.java

@@ -0,0 +1,33 @@
+package org.raddatz.familienarchiv.importing;
+
+import org.springframework.stereotype.Component;
+
+import java.io.File;
+import java.io.FileInputStream;
+import java.io.IOException;
+import java.io.InputStream;
+
+/**
+ * Test seam for opening a {@link File} as an {@link InputStream}. Extracted so the magic-byte
+ * check in {@link DocumentImporter} can be unit-tested for the IO-error branch by injecting a
+ * mock that throws, without needing a Mockito spy on the importer itself.
+ *
+ * <p>Production uses {@link DefaultFileStreamOpener}, a one-line delegate to
+ * {@code new FileInputStream(file)}.
+ */
+@FunctionalInterface
+public interface FileStreamOpener {
+
+    /** Opens {@code file} for sequential reads. Caller closes the returned stream. */
+    InputStream open(File file) throws IOException;
+
+    /** Default production implementation: plain {@code FileInputStream}. */
+    @Component
+    final class DefaultFileStreamOpener implements FileStreamOpener {
+
+        @Override
+        public InputStream open(File file) throws IOException {
+            return new FileInputStream(file);
+        }
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/importing/PersonRegisterImporter.java

@@ -2,6 +2,7 @@ package org.raddatz.familienarchiv.importing;
 
 import lombok.RequiredArgsConstructor;
 import lombok.extern.slf4j.Slf4j;
+import org.raddatz.familienarchiv.person.PersonGeneration;
 import org.raddatz.familienarchiv.person.PersonService;
 import org.raddatz.familienarchiv.person.PersonType;
 import org.raddatz.familienarchiv.person.PersonUpsertCommand;
@@ -11,6 +12,8 @@ import java.io.File;
 import java.time.LocalDate;
 import java.time.format.DateTimeParseException;
 import java.util.List;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
 
 /**
  * Loads {@code canonical-persons.xlsx} (the register) into the person domain via
@@ -25,6 +28,13 @@ public class PersonRegisterImporter {
 
     static final List<String> REQUIRED_HEADERS = List.of("person_id", "last_name", "first_name", "provisional");
 
+    // Matches a leading optional G then a signed integer. Anchored at the
+    // start so noise can't slip in before the number, but tolerant of trailing
+    // commentary cells (e.g. "G 2  de Gruyter") since curated rows sometimes
+    // carry an inline note. Out-of-range values are caught by the post-parse
+    // range guard, not by the regex.
+    private static final Pattern GENERATION_PATTERN = Pattern.compile("^\\s*G?\\s*(-?\\d+)");
+
     private final PersonService personService;
 
     public int load(File artifact) {
@@ -49,11 +59,31 @@ public class PersonRegisterImporter {
                 .notes(blankToNull(row.get("notes")))
                 .birthYear(yearOf(row.get("birth_date")))
                 .deathYear(yearOf(row.get("death_date")))
+                .generation(parseGeneration(row.get("generation"), personId))
                 .personType(PersonType.PERSON)
                 .provisional(Boolean.parseBoolean(row.get("provisional")))
                 .build();
     }
 
+    /**
+     * Parses an optional {@code G n} generation cell. Returns null for blanks,
+     * non-matching strings, and any value outside the {@link PersonGeneration}
+     * bounds (mirroring the V70 CHECK). Out-of-range values log a WARN but
+     * never abort the batch — REQ-IMP-001.
+     */
+    static Integer parseGeneration(String raw, String personId) {
+        if (raw == null || raw.isBlank()) return null;
+        Matcher m = GENERATION_PATTERN.matcher(raw);
+        if (!m.find()) return null;
+        int parsed = Integer.parseInt(m.group(1));
+        if (parsed < PersonGeneration.MIN_GENERATION || parsed > PersonGeneration.MAX_GENERATION) {
+            log.warn("Skipping out-of-range generation '{}' for row {}", raw, personId);
+            return null;
+        }
+        log.debug("Parsed generation '{}' for person {}", raw, personId);
+        return parsed;
+    }
+
     private static Integer yearOf(String isoDate) {
         if (isoDate == null || isoDate.isBlank()) return null;
         try {

backend/src/main/java/org/raddatz/familienarchiv/importing/PersonTreeImporter.java

@@ -7,6 +7,7 @@ import lombok.extern.slf4j.Slf4j;
 import org.raddatz.familienarchiv.exception.DomainException;
 import org.raddatz.familienarchiv.exception.ErrorCode;
 import org.raddatz.familienarchiv.person.Person;
+import org.raddatz.familienarchiv.person.PersonGeneration;
 import org.raddatz.familienarchiv.person.PersonService;
 import org.raddatz.familienarchiv.person.PersonType;
 import org.raddatz.familienarchiv.person.PersonUpsertCommand;
@@ -79,12 +80,29 @@ public class PersonTreeImporter {
                 .notes(blankToNull(text(node, "notes")))
                 .birthYear(intOrNull(node, "birthYear"))
                 .deathYear(intOrNull(node, "deathYear"))
+                .generation(generationOrNull(node, personId))
                 .familyMember(node.path("familyMember").asBoolean(false))
                 .personType(PersonType.PERSON)
                 .provisional(false)
                 .build();
     }
 
+    /**
+     * Returns the JSON {@code generation} value if present and within the
+     * {@link PersonGeneration} bounds; null otherwise. Out-of-range values
+     * log a WARN but never abort the batch — mirrors the register-importer
+     * skip-and-warn policy.
+     */
+    private static Integer generationOrNull(JsonNode node, String personId) {
+        Integer raw = intOrNull(node, "generation");
+        if (raw == null) return null;
+        if (raw < PersonGeneration.MIN_GENERATION || raw > PersonGeneration.MAX_GENERATION) {
+            log.warn("Skipping out-of-range generation '{}' for person {}", raw, personId);
+            return null;
+        }
+        return raw;
+    }
+
     private int createRelationships(JsonNode relationships, Map<String, UUID> idByRowId) {
         int created = 0;
         for (JsonNode node : relationships) {

backend/src/main/java/org/raddatz/familienarchiv/person/Person.java

@@ -52,6 +52,13 @@ public class Person {
     private Integer birthYear;
     private Integer deathYear;
 
+    // Hand-curated generation index from canonical-persons.xlsx (G 0 = oldest).
+    // Nullable for persons outside the curated family graph. Drives the
+    // Stammbaum strict-rank seed (see #689) and re-import preserves human
+    // edits via PersonService.preferHuman (ADR-025).
+    @Column(name = "generation")
+    private Integer generation;
+
     @Column(name = "family_member", nullable = false)
     @Builder.Default
     @Schema(requiredMode = Schema.RequiredMode.REQUIRED)

backend/src/main/java/org/raddatz/familienarchiv/person/PersonGeneration.java

@@ -0,0 +1,16 @@
+package org.raddatz.familienarchiv.person;
+
+/**
+ * Single source of truth for the {@code persons.generation} value range.
+ * The DB CHECK in V70, the {@code PersonUpdateDTO} Bean Validation annotations,
+ * and the canonical importers all reference these constants so a future widening
+ * (e.g. accepting {@code G −1} ancestors) happens in one place. Mirror this file
+ * by hand in the V70 migration comment when adjusting bounds.
+ */
+public final class PersonGeneration {
+
+    public static final int MIN_GENERATION = 0;
+    public static final int MAX_GENERATION = 10;
+
+    private PersonGeneration() {}
+}

backend/src/main/java/org/raddatz/familienarchiv/person/PersonRepository.java

@@ -29,14 +29,36 @@ public interface PersonRepository extends JpaRepository<Person, UUID> {
     // Stammbaum-Knoten: alle Personen mit family_member = true.
     List<Person> findByFamilyMemberTrueOrderByLastNameAscFirstNameAsc();
 
-    // Lookup by full alias string, used during ODS mass import
-    Optional<Person> findByAliasIgnoreCase(String alias);
+    // Exact-case alias lookup — the first resolution step in findOrCreateByAlias.
+    // Case-colliding aliases across persons (müller / Müller) are valid human labels, NOT
+    // duplicates: source_ref is the stable identity (ADR-025/033), alias is editable. Do NOT
+    // add a unique(lower(alias)) constraint — see ADR-033.
+    Optional<Person> findByAlias(String alias);
+
+    // Plural case-insensitive alias lookup — the fallback step. Returns ALL case-folding
+    // siblings so the service can pick a deterministic one (lowest id) instead of letting a
+    // derived Optional<…>IgnoreCase throw NonUniqueResultException. See ADR-033.
+    List<Person> findAllByAliasIgnoreCase(String alias);
 
     // Lookup by the normalizer person_id, used for idempotent canonical re-import (Phase 3).
     Optional<Person> findBySourceRef(String sourceRef);
 
-    // Exact first+last name match, used for filename-based sender lookup
-    Optional<Person> findByFirstNameIgnoreCaseAndLastNameIgnoreCase(String firstName, String lastName);
+    // Exact-case first+last name match — the first step of filename-based sender resolution.
+    // Explicit `=` (HQL, not a derived query) so a null firstName binds as `first_name = NULL`
+    // — never a match — instead of the derived-query fold to `first_name IS NULL`, which would
+    // pull a last-name-only row in as a sender (a provenance defect). See ADR-033.
+    @Query("SELECT p FROM Person p WHERE p.firstName = :firstName AND p.lastName = :lastName")
+    Optional<Person> findByFirstNameAndLastName(@Param("firstName") String firstName,
+                                                @Param("lastName") String lastName);
+
+    // Plural case-insensitive first+last name match — lets findByName bail to empty on 2+ matches
+    // instead of letting a derived Optional<…>IgnoreCase throw NonUniqueResultException. Same
+    // null fail-closed guarantee as above: LOWER(:firstName) is NULL for a null arg, so a null
+    // first name resolves to no match (not first_name IS NULL widening). See ADR-033.
+    @Query("SELECT p FROM Person p WHERE LOWER(p.firstName) = LOWER(:firstName) "
+         + "AND LOWER(p.lastName) = LOWER(:lastName)")
+    List<Person> findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase(@Param("firstName") String firstName,
+                                                                   @Param("lastName") String lastName);
 
     // --- PersonSummaryDTO with document count ---
 
@@ -189,18 +211,15 @@ public interface PersonRepository extends JpaRepository<Person, UUID> {
     List<Person> findCorrespondentsWithFilter(@Param("personId") UUID personId, @Param("q") String q);
 
     // --- Merge helpers (native SQL to bypass JPA entity layer) ---
+    // clearAutomatically + flushAutomatically keep the L1 cache from desyncing: these bulk
+    // updates run beneath Hibernate, and mergePersons follows them with a deleteById whose
+    // ON DELETE CASCADE (V71) also fires beneath the session.
 
-    @Modifying
+    @Modifying(clearAutomatically = true, flushAutomatically = true)
     @Query(value = "UPDATE documents SET sender_id = :target WHERE sender_id = :source", nativeQuery = true)
     void reassignSender(@Param("source") UUID source, @Param("target") UUID target);
 
-    // Used by deletePerson: detach a deleted person from documents they sent, so the hard
-    // delete cannot orphan a documents.sender_id FK (the column is nullable).
-    @Modifying
-    @Query(value = "UPDATE documents SET sender_id = NULL WHERE sender_id = :source", nativeQuery = true)
-    void reassignSenderToNull(@Param("source") UUID source);
-
-    @Modifying
+    @Modifying(clearAutomatically = true, flushAutomatically = true)
     @Query(value = """
             INSERT INTO document_receivers (document_id, person_id)
             SELECT document_id, :target FROM document_receivers
@@ -210,8 +229,4 @@ public interface PersonRepository extends JpaRepository<Person, UUID> {
               )
             """, nativeQuery = true)
     void insertMissingReceiverReference(@Param("source") UUID source, @Param("target") UUID target);
-
-    @Modifying
-    @Query(value = "DELETE FROM document_receivers WHERE person_id = :source", nativeQuery = true)
-    void deleteReceiverReferences(@Param("source") UUID source);
-}
+}

backend/src/main/java/org/raddatz/familienarchiv/person/PersonService.java

@@ -1,5 +1,6 @@
 package org.raddatz.familienarchiv.person;
 
+import java.util.Comparator;
 import java.util.List;
 import java.util.Optional;
 import java.util.UUID;
@@ -68,15 +69,13 @@ public class PersonService {
     }
 
     /**
-     * Hard-deletes a person used by triage. Detaches the person from any documents they
-     * sent (nulls sender_id) and from any received-document references first, so the delete
-     * cannot orphan an FK and fail with a 500.
+     * Hard-deletes a person used by triage. Referential integrity is enforced by the database
+     * (V71's {@code ON DELETE} constraints: sender_id {@code SET NULL}, receiver and @-mention
+     * rows {@code CASCADE}), so the service stays thin — it only verifies existence then deletes.
      */
     @Transactional
     public void deletePerson(UUID id) {
         getById(id);
-        personRepository.reassignSenderToNull(id);
-        personRepository.deleteReceiverReferences(id);
         personRepository.deleteById(id);
     }
 
@@ -100,6 +99,10 @@ public class PersonService {
         return personRepository.findAllById(ids);
     }
 
+    public List<Person> findByDisplayNameContaining(String fragment) {
+        return personRepository.searchByName(fragment);
+    }
+
     public List<Person> findAllFamilyMembers() {
         return personRepository.findByFamilyMemberTrueOrderByLastNameAscFirstNameAsc();
     }
@@ -112,7 +115,19 @@ public class PersonService {
     }
 
     public Optional<Person> findByName(String firstName, String lastName) {
-        return personRepository.findByFirstNameIgnoreCaseAndLastNameIgnoreCase(firstName, lastName);
+        // Same scope as findOrCreateByAlias (#731): a case-collision resolves without throwing;
+        // two byte-identical same-case persons are an out-of-scope data anomaly the exact
+        // Optional below would surface as the opaque INTERNAL_ERROR, not a wrong sender.
+        Optional<Person> exact = personRepository.findByFirstNameAndLastName(firstName, lastName);
+        if (exact.isPresent()) return exact;
+        List<Person> caseInsensitive =
+                personRepository.findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase(firstName, lastName);
+        // Deliberate divergence from findOrCreateByAlias: an ambiguous filename leaves the sender
+        // UNSET rather than picking the lowest id. The archive's value is correct provenance — a
+        // confidently-wrong pre-filled "Hans Müller" is worse than an empty field, because a
+        // reviewer won't re-check a pre-filled value. Do NOT "consistency-clean" this into the
+        // lowest-id fallback. See ADR-033.
+        return caseInsensitive.size() == 1 ? Optional.of(caseInsensitive.get(0)) : Optional.empty();
     }
 
     /** Lookup by the normalizer person_id — used by the canonical importer for register-first matching. */
@@ -127,32 +142,45 @@ public class PersonService {
         PersonType type = PersonTypeClassifier.classify(alias);
         if (type == PersonType.SKIP) return null;
 
-        return personRepository.findByAliasIgnoreCase(alias).orElseGet(() -> {
-            if (type == PersonType.INSTITUTION || type == PersonType.GROUP) {
-                return personRepository.save(Person.builder()
-                        .alias(alias)
-                        .lastName(alias)
-                        .personType(type)
-                        .build());
-            }
+        // Aliases differing only by case (müller / Müller) are valid distinct persons, not
+        // duplicates, so a CASE-COLLISION must not throw: exact-case first, then the lowest-id
+        // case-insensitive sibling, then create. Mirrors the tag path — see ADR-033.
+        // Scope (#731): "ambiguous" means case-insensitive. Two BYTE-IDENTICAL same-case aliases
+        // are a true data anomaly out of scope here; the exact Optional below would surface that
+        // as the opaque INTERNAL_ERROR (never a wrong row), not silently pick one.
+        Optional<Person> exact = personRepository.findByAlias(alias);
+        if (exact.isPresent()) return exact.get();              // exact-case wins
+        List<Person> caseInsensitive = personRepository.findAllByAliasIgnoreCase(alias);
+        if (!caseInsensitive.isEmpty()) {
+            return caseInsensitive.stream().min(Comparator.comparing(Person::getId)).orElseThrow(); // deterministic tie-break — list is non-empty, never throws
+        }
 
-            PersonNameParser.SplitName split = PersonNameParser.split(alias);
-            Person person = personRepository.save(Person.builder()
+        // Create-when-absent: institution/group keep the full label in lastName; a person name
+        // is split and a maiden name (geb. …) becomes a MAIDEN_NAME alias.
+        if (type == PersonType.INSTITUTION || type == PersonType.GROUP) {
+            return personRepository.save(Person.builder()
                     .alias(alias)
-                    .firstName(split.firstName())
-                    .lastName(split.lastName())
+                    .lastName(alias)
+                    .personType(type)
                     .build());
-            if (split.maidenName() != null) {
-                int nextSortOrder = aliasRepository.findMaxSortOrder(person.getId()) + 1;
-                aliasRepository.save(PersonNameAlias.builder()
-                        .person(person)
-                        .lastName(split.maidenName())
-                        .type(PersonNameAliasType.MAIDEN_NAME)
-                        .sortOrder(nextSortOrder)
-                        .build());
-            }
-            return person;
-        });
+        }
+
+        PersonNameParser.SplitName split = PersonNameParser.split(alias);
+        Person person = personRepository.save(Person.builder()
+                .alias(alias)
+                .firstName(split.firstName())
+                .lastName(split.lastName())
+                .build());
+        if (split.maidenName() != null) {
+            int nextSortOrder = aliasRepository.findMaxSortOrder(person.getId()) + 1;
+            aliasRepository.save(PersonNameAlias.builder()
+                    .person(person)
+                    .lastName(split.maidenName())
+                    .type(PersonNameAliasType.MAIDEN_NAME)
+                    .sortOrder(nextSortOrder)
+                    .build());
+        }
+        return person;
     }
 
     /**
@@ -177,6 +205,7 @@ public class PersonService {
                 .notes(blankToNull(cmd.notes()))
                 .birthYear(cmd.birthYear())
                 .deathYear(cmd.deathYear())
+                .generation(cmd.generation())
                 .familyMember(cmd.familyMember())
                 .personType(cmd.personType() == null ? PersonType.PERSON : cmd.personType())
                 .provisional(cmd.provisional())
@@ -200,6 +229,7 @@ public class PersonService {
         existing.setNotes(preferHuman(existing.getNotes(), cmd.notes()));
         existing.setBirthYear(preferHuman(existing.getBirthYear(), cmd.birthYear()));
         existing.setDeathYear(preferHuman(existing.getDeathYear(), cmd.deathYear()));
+        existing.setGeneration(preferHuman(existing.getGeneration(), cmd.generation()));
         if (cmd.personType() != null && existing.getPersonType() == PersonType.PERSON) {
             existing.setPersonType(cmd.personType());
         }
@@ -254,6 +284,7 @@ public class PersonService {
                 .notes(dto.getNotes() == null || dto.getNotes().isBlank() ? null : dto.getNotes().trim())
                 .birthYear(dto.getBirthYear())
                 .deathYear(dto.getDeathYear())
+                .generation(dto.getGeneration())
                 .build();
         return personRepository.save(person);
     }
@@ -286,9 +317,18 @@ public class PersonService {
         person.setNotes(dto.getNotes() == null || dto.getNotes().isBlank() ? null : dto.getNotes().trim());
         person.setBirthYear(dto.getBirthYear());
         person.setDeathYear(dto.getDeathYear());
+        // Form path: a human can clear generation back to null. Unlike the importer
+        // which routes through preferHuman, we write the DTO value verbatim.
+        person.setGeneration(dto.getGeneration());
         return personRepository.save(person);
     }
 
+    /**
+     * Merges the source person into the target, then deletes the source. Sender references move
+     * to the target; receiver references the target lacks are inserted. The source's leftover
+     * receiver join rows are not deleted explicitly — they cascade-drop via V71's
+     * {@code ON DELETE CASCADE} on {@code document_receivers.person_id} when the source is deleted.
+     */
     @Transactional
     public void mergePersons(UUID sourceId, UUID targetId) {
         if (sourceId.equals(targetId)) {
@@ -305,9 +345,7 @@ public class PersonService {
         // Add target as receiver where source is receiver but target is not yet
         personRepository.insertMissingReceiverReference(sourceId, targetId);
 
-        // Remove all remaining source receiver references (duplicates already handled)
-        personRepository.deleteReceiverReferences(sourceId);
-
+        // Source's remaining receiver rows cascade-drop via V71's ON DELETE CASCADE.
         personRepository.deleteById(sourceId);
     }
 

backend/src/main/java/org/raddatz/familienarchiv/person/PersonUpdateDTO.java

@@ -1,5 +1,7 @@
 package org.raddatz.familienarchiv.person;
 
+import jakarta.validation.constraints.Max;
+import jakarta.validation.constraints.Min;
 import jakarta.validation.constraints.NotNull;
 import jakarta.validation.constraints.Size;
 import lombok.Data;
@@ -21,4 +23,9 @@ public class PersonUpdateDTO {
     private String notes;
     private Integer birthYear;
     private Integer deathYear;
+    // Mirror of the persons.generation CHECK constraint (V70). Bounds live in
+    // PersonGeneration so DB, DTO, and importer all read from one place.
+    @Min(PersonGeneration.MIN_GENERATION)
+    @Max(PersonGeneration.MAX_GENERATION)
+    private Integer generation;
 }

backend/src/main/java/org/raddatz/familienarchiv/person/PersonUpsertCommand.java

@@ -18,6 +18,7 @@ public record PersonUpsertCommand(
         String notes,
         Integer birthYear,
         Integer deathYear,
+        Integer generation,
         boolean familyMember,
         PersonType personType,
         boolean provisional

backend/src/main/java/org/raddatz/familienarchiv/person/README.md

@@ -20,8 +20,8 @@ Features: person CRUD, name alias management, person merge (deduplication), fami
 | `getById(UUID)` | document, geschichte, ocr | Fetch one person by ID |
 | `getAllById(List<UUID>)` | document | Bulk fetch for sender/receiver resolution |
 | `findAll(String q)` | document, dashboard | List all persons |
-| `findByName(String firstName, String lastName)` | document | Typeahead search |
-| `findOrCreateByAlias(String rawName)` | importing | Idempotent create during mass import; type classification happens internally |
+| `findByName(String firstName, String lastName)` | document | Filename-based **sender resolution** in `storeDocument`: exact-case match → single case-insensitive match → else **empty** (ambiguous names leave the sender unset; a null first name never matches). See ADR-033. |
+| `findOrCreateByAlias(String rawName)` | importing | Idempotent create during mass import; type classification happens internally. Resolves exact-case → lowest-id case-insensitive sibling → create — never throws on case-colliding aliases. See ADR-033. |
 | `findAllFamilyMembers()` | dashboard | Family member list for stats |
 | `findCorrespondents()` | document | Correspondent list for conversation filter |
 | `count()` | dashboard | Total person count for stats |

backend/src/main/java/org/raddatz/familienarchiv/person/relationship/RelationshipInferenceService.java

@@ -96,7 +96,8 @@ public class RelationshipInferenceService {
             if (p == null) continue;
             List<RelationToken> path = shortestPaths.get(id);
             PersonNodeDTO node = new PersonNodeDTO(
-                    p.getId(), p.getDisplayName(), p.getBirthYear(), p.getDeathYear(), p.isFamilyMember());
+                    p.getId(), p.getDisplayName(), p.getBirthYear(), p.getDeathYear(),
+                    p.getGeneration(), p.isFamilyMember());
             out.add(new InferredRelationshipWithPersonDTO(node, labelFor(path), path.size()));
         }
         out.sort(Comparator.comparingInt(InferredRelationshipWithPersonDTO::hops)

backend/src/main/java/org/raddatz/familienarchiv/person/relationship/RelationshipService.java

@@ -31,6 +31,12 @@ import java.util.UUID;
 @RequiredArgsConstructor
 public class RelationshipService {
 
+    // Single source of truth for which relationship types are part of the family graph.
+    // Consulted by addRelationship (to set family_member on both endpoints) and by
+    // getFamilyNetwork (to filter the edges returned). FRIEND/COLLEAGUE/etc. are excluded.
+    private static final List<RelationType> FAMILY_RELATION_TYPES =
+            List.of(RelationType.PARENT_OF, RelationType.SPOUSE_OF, RelationType.SIBLING_OF);
+
     private final PersonRelationshipRepository relationshipRepository;
     private final PersonService personService;
     private final RelationshipInferenceService inferenceService;
@@ -60,11 +66,12 @@ public class RelationshipService {
         for (Person p : familyMembers) {
             familyIds.add(p.getId());
             nodes.add(new PersonNodeDTO(
-                    p.getId(), p.getDisplayName(), p.getBirthYear(), p.getDeathYear(), true));
+                    p.getId(), p.getDisplayName(), p.getBirthYear(), p.getDeathYear(),
+                    p.getGeneration(), true));
         }
 
         List<PersonRelationship> familyEdges = relationshipRepository.findAllByRelationTypeIn(
-                List.of(RelationType.PARENT_OF, RelationType.SPOUSE_OF, RelationType.SIBLING_OF));
+                FAMILY_RELATION_TYPES);
 
         List<RelationshipDTO> edges = new ArrayList<>();
         for (PersonRelationship r : familyEdges) {
@@ -105,15 +112,23 @@ public class RelationshipService {
                 .notes(blankToNull(dto.notes()))
                 .build();
 
+        PersonRelationship saved;
         try {
             // saveAndFlush so the unique_rel constraint violates synchronously and is
             // caught here, not at commit time outside the @Transactional boundary.
-            return toDTO(relationshipRepository.saveAndFlush(rel));
+            saved = relationshipRepository.saveAndFlush(rel);
         } catch (DataIntegrityViolationException e) {
             throw DomainException.conflict(
                     ErrorCode.DUPLICATE_RELATIONSHIP,
                     "Relationship already exists for (" + personId + ", " + relatedPerson.getId() + ", " + dto.relationType() + ")");
         }
+        // Family-graph edges imply both endpoints are family members. Idempotent: the
+        // setter is a no-op when the person is already flagged, so re-imports stay clean.
+        if (FAMILY_RELATION_TYPES.contains(dto.relationType())) {
+            personService.setFamilyMember(person.getId(), true);
+            personService.setFamilyMember(relatedPerson.getId(), true);
+        }
+        return toDTO(saved);
     }
 
     @Transactional

backend/src/main/java/org/raddatz/familienarchiv/person/relationship/dto/PersonNodeDTO.java

@@ -10,5 +10,6 @@ public record PersonNodeDTO(
         @Schema(requiredMode = Schema.RequiredMode.REQUIRED) String displayName,
         Integer birthYear,
         Integer deathYear,
+        Integer generation,
         @Schema(requiredMode = Schema.RequiredMode.REQUIRED) boolean familyMember
 ) {}

backend/src/main/java/org/raddatz/familienarchiv/search/NlQueryInterpretation.java

@@ -0,0 +1,22 @@
+package org.raddatz.familienarchiv.search;
+
+import io.swagger.v3.oas.annotations.media.Schema;
+
+import java.time.LocalDate;
+import java.util.List;
+
+public record NlQueryInterpretation(
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        List<PersonHint> resolvedPersons,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        List<PersonHint> ambiguousPersons,
+        LocalDate dateFrom,
+        LocalDate dateTo,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        List<String> keywords,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        String rawQuery,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        boolean keywordsApplied
+) {
+}

backend/src/main/java/org/raddatz/familienarchiv/search/NlQueryParserService.java

@@ -0,0 +1,160 @@
+package org.raddatz.familienarchiv.search;
+
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.raddatz.familienarchiv.document.DocumentSearchResult;
+import org.raddatz.familienarchiv.document.DocumentService;
+import org.raddatz.familienarchiv.document.DocumentSort;
+import org.raddatz.familienarchiv.document.SearchFilters;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.raddatz.familienarchiv.person.Person;
+import org.raddatz.familienarchiv.person.PersonService;
+import org.raddatz.familienarchiv.tag.TagOperator;
+import org.springframework.data.domain.Pageable;
+import org.springframework.stereotype.Service;
+
+import java.time.LocalDate;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.UUID;
+
+@Service
+@RequiredArgsConstructor
+@Slf4j
+public class NlQueryParserService {
+
+    private static final int MIN_QUERY = 3;
+    private static final int MAX_QUERY = 500;
+    private static final int MAX_NAME_LENGTH = 200;
+    private static final int MAX_CANDIDATES = 10;
+
+    private final OllamaClient ollamaClient;
+    private final PersonService personService;
+    private final DocumentService documentService;
+
+    public NlSearchResponse search(String query, Pageable pageable) {
+        if (query == null || query.length() < MIN_QUERY || query.length() > MAX_QUERY) {
+            throw DomainException.badRequest(ErrorCode.VALIDATION_ERROR,
+                    "Query must be between " + MIN_QUERY + " and " + MAX_QUERY + " characters");
+        }
+
+        OllamaExtraction ext = ollamaClient.parse(query);
+
+        List<String> personNames = ext.personNames() != null ? ext.personNames() : List.of();
+        List<String> keywords = ext.keywords() != null ? ext.keywords() : List.of();
+
+        NameResolution resolution = resolveNames(personNames);
+
+        if (!resolution.ambiguous().isEmpty()) {
+            NlQueryInterpretation interpretation = new NlQueryInterpretation(
+                    List.of(), resolution.ambiguous(),
+                    ext.dateFrom(), ext.dateTo(),
+                    keywords, ext.rawQuery(), false);
+            return new NlSearchResponse(DocumentSearchResult.of(List.of()), interpretation);
+        }
+
+        List<PersonHint> resolved = resolution.resolved();
+        List<String> noMatchFragments = resolution.noMatchFragments();
+        List<String> extraFragments = resolution.extraFragments();
+
+        String text = buildText(keywords, noMatchFragments, extraFragments, ext.rawQuery());
+
+        if (resolved.size() == 1 && isAnyRole(ext.personRole())) {
+            UUID personId = resolved.get(0).id();
+            DocumentSearchResult docs = documentService.searchDocumentsByPersonId(
+                    personId, ext.dateFrom(), ext.dateTo(), pageable);
+            NlQueryInterpretation interpretation = new NlQueryInterpretation(
+                    resolved, List.of(), ext.dateFrom(), ext.dateTo(), keywords, ext.rawQuery(), false);
+            return new NlSearchResponse(docs, interpretation);
+        }
+
+        UUID sender = buildSender(resolved, ext.personRole());
+        UUID receiver = buildReceiver(resolved, ext.personRole());
+
+        SearchFilters filters = new SearchFilters(
+                text.isBlank() ? null : text,
+                ext.dateFrom(), ext.dateTo(),
+                sender, receiver,
+                List.of(), null,
+                null, TagOperator.AND, false);
+
+        DocumentSearchResult docs = documentService.searchDocuments(filters, DocumentSort.DATE, "desc", pageable);
+        boolean keywordsApplied = !text.isBlank();
+        NlQueryInterpretation interpretation = new NlQueryInterpretation(
+                resolved, List.of(), ext.dateFrom(), ext.dateTo(), keywords, ext.rawQuery(), keywordsApplied);
+        return new NlSearchResponse(docs, interpretation);
+    }
+
+    private NameResolution resolveNames(List<String> personNames) {
+        List<PersonHint> resolved = new ArrayList<>();
+        List<PersonHint> ambiguous = new ArrayList<>();
+        List<String> noMatchFragments = new ArrayList<>();
+        List<String> extraFragments = new ArrayList<>();
+
+        int resolvedIndex = 0;
+        for (String name : personNames) {
+            if (name == null || name.length() > MAX_NAME_LENGTH) {
+                log.debug("Skipping name fragment (too long or null): length={}", name == null ? 0 : name.length());
+                continue;
+            }
+            List<Person> candidates = personService.findByDisplayNameContaining(name);
+            List<Person> capped = candidates.size() > MAX_CANDIDATES
+                    ? candidates.subList(0, MAX_CANDIDATES)
+                    : candidates;
+
+            if (capped.isEmpty()) {
+                noMatchFragments.add(name);
+            } else if (capped.size() == 1) {
+                Person p = capped.get(0);
+                PersonHint hint = new PersonHint(p.getId(), p.getDisplayName());
+                resolvedIndex++;
+                if (resolvedIndex <= 2) {
+                    resolved.add(hint);
+                } else {
+                    extraFragments.add(name);
+                }
+            } else {
+                capped.forEach(p -> ambiguous.add(new PersonHint(p.getId(), p.getDisplayName())));
+            }
+        }
+
+        return new NameResolution(resolved, ambiguous, noMatchFragments, extraFragments);
+    }
+
+    private String buildText(List<String> keywords, List<String> noMatchFragments,
+                             List<String> extraFragments, String rawQuery) {
+        List<String> parts = new ArrayList<>();
+        parts.addAll(keywords);
+        parts.addAll(noMatchFragments);
+        parts.addAll(extraFragments);
+        String text = String.join(" ", parts).strip();
+        if (text.isBlank() && rawQuery != null && !rawQuery.isBlank()) {
+            return rawQuery;
+        }
+        return text;
+    }
+
+    private boolean isAnyRole(String role) {
+        return role == null || "any".equals(role) || (!"sender".equals(role) && !"receiver".equals(role));
+    }
+
+    private UUID buildSender(List<PersonHint> resolved, String role) {
+        if (resolved.size() >= 2) return resolved.get(0).id();
+        if (resolved.size() == 1 && "sender".equals(role)) return resolved.get(0).id();
+        return null;
+    }
+
+    private UUID buildReceiver(List<PersonHint> resolved, String role) {
+        if (resolved.size() >= 2) return resolved.get(1).id();
+        if (resolved.size() == 1 && "receiver".equals(role)) return resolved.get(0).id();
+        return null;
+    }
+
+    private record NameResolution(
+            List<PersonHint> resolved,
+            List<PersonHint> ambiguous,
+            List<String> noMatchFragments,
+            List<String> extraFragments
+    ) {}
+}

backend/src/main/java/org/raddatz/familienarchiv/search/NlSearchController.java

@@ -0,0 +1,28 @@
+package org.raddatz.familienarchiv.search;
+
+import jakarta.validation.Valid;
+import lombok.RequiredArgsConstructor;
+import org.raddatz.familienarchiv.security.Permission;
+import org.raddatz.familienarchiv.security.RequirePermission;
+import org.springframework.data.domain.Pageable;
+import org.springframework.security.core.annotation.AuthenticationPrincipal;
+import org.springframework.security.core.userdetails.UserDetails;
+import org.springframework.web.bind.annotation.*;
+
+@RestController
+@RequestMapping("/api/search/nl")
+@RequiredArgsConstructor
+public class NlSearchController {
+
+    private final NlQueryParserService nlQueryParserService;
+    private final NlSearchRateLimiter rateLimiter;
+
+    @PostMapping
+    @RequirePermission(Permission.READ_ALL)
+    public NlSearchResponse search(@Valid @RequestBody NlSearchRequest request,
+                                   Pageable pageable,
+                                   @AuthenticationPrincipal UserDetails principal) {
+        rateLimiter.checkAndConsume(principal.getUsername());
+        return nlQueryParserService.search(request.query(), pageable);
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/search/NlSearchRateLimitProperties.java

@@ -0,0 +1,12 @@
+package org.raddatz.familienarchiv.search;
+
+import lombok.Data;
+import org.springframework.boot.context.properties.ConfigurationProperties;
+import org.springframework.stereotype.Component;
+
+@Component
+@ConfigurationProperties("app.nl-search.rate-limit")
+@Data
+public class NlSearchRateLimitProperties {
+    private int maxRequestsPerMinute = 5;
+}

backend/src/main/java/org/raddatz/familienarchiv/search/NlSearchRateLimiter.java

@@ -0,0 +1,46 @@
+package org.raddatz.familienarchiv.search;
+
+import com.github.benmanes.caffeine.cache.Caffeine;
+import com.github.benmanes.caffeine.cache.LoadingCache;
+import io.github.bucket4j.Bandwidth;
+import io.github.bucket4j.Bucket;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.springframework.stereotype.Service;
+
+import java.time.Duration;
+import java.util.concurrent.TimeUnit;
+
+@Service
+public class NlSearchRateLimiter {
+
+    private final LoadingCache<String, Bucket> byUser;
+    private final int maxRequestsPerMinute;
+
+    public NlSearchRateLimiter(NlSearchRateLimitProperties props) {
+        this.maxRequestsPerMinute = props.getMaxRequestsPerMinute();
+        this.byUser = Caffeine.newBuilder()
+                .expireAfterAccess(1, TimeUnit.MINUTES)
+                .build(key -> newBucket(maxRequestsPerMinute));
+    }
+
+    public void checkAndConsume(String userKey) {
+        if (!byUser.get(userKey).tryConsume(1)) {
+            throw DomainException.tooManyRequests(ErrorCode.SMART_SEARCH_RATE_LIMITED,
+                    "NL search rate limit exceeded for user: " + userKey, 60L);
+        }
+    }
+
+    void resetForTest() {
+        byUser.invalidateAll();
+    }
+
+    private static Bucket newBucket(int limit) {
+        return Bucket.builder()
+                .addLimit(Bandwidth.builder()
+                        .capacity(limit)
+                        .refillGreedy(limit, Duration.ofMinutes(1))
+                        .build())
+                .build();
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/search/NlSearchRequest.java

@@ -0,0 +1,11 @@
+package org.raddatz.familienarchiv.search;
+
+import jakarta.validation.constraints.NotBlank;
+import jakarta.validation.constraints.Size;
+
+public record NlSearchRequest(
+        @NotBlank
+        @Size(min = 3, max = 500)
+        String query
+) {
+}

backend/src/main/java/org/raddatz/familienarchiv/search/NlSearchResponse.java

@@ -0,0 +1,12 @@
+package org.raddatz.familienarchiv.search;
+
+import io.swagger.v3.oas.annotations.media.Schema;
+import org.raddatz.familienarchiv.document.DocumentSearchResult;
+
+public record NlSearchResponse(
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        DocumentSearchResult result,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        NlQueryInterpretation interpretation
+) {
+}

backend/src/main/java/org/raddatz/familienarchiv/search/OllamaClient.java

@@ -0,0 +1,5 @@
+package org.raddatz.familienarchiv.search;
+
+public interface OllamaClient {
+    OllamaExtraction parse(String query);
+}

backend/src/main/java/org/raddatz/familienarchiv/search/OllamaExtraction.java

@@ -0,0 +1,18 @@
+package org.raddatz.familienarchiv.search;
+
+import java.time.LocalDate;
+import java.util.List;
+
+/**
+ * Raw structured output from Ollama after parsing and sanitising.
+ * personRole is always one of "sender", "receiver", "any" — defensive parsing ensures this.
+ */
+record OllamaExtraction(
+        List<String> personNames,
+        String personRole,
+        LocalDate dateFrom,
+        LocalDate dateTo,
+        List<String> keywords,
+        String rawQuery
+) {
+}

backend/src/main/java/org/raddatz/familienarchiv/search/OllamaHealthClient.java

@@ -0,0 +1,5 @@
+package org.raddatz.familienarchiv.search;
+
+public interface OllamaHealthClient {
+    boolean isHealthy();
+}

backend/src/main/java/org/raddatz/familienarchiv/search/OllamaProperties.java

@@ -0,0 +1,15 @@
+package org.raddatz.familienarchiv.search;
+
+import lombok.Data;
+import org.springframework.boot.context.properties.ConfigurationProperties;
+import org.springframework.stereotype.Component;
+
+@Component
+@ConfigurationProperties("app.ollama")
+@Data
+public class OllamaProperties {
+    private String baseUrl;
+    private String model;
+    private int timeoutSeconds = 30;
+    private int healthCheckTimeoutSeconds = 2;
+}

backend/src/main/java/org/raddatz/familienarchiv/search/PersonHint.java

@@ -0,0 +1,13 @@
+package org.raddatz.familienarchiv.search;
+
+import io.swagger.v3.oas.annotations.media.Schema;
+
+import java.util.UUID;
+
+public record PersonHint(
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        UUID id,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+        String displayName
+) {
+}

backend/src/main/java/org/raddatz/familienarchiv/search/RestClientOllamaClient.java

@@ -0,0 +1,184 @@
+package org.raddatz.familienarchiv.search;
+
+import com.fasterxml.jackson.annotation.JsonIgnoreProperties;
+import com.fasterxml.jackson.annotation.JsonProperty;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import lombok.extern.slf4j.Slf4j;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.springframework.http.client.JdkClientHttpRequestFactory;
+import org.springframework.stereotype.Service;
+import org.springframework.web.client.RestClient;
+import org.springframework.web.client.RestClientException;
+
+import java.net.http.HttpClient;
+import java.time.Duration;
+import java.time.LocalDate;
+import java.time.Year;
+import java.time.format.DateTimeFormatter;
+import java.time.format.DateTimeParseException;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+
+@Service
+@Slf4j
+public class RestClientOllamaClient implements OllamaClient, OllamaHealthClient {
+
+    private static final ObjectMapper MAPPER = new ObjectMapper();
+    private static final Set<String> VALID_ROLES = Set.of("sender", "receiver", "any");
+    private static final int MAX_NAME_LENGTH = 200;
+    private static final int MAX_KEYWORD_LENGTH = 100;
+
+    private static final Map<String, Object> JSON_SCHEMA = Map.of(
+            "type", "object",
+            "required", List.of("personNames", "personRole", "keywords"),
+            "properties", Map.of(
+                    "personNames", Map.of("type", "array", "items", Map.of("type", "string", "maxLength", MAX_NAME_LENGTH)),
+                    "personRole", Map.of("type", "string", "enum", List.of("sender", "receiver", "any")),
+                    "dateFrom", Map.of("type", List.of("string", "null"), "maxLength", 20),
+                    "dateTo", Map.of("type", List.of("string", "null"), "maxLength", 20),
+                    "keywords", Map.of("type", "array", "items", Map.of("type", "string", "maxLength", MAX_KEYWORD_LENGTH))
+            )
+    );
+
+    private final RestClient inferenceClient;
+    private final RestClient healthClient;
+    private final OllamaProperties props;
+
+    public RestClientOllamaClient(OllamaProperties props) {
+        this.props = props;
+
+        HttpClient inferenceHttp = HttpClient.newBuilder()
+                .version(HttpClient.Version.HTTP_1_1)
+                .connectTimeout(Duration.ofSeconds(10))
+                .build();
+        JdkClientHttpRequestFactory inferenceFactory = new JdkClientHttpRequestFactory(inferenceHttp);
+        inferenceFactory.setReadTimeout(Duration.ofSeconds(props.getTimeoutSeconds()));
+        this.inferenceClient = RestClient.builder()
+                .baseUrl(props.getBaseUrl())
+                .requestFactory(inferenceFactory)
+                .build();
+
+        HttpClient healthHttp = HttpClient.newBuilder()
+                .version(HttpClient.Version.HTTP_1_1)
+                .connectTimeout(Duration.ofSeconds(props.getHealthCheckTimeoutSeconds()))
+                .build();
+        JdkClientHttpRequestFactory healthFactory = new JdkClientHttpRequestFactory(healthHttp);
+        healthFactory.setReadTimeout(Duration.ofSeconds(props.getHealthCheckTimeoutSeconds()));
+        this.healthClient = RestClient.builder()
+                .baseUrl(props.getBaseUrl())
+                .requestFactory(healthFactory)
+                .build();
+    }
+
+    @Override
+    public OllamaExtraction parse(String query) {
+        try {
+            OllamaGenerateRequest request = new OllamaGenerateRequest(
+                    props.getModel(), query, JSON_SCHEMA, false);
+            String responseBody = inferenceClient.post()
+                    .uri("/api/generate")
+                    .contentType(org.springframework.http.MediaType.APPLICATION_JSON)
+                    .body(request)
+                    .retrieve()
+                    .body(String.class);
+            return parseOllamaResponse(responseBody, query);
+        } catch (DomainException e) {
+            throw e;
+        } catch (Exception e) {
+            log.warn("Ollama inference failed: {}", e.getClass().getSimpleName());
+            throw DomainException.serviceUnavailable(ErrorCode.SMART_SEARCH_UNAVAILABLE,
+                    "Ollama unavailable: " + e.getClass().getSimpleName());
+        }
+    }
+
+    @Override
+    public boolean isHealthy() {
+        try {
+            healthClient.get().uri("/api/tags").retrieve().toBodilessEntity();
+            return true;
+        } catch (Exception e) {
+            return false;
+        }
+    }
+
+    private OllamaExtraction parseOllamaResponse(String responseBody, String rawQuery) {
+        try {
+            OllamaGenerateResponse response = MAPPER.readValue(responseBody, OllamaGenerateResponse.class);
+            String inner = response.response();
+            if (inner == null || inner.isBlank()) {
+                return fallbackExtraction(rawQuery);
+            }
+            RawOllamaOutput raw = MAPPER.readValue(inner, RawOllamaOutput.class);
+            return toExtraction(raw, rawQuery);
+        } catch (Exception e) {
+            log.warn("Failed to parse Ollama response: {}", e.getClass().getSimpleName());
+            throw DomainException.serviceUnavailable(ErrorCode.SMART_SEARCH_UNAVAILABLE,
+                    "Failed to parse Ollama response: " + e.getClass().getSimpleName());
+        }
+    }
+
+    private OllamaExtraction toExtraction(RawOllamaOutput raw, String rawQuery) {
+        List<String> names = raw.personNames() == null ? List.of() : raw.personNames().stream()
+                .filter(n -> n != null && n.length() <= MAX_NAME_LENGTH)
+                .toList();
+        List<String> keywords = raw.keywords() == null ? List.of() : raw.keywords().stream()
+                .filter(k -> k != null && k.length() <= MAX_KEYWORD_LENGTH)
+                .toList();
+        String role = sanitiseRole(raw.personRole());
+        LocalDate dateFrom = parseDate(raw.dateFrom(), true);
+        LocalDate dateTo = parseDate(raw.dateTo(), false);
+        return new OllamaExtraction(names, role, dateFrom, dateTo, keywords, rawQuery);
+    }
+
+    private OllamaExtraction fallbackExtraction(String rawQuery) {
+        return new OllamaExtraction(List.of(), "any", null, null, List.of(), rawQuery);
+    }
+
+    private String sanitiseRole(String role) {
+        if (role != null && VALID_ROLES.contains(role)) {
+            return role;
+        }
+        log.warn("Unexpected personRole from Ollama: {}", role);
+        return "any";
+    }
+
+    private LocalDate parseDate(String raw, boolean isFrom) {
+        if (raw == null || raw.isBlank()) return null;
+        try {
+            return LocalDate.parse(raw, DateTimeFormatter.ISO_LOCAL_DATE);
+        } catch (DateTimeParseException ignored) {
+        }
+        try {
+            int year = Integer.parseInt(raw.strip());
+            if (year > 1000 && year < 3000) {
+                return isFrom ? Year.of(year).atDay(1) : Year.of(year).atMonth(12).atEndOfMonth();
+            }
+        } catch (NumberFormatException ignored) {
+        }
+        return null;
+    }
+
+    @JsonIgnoreProperties(ignoreUnknown = true)
+    private record OllamaGenerateResponse(String response) {
+    }
+
+    @JsonIgnoreProperties(ignoreUnknown = true)
+    private record RawOllamaOutput(
+            @JsonProperty("personNames") List<String> personNames,
+            @JsonProperty("personRole") String personRole,
+            @JsonProperty("dateFrom") String dateFrom,
+            @JsonProperty("dateTo") String dateTo,
+            @JsonProperty("keywords") List<String> keywords
+    ) {
+    }
+
+    private record OllamaGenerateRequest(
+            String model,
+            String prompt,
+            Object format,
+            boolean stream
+    ) {
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/tag/README.md

@@ -7,6 +7,13 @@ Hierarchical document categories. Tags form a tree via a self-referencing `paren
 Entity: `Tag` (self-referencing `parent_id` tree).
 Features: tag CRUD, hierarchical deletion (cascade to descendants), tag typeahead, admin tag management (rename, reparent, merge).
 
+## Tag tree counts (`getTagTree`)
+
+`GET /api/tags/tree` returns each node with **two** document counts, from two aggregate queries (no N+1):
+
+- `documentCount` — documents tagged with that **exact** tag (direct). Read by the admin surfaces (sidebar tree, merge preview, delete-impact guard), which describe direct-document operations.
+- `subtreeDocumentCount` — **distinct** documents tagged with that tag **or any descendant** (subtree rollup, recursive-CTE closure, depth guard ≤50). Read by the reader surfaces (`/themen` page, dashboard `ThemenWidget`) so the box number matches what `/documents?tag=X` actually finds.
+
 ## What this domain does NOT own
 
 - Documents — the `document_tags` join table is on the document side. `Tag` does not hold document references.

backend/src/main/java/org/raddatz/familienarchiv/tag/TagRepository.java

@@ -20,7 +20,14 @@ public interface TagRepository extends JpaRepository<Tag, UUID> {
     }
 
 
-    Optional<Tag> findByNameIgnoreCase(String name);
+    // Tag-name resolution (see TagService.findOrCreate). Names that collide case-insensitively across
+    // the canonical tree are VALID — a parent and its same-named lowercase child (e.g. "Geburt" /
+    // "Geburt/geburt") are distinct nodes with their own source_ref and document attachments. So
+    // resolution must be exact-case first, then a non-throwing list for the case-insensitive fallback.
+    // Do NOT add a unique(lower(name)) constraint — it would reject these legitimate rows. See #730.
+    Optional<Tag> findByName(String name);
+
+    List<Tag> findAllByNameIgnoreCase(String name);
 
     // Lookup by the canonical tag_path, used for idempotent canonical re-import (Phase 3).
     Optional<Tag> findBySourceRef(String sourceRef);
@@ -126,4 +133,31 @@ public interface TagRepository extends JpaRepository<Tag, UUID> {
      */
     @Query(value = "SELECT tag_id AS tagId, COUNT(*) AS count FROM document_tags GROUP BY tag_id", nativeQuery = true)
     List<TagCount> findDocumentCountsPerTag();
+
+    /**
+     * Returns (tagId, count) pairs where count is the number of <b>distinct</b> documents tagged
+     * with that tag <b>or any of its descendants</b> (full subtree rollup).
+     * <p>
+     * Builds a tag closure of (ancestor_id, descendant_id) pairs via a recursive CTE — each tag is
+     * its own ancestor at depth 0, then descends into children (depth guard of 50 levels prevents a
+     * cycle or pathological depth from running away) — joins it to {@code document_tags} on the
+     * descendant, and counts distinct documents per ancestor. A document tagged with several tags in
+     * the same subtree is therefore counted once. Tags whose entire subtree holds no documents do
+     * not appear in the result (they default to 0 in the tree). One aggregate query for all tags.
+     */
+    @Query(value = """
+            WITH RECURSIVE closure AS (
+                SELECT id AS ancestor_id, id AS descendant_id, 0 AS depth FROM tag
+                UNION ALL
+                SELECT c.ancestor_id, t.id AS descendant_id, c.depth + 1
+                FROM tag t
+                JOIN closure c ON t.parent_id = c.descendant_id
+                WHERE c.depth < 50
+            )
+            SELECT c.ancestor_id AS tagId, COUNT(DISTINCT dt.document_id) AS count
+            FROM closure c
+            JOIN document_tags dt ON dt.tag_id = c.descendant_id
+            GROUP BY c.ancestor_id
+            """, nativeQuery = true)
+    List<TagCount> findSubtreeDocumentCountsPerTag();
 }

backend/src/main/java/org/raddatz/familienarchiv/tag/TagService.java

@@ -2,6 +2,7 @@ package org.raddatz.familienarchiv.tag;
 
 import java.util.ArrayList;
 import java.util.Collection;
+import java.util.Comparator;
 import java.util.HashMap;
 import java.util.HashSet;
 import java.util.LinkedHashMap;
@@ -55,10 +56,21 @@ public class TagService {
         return tagRepository.findBySourceRef(sourceRef);
     }
 
+    /**
+     * Resolves a tag name to a single tag, creating one when absent. Never throws on case-insensitive
+     * collisions: names that differ only by case are valid distinct nodes in the canonical tree (a
+     * parent and its same-named lowercase child), so resolution prefers an exact-case match, then
+     * falls back to the lowest-id case-insensitive match, then creates. See #730.
+     */
     public Tag findOrCreate(String name) {
         String cleanName = name.trim();
-        return tagRepository.findByNameIgnoreCase(cleanName)
-                .orElseGet(() -> tagRepository.save(Tag.builder().name(cleanName).build()));
+        Optional<Tag> exact = tagRepository.findByName(cleanName);
+        if (exact.isPresent()) return exact.get();                          // exact-case wins (edit round-trip replays the stored name)
+        List<Tag> caseInsensitive = tagRepository.findAllByNameIgnoreCase(cleanName);
+        if (!caseInsensitive.isEmpty()) {
+            return caseInsensitive.stream().min(Comparator.comparing(Tag::getId)).orElseThrow(); // deterministic tie-break by id — list is non-empty, never throws
+        }
+        return tagRepository.save(Tag.builder().name(cleanName).build());   // create-when-absent (orphan tag: null sourceRef/parentId)
     }
 
     /**
@@ -172,19 +184,27 @@ public class TagService {
     }
 
     /**
-     * Returns all tags assembled into a tree with document counts per node.
-     * Uses a single aggregate query to avoid N+1 behaviour.
-     * NOTE: document counts are global per tag, not scoped to any search filter.
-     * The tree endpoint is only used for the admin sidebar, so this is intentional.
+     * Returns all tags assembled into a tree, each node carrying two counts:
+     * {@code documentCount} — documents tagged with that exact tag (direct) — and
+     * {@code subtreeDocumentCount} — distinct documents tagged with that tag or any descendant
+     * (subtree rollup). Each count comes from one aggregate query (no N+1).
+     * NOTE: counts are global per tag, not scoped to any search filter.
+     * Consumed by the reader surfaces (/themen page, dashboard ThemenWidget — which read the
+     * subtree rollup) as well as the admin sidebar and tag operation previews (which read the
+     * direct count).
      */
     public List<TagTreeNodeDTO> getTagTree() {
         List<Tag> all = tagRepository.findAll();
-        Map<UUID, Long> counts = tagRepository.findDocumentCountsPerTag().stream()
-                .collect(Collectors.toMap(
-                        TagRepository.TagCount::getTagId,
-                        TagRepository.TagCount::getCount
-                ));
-        return buildTree(all, counts);
+        Map<UUID, Long> counts = toCountMap(tagRepository.findDocumentCountsPerTag());
+        Map<UUID, Long> subtreeCounts = toCountMap(tagRepository.findSubtreeDocumentCountsPerTag());
+        return buildTree(all, counts, subtreeCounts);
+    }
+
+    private static Map<UUID, Long> toCountMap(List<TagRepository.TagCount> counts) {
+        return counts.stream().collect(Collectors.toMap(
+                TagRepository.TagCount::getTagId,
+                TagRepository.TagCount::getCount
+        ));
     }
 
     // ─── private helpers ─────────────────────────────────────────────────────
@@ -259,12 +279,14 @@ public class TagService {
         }
     }
 
-    private List<TagTreeNodeDTO> buildTree(List<Tag> tags, Map<UUID, Long> counts) {
+    private List<TagTreeNodeDTO> buildTree(List<Tag> tags, Map<UUID, Long> counts,
+                                           Map<UUID, Long> subtreeCounts) {
         Map<UUID, TagTreeNodeDTO> nodeById = new LinkedHashMap<>();
         for (Tag tag : tags) {
             int documentCount = counts.getOrDefault(tag.getId(), 0L).intValue();
+            int subtreeDocumentCount = subtreeCounts.getOrDefault(tag.getId(), 0L).intValue();
             nodeById.put(tag.getId(), new TagTreeNodeDTO(
-                    tag.getId(), tag.getName(), tag.getColor(), documentCount,
+                    tag.getId(), tag.getName(), tag.getColor(), documentCount, subtreeDocumentCount,
                     new ArrayList<>(), tag.getParentId()
             ));
         }

backend/src/main/java/org/raddatz/familienarchiv/tag/TagTreeNodeDTO.java

@@ -10,5 +10,8 @@ public record TagTreeNodeDTO(
         @Schema(requiredMode = Schema.RequiredMode.REQUIRED) String name,
         String color,
         @Schema(requiredMode = Schema.RequiredMode.REQUIRED) int documentCount,
+        @Schema(requiredMode = Schema.RequiredMode.REQUIRED,
+                description = "Distinct documents tagged with this tag or any descendant tag (subtree rollup)")
+        int subtreeDocumentCount,
         List<TagTreeNodeDTO> children,
         @Schema(description = "Parent tag ID, null for root tags") UUID parentId) {}

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

@@ -51,6 +51,12 @@ public class AdminController {
         return ResponseEntity.ok(new BackfillResult(count));
     }
 
+    @PostMapping("/backfill-titles")
+    public ResponseEntity<BackfillResult> backfillTitles() {
+        int count = documentService.backfillTitles();
+        return ResponseEntity.ok(new BackfillResult(count));
+    }
+
     @PostMapping("/generate-thumbnails")
     public ResponseEntity<ThumbnailBackfillService.BackfillStatus> generateThumbnails() {
         thumbnailBackfillService.runBackfillAsync();

backend/src/main/resources/application-dev.yaml

@@ -11,3 +11,7 @@ springdoc:
   swagger-ui:
     enabled: true
     path: /swagger-ui.html
+
+app:
+  ollama:
+    base-url: http://localhost:11434

backend/src/main/resources/application.yaml

@@ -130,6 +130,16 @@ app:
     # The loader maps columns by header name — no positional indices (see ADR-025).
     dir: ${IMPORT_DIR:/import}
 
+  ollama:
+    base-url: http://ollama:11434
+    model: qwen2.5:7b-instruct-q4_K_M
+    timeout-seconds: 30
+    health-check-timeout-seconds: 2
+
+  nl-search:
+    rate-limit:
+      max-requests-per-minute: 5
+
 ocr:
   sender-model:
     activation-threshold: 100

backend/src/main/resources/db/migration/V70__add_generation_to_persons.sql

@@ -0,0 +1,26 @@
+-- #689: persist the hand-curated "G 0…G 5" generation index from
+-- canonical-persons.xlsx so the Stammbaum layout can use it as a strict
+-- rank anchor (replacing the current iterative longest-path heuristic that
+-- silently misplaces loose spouses with their own parents in the graph).
+--
+-- Nullable: pre-import rows and persons outside the curated family graph
+-- legitimately have no generation. The canonical importer back-fills via
+-- preferHuman on the next run; a human-edited value is never overwritten
+-- (see ADR-025).
+
+ALTER TABLE persons ADD COLUMN generation SMALLINT;
+
+-- Allowlist of valid generation indices. The 0..10 bounds mirror
+-- PersonGeneration.MIN_GENERATION / MAX_GENERATION in Java — keep the
+-- two in sync (the DTO @Min/@Max and both importer range guards read from
+-- those Java constants). Current data tops out at G 5, but a future G 6 →
+-- G 10 widening needs no migration. A G −1 ancestor would require a
+-- separate one-shot shift migration (out of scope here; the layout's
+-- normalise step already handles negative seeds at render time).
+ALTER TABLE persons ADD CONSTRAINT chk_generation_range
+    CHECK (generation IS NULL OR generation BETWEEN 0 AND 10);
+
+-- Partial index: only the curated rows (≈ 163 of 1,105) ever get a value,
+-- and the layout only ever queries for non-null rows.
+CREATE INDEX idx_persons_generation ON persons (generation)
+    WHERE generation IS NOT NULL;

backend/src/main/resources/db/migration/V71__person_delete_on_delete_fk.sql

@@ -0,0 +1,53 @@
+-- Move person-delete referential integrity from application code into the database (#684).
+--
+-- Before this migration, PersonService.deletePerson nulled documents.sender_id and removed
+-- document_receivers rows in Java before deleting the person, because the two V1 FKs into
+-- persons had no ON DELETE behaviour. Any other delete path (a future endpoint, a manual
+-- psql, a batch job) could still orphan rows or 500. This migration makes the database the
+-- single source of truth so a person delete is safe from every path.
+--
+-- Cascade boundary: the cascade stays STRICTLY at the join/reference layer and NEVER reaches
+-- documents rows — a cascade into documents would destroy historical letters. sender_id is
+-- SET NULL (documents.senderText preserves the raw textual attribution); the receiver join
+-- row and the @-mention sidecar row are dropped.
+--
+-- No NOT VALID + VALIDATE two-step: these tables are small (thousands of rows → sub-second
+-- ACCESS EXCLUSIVE lock). Do NOT copy this drop-and-recreate pattern onto a large table.
+--
+-- Not audit-logged: a DB ON DELETE cascade runs below AuditService — a known, accepted trade.
+-- The person-delete action itself is still logged at the service layer.
+
+-- documents.sender_id → ON DELETE SET NULL (deleted sender clears the link; the document survives).
+ALTER TABLE public.documents
+    DROP CONSTRAINT fkl5xhww7es3b4um01vmly4y18m,
+    ADD CONSTRAINT fkl5xhww7es3b4um01vmly4y18m
+        FOREIGN KEY (sender_id) REFERENCES public.persons(id) ON DELETE SET NULL;
+
+-- document_receivers.person_id → ON DELETE CASCADE (drop the join row), the symmetric
+-- completion of V14, which added the same to the document_id side of this table.
+ALTER TABLE public.document_receivers
+    DROP CONSTRAINT fkcg7r68qvosqricx1betgrlt7s,
+    ADD CONSTRAINT fkcg7r68qvosqricx1betgrlt7s
+        FOREIGN KEY (person_id) REFERENCES public.persons(id) ON DELETE CASCADE;
+
+-- Soft reference fix: transcription_block_mentioned_persons.person_id was a UUID with no FK
+-- (V56), so deleting a person left dangling mention rows. Give it a real FK with CASCADE.
+-- This reverses V56's deliberate "no FK on person_id" choice — that comment is now historical
+-- but is intentionally left untouched, because editing an already-applied migration changes its
+-- Flyway checksum and would fail validateOnMigrate in prod. ADR-032 is the authoritative record.
+-- Clean up pre-existing orphans first — production likely holds dangling rows because the old
+-- deletePerson never cleaned mention rows, and the ADD CONSTRAINT validation scan fails on them.
+-- A DO block with RAISE NOTICE surfaces the purge count: Flyway runs each statement via JDBC
+-- and discards a trailing SELECT's result set, so a "SELECT count(*)" would log nothing.
+DO $$
+DECLARE removed int;
+BEGIN
+    DELETE FROM transcription_block_mentioned_persons m
+    WHERE NOT EXISTS (SELECT 1 FROM persons p WHERE p.id = m.person_id);
+    GET DIAGNOSTICS removed = ROW_COUNT;
+    RAISE NOTICE 'V71 orphaned_mention_rows_removed=%', removed;
+END $$;
+
+ALTER TABLE public.transcription_block_mentioned_persons
+    ADD CONSTRAINT fk_tbmp_person
+        FOREIGN KEY (person_id) REFERENCES public.persons(id) ON DELETE CASCADE;

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

@@ -38,7 +38,6 @@ import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
 import static org.mockito.ArgumentMatchers.any;
-import static org.mockito.ArgumentMatchers.anyBoolean;
 import static org.mockito.ArgumentMatchers.anyInt;
 import static org.mockito.ArgumentMatchers.eq;
 import static org.mockito.Mockito.verify;
@@ -76,7 +75,7 @@ class DocumentControllerTest {
     @Test
     @WithMockUser
     void search_returns200_whenAuthenticated() throws Exception {
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search"))
@@ -88,7 +87,7 @@ class DocumentControllerTest {
     void search_undatedTrue_isReachableByAuthenticatedUser() throws Exception {
         // The read GET must stay reachable for READ_ALL users — guards against a
         // future refactor accidentally write-guarding the undated triage path (#668).
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search").param("undated", "true"))
@@ -104,41 +103,43 @@ class DocumentControllerTest {
     @Test
     @WithMockUser
     void search_undatedTrue_isForwardedToServiceAsTrue() throws Exception {
-        ArgumentCaptor<Boolean> undatedCaptor = ArgumentCaptor.forClass(Boolean.class);
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        ArgumentCaptor<SearchFilters> filtersCaptor = ArgumentCaptor.forClass(SearchFilters.class);
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search").param("undated", "true"))
                 .andExpect(status().isOk());
 
-        verify(documentService).searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), undatedCaptor.capture(), any());
-        assertThat(undatedCaptor.getValue()).isTrue();
+        verify(documentService).searchDocuments(filtersCaptor.capture(), any(), any(), any());
+        assertThat(filtersCaptor.getValue().undated()).isTrue();
     }
 
     @Test
     @WithMockUser
     void search_withoutUndatedParam_forwardsFalseToService() throws Exception {
-        ArgumentCaptor<Boolean> undatedCaptor = ArgumentCaptor.forClass(Boolean.class);
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        ArgumentCaptor<SearchFilters> filtersCaptor = ArgumentCaptor.forClass(SearchFilters.class);
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search"))
                 .andExpect(status().isOk());
 
-        verify(documentService).searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), undatedCaptor.capture(), any());
-        assertThat(undatedCaptor.getValue()).isFalse();
+        verify(documentService).searchDocuments(filtersCaptor.capture(), any(), any(), any());
+        assertThat(filtersCaptor.getValue().undated()).isFalse();
     }
 
     @Test
     @WithMockUser
     void search_withStatusParam_passesItToService() throws Exception {
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), eq(DocumentStatus.REVIEWED), any(), any(), any(), anyBoolean(), any()))
+        ArgumentCaptor<SearchFilters> filtersCaptor = ArgumentCaptor.forClass(SearchFilters.class);
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search").param("status", "REVIEWED"))
                 .andExpect(status().isOk());
 
-        verify(documentService).searchDocuments(any(), any(), any(), any(), any(), any(), any(), eq(DocumentStatus.REVIEWED), any(), any(), any(), anyBoolean(), any());
+        verify(documentService).searchDocuments(filtersCaptor.capture(), any(), any(), any());
+        assertThat(filtersCaptor.getValue().status()).isEqualTo(DocumentStatus.REVIEWED);
     }
 
     @Test
@@ -165,7 +166,7 @@ class DocumentControllerTest {
     @Test
     @WithMockUser
     void search_responseContainsTotalCount() throws Exception {
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search"))
@@ -180,12 +181,13 @@ class DocumentControllerTest {
         UUID docId = UUID.randomUUID();
         var matchData = new SearchMatchData(
                 "Er schrieb einen langen Brief", List.of(), false, List.of(), List.of(), List.of(), null, List.of());
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of(new DocumentListItem(
                         docId, "Brief an Anna", "brief.pdf", null, null,
                         DatePrecision.UNKNOWN, null, null,
                         List.of(), List.of(), null, null, null, null,
-                        0, List.of(), matchData))));
+                        0, List.of(), matchData,
+                        LocalDateTime.of(2026, 1, 15, 10, 0), LocalDateTime.of(2026, 1, 15, 10, 0)))));
 
         mockMvc.perform(get("/api/documents/search").param("q", "Brief"))
                 .andExpect(status().isOk())
@@ -199,12 +201,13 @@ class DocumentControllerTest {
     void search_returns_flat_item_with_id_and_without_sensitive_fields() throws Exception {
         UUID docId = UUID.randomUUID();
         var matchData = new SearchMatchData(null, List.of(), false, List.of(), List.of(), List.of(), null, List.of());
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of(new DocumentListItem(
                         docId, "Brief an Anna", "brief.pdf", null, null,
                         DatePrecision.UNKNOWN, null, null,
                         List.of(), List.of(), null, null, null, null,
-                        0, List.of(), matchData))));
+                        0, List.of(), matchData,
+                        LocalDateTime.of(2026, 1, 15, 10, 0), LocalDateTime.of(2026, 1, 15, 10, 0)))));
 
         mockMvc.perform(get("/api/documents/search"))
                 .andExpect(status().isOk())
@@ -221,7 +224,7 @@ class DocumentControllerTest {
     @Test
     @WithMockUser
     void search_responseExposesPagingFields() throws Exception {
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search"))
@@ -266,7 +269,7 @@ class DocumentControllerTest {
     @Test
     @WithMockUser
     void search_passesPageRequestToService() throws Exception {
-        when(documentService.searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), any()))
+        when(documentService.searchDocuments(any(), any(), any(), any()))
                 .thenReturn(DocumentSearchResult.of(List.of()));
 
         mockMvc.perform(get("/api/documents/search").param("page", "2").param("size", "25"))
@@ -274,7 +277,7 @@ class DocumentControllerTest {
 
         org.mockito.ArgumentCaptor<org.springframework.data.domain.Pageable> captor =
                 org.mockito.ArgumentCaptor.forClass(org.springframework.data.domain.Pageable.class);
-        verify(documentService).searchDocuments(any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean(), captor.capture());
+        verify(documentService).searchDocuments(any(), any(), any(), captor.capture());
         org.springframework.data.domain.Pageable pageable = captor.getValue();
         org.assertj.core.api.Assertions.assertThat(pageable.getPageNumber()).isEqualTo(2);
         org.assertj.core.api.Assertions.assertThat(pageable.getPageSize()).isEqualTo(25);
@@ -295,6 +298,13 @@ class DocumentControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void createDocument_returns403_forReaderOnly() throws Exception {
+        mockMvc.perform(multipart("/api/documents").with(csrf()))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void createDocument_returns200_whenHasWritePermission() throws Exception {
@@ -412,6 +422,13 @@ class DocumentControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void quickUpload_returns403_forReaderOnly() throws Exception {
+        mockMvc.perform(multipart("/api/documents/quick-upload").with(csrf()))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void quickUpload_returns200_withValidPdfFile() throws Exception {
@@ -1192,7 +1209,7 @@ class DocumentControllerTest {
     void getDocumentIds_returns200_andDelegatesToService() throws Exception {
         when(userService.findByEmail(any())).thenReturn(AppUser.builder().id(UUID.randomUUID()).build());
         UUID id = UUID.randomUUID();
-        when(documentService.findIdsForFilter(any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean()))
+        when(documentService.findIdsForFilter(any()))
                 .thenReturn(List.of(id));
 
         mockMvc.perform(get("/api/documents/ids"))
@@ -1205,13 +1222,33 @@ class DocumentControllerTest {
     void getDocumentIds_passesSenderIdParamToService() throws Exception {
         when(userService.findByEmail(any())).thenReturn(AppUser.builder().id(UUID.randomUUID()).build());
         UUID senderId = UUID.randomUUID();
-        when(documentService.findIdsForFilter(any(), any(), any(), eq(senderId), any(), any(), any(), any(), any(), anyBoolean()))
+        ArgumentCaptor<SearchFilters> filtersCaptor = ArgumentCaptor.forClass(SearchFilters.class);
+        when(documentService.findIdsForFilter(any()))
                 .thenReturn(List.of());
 
         mockMvc.perform(get("/api/documents/ids").param("senderId", senderId.toString()))
                 .andExpect(status().isOk());
 
-        verify(documentService).findIdsForFilter(any(), any(), any(), eq(senderId), any(), any(), any(), any(), any(), anyBoolean());
+        verify(documentService).findIdsForFilter(filtersCaptor.capture());
+        assertThat(filtersCaptor.getValue().sender()).isEqualTo(senderId);
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void getDocumentIds_withoutUndatedParam_coercesNullToFalse() throws Exception {
+        // The controller coerces a null boxed Boolean to primitive false
+        // (Boolean.TRUE.equals(undated)) so the absent param never NPEs and the
+        // record always holds a concrete boolean.
+        when(userService.findByEmail(any())).thenReturn(AppUser.builder().id(UUID.randomUUID()).build());
+        ArgumentCaptor<SearchFilters> filtersCaptor = ArgumentCaptor.forClass(SearchFilters.class);
+        when(documentService.findIdsForFilter(any()))
+                .thenReturn(List.of());
+
+        mockMvc.perform(get("/api/documents/ids"))
+                .andExpect(status().isOk());
+
+        verify(documentService).findIdsForFilter(filtersCaptor.capture());
+        assertThat(filtersCaptor.getValue().undated()).isFalse();
     }
 
     @Test
@@ -1221,7 +1258,7 @@ class DocumentControllerTest {
         // Service returns 5001 IDs — one over BULK_EDIT_FILTER_MAX_IDS (5000).
         java.util.List<UUID> tooMany = new java.util.ArrayList<>(5001);
         for (int i = 0; i < 5001; i++) tooMany.add(UUID.randomUUID());
-        when(documentService.findIdsForFilter(any(), any(), any(), any(), any(), any(), any(), any(), any(), anyBoolean()))
+        when(documentService.findIdsForFilter(any()))
                 .thenReturn(tooMany);
 
         mockMvc.perform(get("/api/documents/ids"))
@@ -1386,16 +1423,16 @@ class DocumentControllerTest {
 
     @Test
     @WithMockUser
-    void density_emitsPrivateCacheControlHeader() throws Exception {
+    void density_isNeverBrowserCached() throws Exception {
         when(documentService.getDensity(any())).thenReturn(
                 new DocumentDensityResult(List.of(), null, null));
 
+        // The endpoint sets no explicit Cache-Control, so Spring Security's
+        // default no-store directive applies — the density chart is always fresh.
         mockMvc.perform(get("/api/documents/density"))
                 .andExpect(status().isOk())
                 .andExpect(header().string("Cache-Control",
-                        org.hamcrest.Matchers.containsString("max-age=300")))
-                .andExpect(header().string("Cache-Control",
-                        org.hamcrest.Matchers.containsString("private")));
+                        "no-cache, no-store, max-age=0, must-revalidate"));
     }
 
     @Test

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

@@ -24,6 +24,7 @@ import java.util.Set;
 import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
+import static org.raddatz.familienarchiv.document.SearchFiltersFixtures.noFilters;
 import static org.assertj.core.api.Assertions.assertThatCode;
 import static org.mockito.ArgumentMatchers.any;
 import static org.mockito.Mockito.when;
@@ -122,8 +123,8 @@ class DocumentLazyLoadingTest {
         savedDocument("SrDoc", "sr_doc.pdf", sender, Set.of(receiver), Set.of(tag));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.RECEIVER, "asc", null, false, PageRequest.of(0, 20));
+                noFilters(),
+                DocumentSort.RECEIVER, "asc", PageRequest.of(0, 20));
         assertThat(result.totalElements()).isGreaterThan(0);
         assertThatCode(() ->
                 result.items().forEach(i -> { if (i.sender() != null) i.sender().getLastName(); }))
@@ -137,8 +138,8 @@ class DocumentLazyLoadingTest {
         savedDocument("SsDoc", "ss_doc.pdf", sender, Set.of(), Set.of(tag));
 
         assertThatCode(() -> documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.SENDER, "asc", null, false, PageRequest.of(0, 20)))
+                noFilters(),
+                DocumentSort.SENDER, "asc", PageRequest.of(0, 20)))
                 .doesNotThrowAnyException();
     }
 

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

@@ -17,6 +17,7 @@ import java.util.HashSet;
 import java.util.Set;
 
 import static org.assertj.core.api.Assertions.assertThat;
+import static org.raddatz.familienarchiv.document.SearchFiltersFixtures.noFilters;
 import static org.assertj.core.api.Assertions.assertThatCode;
 
 /**
@@ -55,8 +56,8 @@ class DocumentListItemIntegrationTest {
                 .build());
 
         assertThatCode(() -> documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50)))
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50)))
             .doesNotThrowAnyException();
     }
 
@@ -70,8 +71,8 @@ class DocumentListItemIntegrationTest {
                 .build());
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
 
         assertThat(result.totalElements()).isGreaterThan(0);
         DocumentListItem item = result.items().get(0);
@@ -91,8 +92,8 @@ class DocumentListItemIntegrationTest {
                 .build());
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
 
         DocumentListItem item = result.items().stream()
                 .filter(i -> i.title().equals("Range Brief")).findFirst().orElseThrow();

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

@@ -38,7 +38,10 @@ import java.util.Optional;
 import java.util.Set;
 import java.util.UUID;
 
+import org.springframework.dao.DataIntegrityViolationException;
+
 import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
 
 @DataJpaTest
 @AutoConfigureTestDatabase(replace = AutoConfigureTestDatabase.Replace.NONE)
@@ -259,67 +262,6 @@ class DocumentRepositoryTest {
         assertThat(result.getContent()).allMatch(d -> !d.isMetadataComplete());
     }
 
-    // ─── findSinglePersonCorrespondence — DISTINCT / multi-receiver safety ────
-
-    @Test
-    void findSinglePersonCorrespondence_returnsExactlyOneResult_whenDocumentHasThreeReceiversAndOneMatchesPersonId() {
-        Person sender = personRepository.save(Person.builder()
-                .firstName("Hans").lastName("Müller").build());
-        Person receiver1 = personRepository.save(Person.builder()
-                .firstName("Anna").lastName("Schmidt").build());
-        Person receiver2 = personRepository.save(Person.builder()
-                .firstName("Bertha").lastName("Wagner").build());
-        Person receiver3 = personRepository.save(Person.builder()
-                .firstName("Clara").lastName("Koch").build());
-
-        // Document addressed to all three receivers
-        Document doc = documentRepository.save(Document.builder()
-                .title("Rundschreiben")
-                .originalFilename("rundschreiben.pdf")
-                .status(DocumentStatus.UPLOADED)
-                .sender(sender)
-                .receivers(new HashSet<>(Set.of(receiver1, receiver2, receiver3)))
-                .documentDate(LocalDate.of(1950, 6, 1))
-                .build());
-
-        Sort sort = Sort.by(Sort.Direction.DESC, "documentDate");
-        LocalDate from = LocalDate.of(1900, 1, 1);
-        LocalDate to = LocalDate.of(2000, 1, 1);
-
-        // Query for receiver1 — the DISTINCT must collapse the 3 JOIN rows into 1 result
-        List<Document> results = documentRepository.findSinglePersonCorrespondence(
-                receiver1.getId(), from, to, sort);
-
-        assertThat(results).hasSize(1);
-        assertThat(results.get(0).getId()).isEqualTo(doc.getId());
-    }
-
-    @Test
-    void findSinglePersonCorrespondence_includesDocumentsWherePerson_isSender() {
-        Person sender = personRepository.save(Person.builder()
-                .firstName("Hans").lastName("Müller").build());
-        Person receiver = personRepository.save(Person.builder()
-                .firstName("Anna").lastName("Schmidt").build());
-
-        documentRepository.save(Document.builder()
-                .title("Brief als Absender")
-                .originalFilename("brief_absender.pdf")
-                .status(DocumentStatus.UPLOADED)
-                .sender(sender)
-                .receivers(new HashSet<>(Set.of(receiver)))
-                .documentDate(LocalDate.of(1950, 6, 1))
-                .build());
-
-        Sort sort = Sort.by(Sort.Direction.DESC, "documentDate");
-        LocalDate from = LocalDate.of(1900, 1, 1);
-        LocalDate to = LocalDate.of(2000, 1, 1);
-
-        List<Document> results = documentRepository.findSinglePersonCorrespondence(
-                sender.getId(), from, to, sort);
-
-        assertThat(results).hasSize(1);
-    }
-
     // ─── findSegmentationQueue ────────────────────────────────────────────────
 
     @Test
@@ -612,6 +554,48 @@ class DocumentRepositoryTest {
                 .isLessThanOrEqualTo(5);
     }
 
+    // ─── V69 date-range CHECK constraints (#678) ──────────────────────────────
+
+    @Test
+    void save_acceptsRange_whenEndEqualsStart() {
+        // chk_meta_date_end_after_start is end >= start, so equal dates are valid.
+        // Real Postgres + Flyway here (H2 would not enforce the CHECK) pins the
+        // app guard's isBefore semantics to the actual constraint — guards drift (AC2).
+        LocalDate day = LocalDate.of(1917, 1, 10);
+        Document saved = documentRepository.saveAndFlush(Document.builder()
+                .title("Gleicher Tag")
+                .originalFilename("gleicher_tag.pdf")
+                .status(DocumentStatus.UPLOADED)
+                .documentDate(day)
+                .metaDatePrecision(DatePrecision.RANGE)
+                .metaDateEnd(day)
+                .build());
+
+        Document found = documentRepository.findById(saved.getId()).orElseThrow();
+        assertThat(found.getDocumentDate()).isEqualTo(day);
+        assertThat(found.getMetaDateEnd()).isEqualTo(day);
+        assertThat(found.getMetaDatePrecision()).isEqualTo(DatePrecision.RANGE);
+    }
+
+    @Test
+    void save_rejectsRange_whenEndBeforeStart_atDbLevel() {
+        // The app guard normally intercepts this, so the DB CHECK never fires in practice.
+        // Persisting directly proves chk_meta_date_end_after_start actually rejects end < start
+        // (H2 would not) — if the app guard ever regresses, a bad row still can't reach the table,
+        // and this is exactly the violation the GlobalExceptionHandler backstop turns into a 400.
+        Document doc = Document.builder()
+                .title("Verdrehte Spanne")
+                .originalFilename("verdreht.pdf")
+                .status(DocumentStatus.UPLOADED)
+                .documentDate(LocalDate.of(1917, 1, 11))
+                .metaDatePrecision(DatePrecision.RANGE)
+                .metaDateEnd(LocalDate.of(1917, 1, 10))
+                .build();
+
+        assertThatThrownBy(() -> documentRepository.saveAndFlush(doc))
+                .isInstanceOf(DataIntegrityViolationException.class);
+    }
+
     // ─── seeding helpers ─────────────────────────────────────────────────────
 
     private Document uploaded(String title) {
@@ -640,4 +624,88 @@ class DocumentRepositoryTest {
                 .reviewed(reviewed)
                 .build();
     }
+
+    // ─── searchDocumentsByPersonId (via Specification) ───────────────────────
+
+    private Page<Document> searchByPerson(Person person, LocalDate from, LocalDate to) {
+        Specification<Document> spec = (root, query, cb) -> {
+            if (query != null) query.distinct(true);
+            var receiversJoin = root.join("receivers", jakarta.persistence.criteria.JoinType.LEFT);
+            var personPredicate = cb.or(
+                    cb.equal(root.get("sender"), person),
+                    cb.equal(receiversJoin, person));
+            var predicates = new java.util.ArrayList<>(java.util.List.of(personPredicate));
+            if (from != null) predicates.add(cb.greaterThanOrEqualTo(root.get("documentDate"), from));
+            if (to != null) predicates.add(cb.lessThanOrEqualTo(root.get("documentDate"), to));
+            return cb.and(predicates.toArray(new jakarta.persistence.criteria.Predicate[0]));
+        };
+        return documentRepository.findAll(spec, PageRequest.of(0, 10));
+    }
+
+    @Test
+    void searchByPersonSpec_returnsDocument_whenPersonIsSender() {
+        Person person = personRepository.save(Person.builder().lastName("Raddatz").build());
+        Document doc = documentRepository.save(Document.builder()
+                .title("Senderbrief").originalFilename("sender.pdf")
+                .status(DocumentStatus.UPLOADED).sender(person).build());
+
+        Page<Document> result = searchByPerson(person, null, null);
+
+        assertThat(result.getContent()).extracting(Document::getId).containsExactly(doc.getId());
+    }
+
+    @Test
+    void searchByPersonSpec_returnsDocument_whenPersonIsReceiver() {
+        Person person = personRepository.save(Person.builder().lastName("Raddatz").build());
+        Document doc = documentRepository.save(Document.builder()
+                .title("Empfängerbrief").originalFilename("receiver.pdf")
+                .status(DocumentStatus.UPLOADED)
+                .receivers(new java.util.HashSet<>(List.of(person))).build());
+
+        Page<Document> result = searchByPerson(person, null, null);
+
+        assertThat(result.getContent()).extracting(Document::getId).containsExactly(doc.getId());
+    }
+
+    @Test
+    void searchByPersonSpec_returnsDocumentOnce_whenPersonIsBothSenderAndReceiver() {
+        Person person = personRepository.save(Person.builder().lastName("Raddatz").build());
+        Document doc = documentRepository.save(Document.builder()
+                .title("SenderEmpfänger").originalFilename("both.pdf")
+                .status(DocumentStatus.UPLOADED).sender(person)
+                .receivers(new java.util.HashSet<>(List.of(person))).build());
+
+        Page<Document> result = searchByPerson(person, null, null);
+
+        assertThat(result.getContent()).hasSize(1);
+        assertThat(result.getContent().get(0).getId()).isEqualTo(doc.getId());
+    }
+
+    @Test
+    void searchByPersonSpec_excludesDocuments_outsideDateRange() {
+        Person person = personRepository.save(Person.builder().lastName("Raddatz").build());
+        Document inside = documentRepository.save(Document.builder()
+                .title("Innen").originalFilename("inside.pdf").status(DocumentStatus.UPLOADED)
+                .sender(person).documentDate(LocalDate.of(1918, 6, 15)).build());
+        documentRepository.save(Document.builder()
+                .title("Außen").originalFilename("outside.pdf").status(DocumentStatus.UPLOADED)
+                .sender(person).documentDate(LocalDate.of(1920, 1, 1)).build());
+
+        Page<Document> result = searchByPerson(person, LocalDate.of(1914, 1, 1), LocalDate.of(1918, 12, 31));
+
+        assertThat(result.getContent()).extracting(Document::getId).containsExactly(inside.getId());
+    }
+
+    @Test
+    void searchByPersonSpec_returnsEmpty_whenNoMatchingDocuments() {
+        Person person = personRepository.save(Person.builder().lastName("Raddatz").build());
+        Person other = personRepository.save(Person.builder().lastName("Braun").build());
+        documentRepository.save(Document.builder()
+                .title("Fremder Brief").originalFilename("other.pdf")
+                .status(DocumentStatus.UPLOADED).sender(other).build());
+
+        Page<Document> result = searchByPerson(person, null, null);
+
+        assertThat(result.getContent()).isEmpty();
+    }
 }

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

@@ -21,6 +21,7 @@ import java.time.LocalDate;
 import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
+import static org.raddatz.familienarchiv.document.SearchFiltersFixtures.noFilters;
 
 /**
  * End-to-end paged search test with real PostgreSQL (Testcontainers). Covers the
@@ -61,8 +62,8 @@ class DocumentSearchPagedIntegrationTest {
     @Test
     void search_firstPage_returnsExactlyPageSizeItems_andCorrectTotalElements() {
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
 
         assertThat(result.items()).hasSize(50);
         assertThat(result.totalElements()).isEqualTo(FIXTURE_SIZE);
@@ -74,8 +75,8 @@ class DocumentSearchPagedIntegrationTest {
     @Test
     void search_lastPartialPage_returnsRemainingItems() {
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(2, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(2, 50));
 
         // Page 2 (offset 100) of 120 docs → exactly 20 items on the tail.
         assertThat(result.items()).hasSize(20);
@@ -86,8 +87,8 @@ class DocumentSearchPagedIntegrationTest {
     @Test
     void search_pageBeyondLast_returnsEmptyContent_totalElementsStillCorrect() {
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(99, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(99, 50));
 
         assertThat(result.items()).isEmpty();
         assertThat(result.totalElements()).isEqualTo(FIXTURE_SIZE);
@@ -99,8 +100,8 @@ class DocumentSearchPagedIntegrationTest {
         // comment in DocumentService). Proves that the in-memory slice path
         // returns the correct total from a real repository fetch.
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.SENDER, "asc", null, false, PageRequest.of(1, 50));
+                noFilters(),
+                DocumentSort.SENDER, "asc", PageRequest.of(1, 50));
 
         assertThat(result.items()).hasSize(50);
         assertThat(result.totalElements()).isEqualTo(FIXTURE_SIZE);
@@ -125,8 +126,8 @@ class DocumentSearchPagedIntegrationTest {
         }
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
 
         // Global undated count is the full undated total, independent of page size.
         assertThat(result.undatedCount()).isEqualTo(undatedTotal);
@@ -153,11 +154,11 @@ class DocumentSearchPagedIntegrationTest {
         }
 
         DocumentSearchResult unfiltered = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
         DocumentSearchResult undatedOnly = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, true, PageRequest.of(0, 50));
+                noFilters().withUndated(true),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
 
         assertThat(unfiltered.undatedCount()).isEqualTo(undatedTotal);
         assertThat(undatedOnly.undatedCount()).isEqualTo(undatedTotal);
@@ -178,9 +179,9 @@ class DocumentSearchPagedIntegrationTest {
         }
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, LocalDate.of(1900, 1, 1), LocalDate.of(2000, 12, 31),
-                null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50));
+                new SearchFilters(null, LocalDate.of(1900, 1, 1), LocalDate.of(2000, 12, 31),
+                        null, null, null, null, null, null, false),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
 
         assertThat(result.undatedCount()).isZero();
     }
@@ -188,11 +189,11 @@ class DocumentSearchPagedIntegrationTest {
     @Test
     void search_differentPagesReturnDisjointSlices() {
         DocumentSearchResult page0 = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(0, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(0, 50));
         DocumentSearchResult page1 = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, PageRequest.of(1, 50));
+                noFilters(),
+                DocumentSort.DATE, "DESC", PageRequest.of(1, 50));
 
         // No document id should appear on both pages — slicing must be exclusive.
         var idsOnPage0 = page0.items().stream()

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

@@ -5,6 +5,7 @@ import org.junit.jupiter.api.Test;
 import org.raddatz.familienarchiv.audit.ActivityActorDTO;
 import org.springframework.data.domain.PageRequest;
 
+import java.time.LocalDateTime;
 import java.util.List;
 import java.util.UUID;
 
@@ -17,7 +18,8 @@ class DocumentSearchResultTest {
                 docId, "Test", "test.pdf", null, null,
                 DatePrecision.UNKNOWN, null, null,
                 List.of(), List.of(), null, null, null, null,
-                0, List.of(), SearchMatchData.empty());
+                0, List.of(), SearchMatchData.empty(),
+                LocalDateTime.of(2026, 1, 15, 10, 0), LocalDateTime.of(2026, 1, 15, 10, 0));
     }
 
     @Test
@@ -68,7 +70,8 @@ class DocumentSearchResultTest {
                 id, "T", "t.pdf", null, null,
                 DatePrecision.UNKNOWN, null, null,
                 List.of(), List.of(), null, null, null, null,
-                75, List.of(actor), SearchMatchData.empty());
+                75, List.of(actor), SearchMatchData.empty(),
+                LocalDateTime.of(2026, 1, 15, 10, 0), LocalDateTime.of(2026, 1, 15, 10, 0));
 
         DocumentSearchResult result = DocumentSearchResult.of(List.of(item));
 

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

@@ -67,7 +67,8 @@ class DocumentServiceSortTest {
                 .thenReturn(new PageImpl<>(List.of(newer, older)));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                "Brief", null, null, null, null, null, null, null, DocumentSort.DATE, "DESC", null, false, PAGE);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                DocumentSort.DATE, "DESC", PAGE);
 
         assertThat(result.items()).hasSize(2);
         assertThat(result.items().get(0).id()).isEqualTo(id2);  // newer first
@@ -84,7 +85,8 @@ class DocumentServiceSortTest {
                 .thenReturn(List.of(doc(id1)));
 
         documentService.searchDocuments(
-                "Brief", null, null, null, null, null, null, null, DocumentSort.RELEVANCE, null, null, false, PAGE);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                DocumentSort.RELEVANCE, null, PAGE);
 
         verify(documentRepository).findFtsPageRaw(anyString(), anyInt(), anyInt());
         verify(documentRepository, never()).findAllMatchingIdsByFts(anyString());
@@ -102,7 +104,8 @@ class DocumentServiceSortTest {
         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, false, PAGE);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                DocumentSort.RELEVANCE, null, PAGE);
 
         assertThat(result.items().get(0).id()).isEqualTo(id1);
     }
@@ -119,7 +122,8 @@ class DocumentServiceSortTest {
         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, false, PAGE);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                null, null, PAGE);
 
         assertThat(result.items().get(0).id()).isEqualTo(id1);
     }
@@ -132,8 +136,8 @@ class DocumentServiceSortTest {
         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, false, hugePage);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                DocumentSort.RELEVANCE, null, hugePage);
 
         assertThat(result.items()).isEmpty();
         verify(documentRepository, never()).findFtsPageRaw(anyString(), anyInt(), anyInt());
@@ -152,8 +156,8 @@ class DocumentServiceSortTest {
         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, false, PAGE);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                DocumentSort.RELEVANCE, null, PAGE);
 
         assertThat(result.items()).hasSize(1);
         assertThat(result.items().get(0).id()).isEqualTo(uuidId);
@@ -173,7 +177,8 @@ class DocumentServiceSortTest {
         // 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, false, PAGE);
+                new SearchFilters("Brief", from, null, null, null, null, null, null, null, false),
+                DocumentSort.RELEVANCE, null, PAGE);
 
         verify(documentRepository, never()).findFtsPageRaw(anyString(), anyInt(), anyInt());
         verify(documentRepository).findAllMatchingIdsByFts("Brief");

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

@@ -5,6 +5,7 @@ import org.junit.jupiter.api.extension.ExtendWith;
 import org.mockito.ArgumentCaptor;
 import org.mockito.InjectMocks;
 import org.mockito.Mock;
+import org.mockito.Spy;
 import org.mockito.junit.jupiter.MockitoExtension;
 import org.raddatz.familienarchiv.audit.AuditKind;
 import org.raddatz.familienarchiv.audit.AuditLogQueryService;
@@ -20,6 +21,7 @@ import org.raddatz.familienarchiv.document.MatchOffset;
 import org.raddatz.familienarchiv.document.SearchMatchData;
 import org.raddatz.familienarchiv.tag.TagOperator;
 import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
 import org.raddatz.familienarchiv.document.Document;
 import org.raddatz.familienarchiv.document.DocumentStatus;
 import org.raddatz.familienarchiv.person.Person;
@@ -45,6 +47,7 @@ import java.util.Set;
 import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
+import static org.raddatz.familienarchiv.document.SearchFiltersFixtures.noFilters;
 import static org.assertj.core.api.Assertions.assertThatThrownBy;
 import static org.mockito.ArgumentMatchers.any;
 import static org.mockito.ArgumentMatchers.anyInt;
@@ -72,6 +75,9 @@ class DocumentServiceTest {
     @Mock AuditLogQueryService auditLogQueryService;
     @Mock TranscriptionBlockQueryService transcriptionBlockQueryService;
     @Mock ThumbnailAsyncRunner thumbnailAsyncRunner;
+    // Real factory (pure, dependency-free) so save-time title-regeneration tests exercise the
+    // shared composition rather than a stub — the #726 single source of truth.
+    @Spy DocumentTitleFactory documentTitleFactory = new DocumentTitleFactory();
     @InjectMocks DocumentService documentService;
 
     // ─── deleteDocument ───────────────────────────────────────────────────────
@@ -118,6 +124,37 @@ class DocumentServiceTest {
         assertThat(documentService.getDocumentById(id)).isEqualTo(doc);
     }
 
+    @Test
+    void getDocumentById_doesNotQueryTranscription() {
+        UUID id = UUID.randomUUID();
+        Document doc = Document.builder().id(id).title("Test").build();
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+
+        documentService.getDocumentById(id);
+
+        verifyNoInteractions(transcriptionBlockQueryService);
+    }
+
+    @Test
+    void getDocumentDetail_setsHasTranscriptionTrue_whenBlocksExist() {
+        UUID id = UUID.randomUUID();
+        Document doc = Document.builder().id(id).title("Test").build();
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+        when(transcriptionBlockQueryService.hasBlocks(id)).thenReturn(true);
+
+        assertThat(documentService.getDocumentDetail(id).isHasTranscription()).isTrue();
+    }
+
+    @Test
+    void getDocumentDetail_setsHasTranscriptionFalse_whenNoBlocksExist() {
+        UUID id = UUID.randomUUID();
+        Document doc = Document.builder().id(id).title("Test").build();
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+        when(transcriptionBlockQueryService.hasBlocks(id)).thenReturn(false);
+
+        assertThat(documentService.getDocumentDetail(id).isHasTranscription()).isFalse();
+    }
+
     // ─── updateDocument ───────────────────────────────────────────────────────
 
     @Test
@@ -171,10 +208,12 @@ class DocumentServiceTest {
         // Editing a doc (e.g. fixing a location typo) without touching the precision
         // controls must NOT fabricate a precision. The form omits the three precision
         // fields → they arrive null on the DTO → the stored values must be preserved.
+        // Stored combo is RANGE + end: the only DB-valid way to have a non-null end
+        // (chk_meta_date_end_only_for_range), so the carried-over state passes the guard.
         UUID id = UUID.randomUUID();
         Document doc = Document.builder()
                 .id(id)
-                .metaDatePrecision(DatePrecision.MONTH)
+                .metaDatePrecision(DatePrecision.RANGE)
                 .metaDateEnd(LocalDate.of(1916, 6, 30))
                 .metaDateRaw("Juni 1916")
                 .receivers(new HashSet<>())
@@ -188,11 +227,329 @@ class DocumentServiceTest {
 
         documentService.updateDocument(id, dto, null, null);
 
-        assertThat(doc.getMetaDatePrecision()).isEqualTo(DatePrecision.MONTH);
+        assertThat(doc.getMetaDatePrecision()).isEqualTo(DatePrecision.RANGE);
         assertThat(doc.getMetaDateEnd()).isEqualTo(LocalDate.of(1916, 6, 30));
         assertThat(doc.getMetaDateRaw()).isEqualTo("Juni 1916");
     }
 
+    // ─── updateDocument save-time auto-title regeneration (#726) ──────────────
+    //
+    // Exact old-vs-new comparison: the title is the catalog auto-title iff the submitted
+    // title equals what the factory builds from the CURRENTLY-persisted state. The edit form
+    // round-trips the stored title verbatim when untouched, so an equal submission means the
+    // user did not type over it. makeStored() seeds index/date/precision/location and sets the
+    // stored title to the matching auto-title, mirroring a freshly-imported row.
+
+    private Document makeStored(String index, LocalDate date, DatePrecision precision, String location) {
+        Document doc = Document.builder()
+                .id(UUID.randomUUID())
+                .originalFilename(index)
+                .documentDate(date)
+                .metaDatePrecision(precision)
+                .location(location)
+                .receivers(new HashSet<>())
+                .tags(new HashSet<>())
+                .build();
+        doc.setTitle(documentTitleFactory.build(doc));
+        return doc;
+    }
+
+    /** A DTO that round-trips the stored auto-title untouched, with new date/precision/location. */
+    private static DocumentUpdateDTO editDto(String submittedTitle, LocalDate date,
+                                             DatePrecision precision, String location) {
+        DocumentUpdateDTO dto = new DocumentUpdateDTO();
+        dto.setTitle(submittedTitle);
+        dto.setDocumentDate(date);
+        dto.setMetaDatePrecision(precision);
+        dto.setLocation(location);
+        return dto;
+    }
+
+    private Document runUpdate(Document stored, DocumentUpdateDTO dto) throws Exception {
+        when(documentRepository.findById(stored.getId())).thenReturn(Optional.of(stored));
+        when(documentRepository.save(any())).thenReturn(stored);
+        documentService.updateDocument(stored.getId(), dto, null, null);
+        return stored;
+    }
+
+    @Test
+    void updateDocument_regeneratesAutoTitle_whenDateChanges() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(2028, 1, 1), DatePrecision.YEAR, "Berlin");
+        // title untouched ("C-0029 – 2028 – Berlin"), date corrected to 1928
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – 1928 – Berlin");
+    }
+
+    @Test
+    void updateDocument_keepsHandWrittenTitle_whenDateChanges() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, null);
+        stored.setTitle("C-0029 – Brief an Mutter"); // hand-written, ≠ auto-title
+        DocumentUpdateDTO dto = editDto("C-0029 – Brief an Mutter", LocalDate.of(1930, 1, 1), DatePrecision.YEAR, null);
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – Brief an Mutter");
+    }
+
+    @Test
+    void updateDocument_freshlyTypedTitleWins_overRegeneration() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(2028, 1, 1), DatePrecision.YEAR, "Berlin");
+        // user changed the date AND typed a new title in the same save
+        DocumentUpdateDTO dto = editDto("Geburtsanzeige", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("Geburtsanzeige");
+    }
+
+    @Test
+    void updateDocument_regeneratesWithNewDateAndLocation() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(2028, 1, 1), DatePrecision.YEAR, "Berlin");
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "München");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – 1928 – München");
+    }
+
+    @Test
+    void updateDocument_dropsTrailingLocationSegment_whenLocationCleared() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+        // location cleared (null), title untouched
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1928, 1, 1), DatePrecision.YEAR, null);
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – 1928");
+    }
+
+    @Test
+    void updateDocument_regeneratedTitle_doesNotContainOldDate() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(2028, 1, 1), DatePrecision.YEAR, "Berlin");
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).doesNotContain("2028");
+    }
+
+    @Test
+    void updateDocument_relabelsOnPrecisionChange_yearToDay() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, null);
+        // stored auto-title "C-0029 – 1928"; set a full day at DAY precision
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1928, 1, 15), DatePrecision.DAY, null);
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – 15. Januar 1928");
+    }
+
+    @Test
+    void updateDocument_populatesTitle_whenDateAddedToUnknownRow() throws Exception {
+        Document stored = makeStored("C-0029", null, DatePrecision.UNKNOWN, null);
+        // stored auto-title is just "C-0029"; add a 1928 YEAR date
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1928, 1, 1), DatePrecision.YEAR, null);
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – 1928");
+    }
+
+    @Test
+    void updateDocument_roundTripsSeasonLabel() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(1943, 4, 1), DatePrecision.SEASON, null);
+        stored.setMetaDateRaw("Frühling 1943");
+        stored.setTitle(documentTitleFactory.build(stored)); // "C-0029 – Frühling 1943"
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1943, 4, 1), DatePrecision.SEASON, null);
+        dto.setMetaDateRaw("Frühling 1943");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – Frühling 1943");
+    }
+
+    @Test
+    void updateDocument_carriesStoredPrecisionAndRaw_whenDtoOmitsThem() throws Exception {
+        // Only the year changes; precision/end/raw are omitted from the DTO, so projectedState
+        // must carry them from the entity (exercises the skip-null effective* resolvers).
+        Document stored = makeStored("C-0029", LocalDate.of(1943, 4, 1), DatePrecision.SEASON, null);
+        stored.setMetaDateRaw("Frühling 1943");
+        stored.setTitle(documentTitleFactory.build(stored)); // "C-0029 – Frühling 1943"
+        DocumentUpdateDTO dto = editDto(stored.getTitle(), LocalDate.of(1944, 4, 1), null, null);
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – Frühling 1944");
+    }
+
+    @Test
+    void updateDocument_roundTripsRangeLabel_atSaveTime() throws Exception {
+        Document stored = Document.builder()
+                .id(UUID.randomUUID())
+                .originalFilename("C-0029")
+                .documentDate(LocalDate.of(1917, 1, 10))
+                .metaDatePrecision(DatePrecision.RANGE)
+                .metaDateEnd(LocalDate.of(1917, 1, 11))
+                .receivers(new HashSet<>())
+                .tags(new HashSet<>())
+                .build();
+        stored.setTitle(documentTitleFactory.build(stored)); // "C-0029 – 10.–11. Jan. 1917"
+        DocumentUpdateDTO dto = new DocumentUpdateDTO();
+        dto.setTitle(stored.getTitle());
+        dto.setDocumentDate(LocalDate.of(1918, 1, 10));
+        dto.setMetaDatePrecision(DatePrecision.RANGE);
+        dto.setMetaDateEnd(LocalDate.of(1918, 1, 11));
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – 10.–11. Jan. 1918");
+    }
+
+    @Test
+    void updateDocument_doesNotRegenerateToBlank_whenSubmittedTitleEmpty() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+        DocumentUpdateDTO dto = editDto("", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isNotBlank();
+    }
+
+    @Test
+    void updateDocument_treatsFileReplacedDoc_asManual() throws Exception {
+        // originalFilename was reassigned by an earlier file-replace, so the stored title (built
+        // at import from the old index) no longer matches build(currentState) → treated as manual.
+        Document stored = makeStored("scan_2024.pdf", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+        stored.setTitle("C-0029 – 1928 – Berlin"); // legacy import title, ≠ build("scan_2024.pdf"…)
+        DocumentUpdateDTO dto = editDto("C-0029 – 1928 – Berlin", LocalDate.of(1930, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo("C-0029 – 1928 – Berlin");
+    }
+
+    @Test
+    void updateDocument_idempotent_whenNothingChanges() throws Exception {
+        Document stored = makeStored("C-0029", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+        String before = stored.getTitle();
+        DocumentUpdateDTO dto = editDto(before, LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        runUpdate(stored, dto);
+
+        assertThat(stored.getTitle()).isEqualTo(before);
+    }
+
+    // ─── updateDocument date-range validation (#678) ──────────────────────────
+
+    /** Builds a stored doc ready for an updateDocument call (collections initialised). */
+    private static Document docForRangeUpdate(UUID id) {
+        return Document.builder().id(id).receivers(new HashSet<>()).tags(new HashSet<>()).build();
+    }
+
+    private static DocumentUpdateDTO rangeDto(LocalDate start, LocalDate end) {
+        DocumentUpdateDTO dto = new DocumentUpdateDTO();
+        dto.setDocumentDate(start);
+        dto.setMetaDatePrecision(DatePrecision.RANGE);
+        dto.setMetaDateEnd(end);
+        return dto;
+    }
+
+    @Test
+    void updateDocument_rejectsRange_whenEndBeforeStart() {
+        UUID id = UUID.randomUUID();
+        Document doc = docForRangeUpdate(id);
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+
+        DocumentUpdateDTO dto = rangeDto(LocalDate.of(1917, 1, 11), LocalDate.of(1917, 1, 10));
+
+        assertThatThrownBy(() -> documentService.updateDocument(id, dto, null, null))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.INVALID_DATE_RANGE);
+        verify(documentRepository, never()).save(any());
+    }
+
+    @Test
+    void updateDocument_acceptsRange_whenEndEqualsStart() throws Exception {
+        // AC2: the DB CHECK is end >= start, so equal dates are valid.
+        UUID id = UUID.randomUUID();
+        Document doc = docForRangeUpdate(id);
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+        when(documentRepository.save(any())).thenReturn(doc);
+
+        LocalDate same = LocalDate.of(1917, 1, 10);
+        documentService.updateDocument(id, rangeDto(same, same), null, null);
+
+        assertThat(doc.getMetaDateEnd()).isEqualTo(same);
+        verify(documentRepository, atLeastOnce()).save(any());
+    }
+
+    @Test
+    void updateDocument_acceptsRange_whenEndAfterStart() throws Exception {
+        UUID id = UUID.randomUUID();
+        Document doc = docForRangeUpdate(id);
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+        when(documentRepository.save(any())).thenReturn(doc);
+
+        documentService.updateDocument(id,
+                rangeDto(LocalDate.of(1917, 1, 10), LocalDate.of(1917, 1, 11)), null, null);
+
+        assertThat(doc.getMetaDateEnd()).isEqualTo(LocalDate.of(1917, 1, 11));
+        verify(documentRepository, atLeastOnce()).save(any());
+    }
+
+    @Test
+    void updateDocument_acceptsRange_whenEndIsNull_openEnded() throws Exception {
+        // AC3: an open-ended range (no end) is valid.
+        UUID id = UUID.randomUUID();
+        Document doc = docForRangeUpdate(id);
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+        when(documentRepository.save(any())).thenReturn(doc);
+
+        documentService.updateDocument(id,
+                rangeDto(LocalDate.of(1917, 1, 10), null), null, null);
+
+        verify(documentRepository, atLeastOnce()).save(any());
+    }
+
+    @Test
+    void updateDocument_acceptsRange_whenStartNullAndEndSet() throws Exception {
+        // AC4: mirrors the DB "meta_date IS NULL" escape — must NOT reject (and must not NPE).
+        UUID id = UUID.randomUUID();
+        Document doc = docForRangeUpdate(id);
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+        when(documentRepository.save(any())).thenReturn(doc);
+
+        documentService.updateDocument(id,
+                rangeDto(null, LocalDate.of(1917, 1, 11)), null, null);
+
+        assertThat(doc.getMetaDateEnd()).isEqualTo(LocalDate.of(1917, 1, 11));
+        verify(documentRepository, atLeastOnce()).save(any());
+    }
+
+    @Test
+    void updateDocument_rejectsEndDate_whenPrecisionNotRange() {
+        // AC6: an end date only makes sense for RANGE (mirrors chk_meta_date_end_only_for_range).
+        // API-only — the edit form clears the end field off-RANGE — so close the 500 class here too.
+        UUID id = UUID.randomUUID();
+        Document doc = docForRangeUpdate(id);
+        when(documentRepository.findById(id)).thenReturn(Optional.of(doc));
+
+        DocumentUpdateDTO dto = new DocumentUpdateDTO();
+        dto.setDocumentDate(LocalDate.of(1917, 1, 10));
+        dto.setMetaDatePrecision(DatePrecision.MONTH);
+        dto.setMetaDateEnd(LocalDate.of(1917, 1, 31));
+
+        assertThatThrownBy(() -> documentService.updateDocument(id, dto, null, null))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.INVALID_DATE_RANGE);
+        verify(documentRepository, never()).save(any());
+    }
+
     // ─── deleteTagCascading ───────────────────────────────────────────────────
 
     @Test
@@ -338,6 +695,59 @@ class DocumentServiceTest {
         verify(documentVersionService).recordVersion(any(Document.class));
     }
 
+    // ─── backfillTitles — one-time stale-title cleanup (#726, FR-003) ─────────
+
+    @Test
+    void backfillTitles_rewritesStaleAutoTitle_andCountsIt() {
+        Document stale = makeStored("C-0029", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+        stale.setTitle("C-0029 – 2028 – Berlin"); // stale stored title (date typo never fixed)
+        when(documentRepository.findAll()).thenReturn(List.of(stale));
+        when(documentRepository.save(any())).thenReturn(stale);
+
+        int count = documentService.backfillTitles();
+
+        assertThat(count).isEqualTo(1);
+        assertThat(stale.getTitle()).isEqualTo("C-0029 – 1928 – Berlin");
+        verify(documentRepository).save(stale);
+    }
+
+    @Test
+    void backfillTitles_skipsProse() {
+        Document prose = makeStored("C-0030", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, null);
+        prose.setTitle("C-0030 – Brief an Mutter");
+        when(documentRepository.findAll()).thenReturn(List.of(prose));
+
+        int count = documentService.backfillTitles();
+
+        assertThat(count).isZero();
+        assertThat(prose.getTitle()).isEqualTo("C-0030 – Brief an Mutter");
+        verify(documentRepository, never()).save(any());
+    }
+
+    @Test
+    void backfillTitles_isIdempotent_forAlreadyCorrectTitle() {
+        Document fresh = makeStored("C-0031", LocalDate.of(1940, 1, 1), DatePrecision.YEAR, null);
+        // title already equals build(current state) → nothing to do
+        when(documentRepository.findAll()).thenReturn(List.of(fresh));
+
+        int count = documentService.backfillTitles();
+
+        assertThat(count).isZero();
+        verify(documentRepository, never()).save(any());
+    }
+
+    @Test
+    void backfillTitles_neverRecordsVersions() {
+        Document stale = makeStored("C-0029", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+        stale.setTitle("C-0029 – 2028 – Berlin");
+        when(documentRepository.findAll()).thenReturn(List.of(stale));
+        when(documentRepository.save(any())).thenReturn(stale);
+
+        documentService.backfillTitles();
+
+        verify(documentVersionService, never()).recordVersion(any());
+    }
+
     // ─── thumbnail dispatch ───────────────────────────────────────────────────
 
     @Test
@@ -985,53 +1395,6 @@ class DocumentServiceTest {
                 .isEqualTo("19650332_Mueller_Hans");
     }
 
-    // ─── getConversationFiltered ───────────────────────────────────────────────
-
-    @Test
-    void getConversationFiltered_passesGivenDates_whenFromAndToAreProvided() {
-        UUID senderId = UUID.randomUUID();
-        UUID receiverId = UUID.randomUUID();
-        LocalDate from = LocalDate.of(1940, 1, 1);
-        LocalDate to = LocalDate.of(1960, 12, 31);
-        Sort sort = Sort.by(Sort.Direction.ASC, "documentDate");
-        when(documentRepository.findConversation(senderId, receiverId, from, to, sort))
-                .thenReturn(List.of());
-
-        documentService.getConversationFiltered(senderId, receiverId, from, to, sort);
-
-        verify(documentRepository).findConversation(senderId, receiverId, from, to, sort);
-    }
-
-    @Test
-    void getConversationFiltered_usesMinDateForFrom_whenFromIsNull() {
-        UUID senderId = UUID.randomUUID();
-        UUID receiverId = UUID.randomUUID();
-        Sort sort = Sort.by(Sort.Direction.ASC, "documentDate");
-        when(documentRepository.findConversation(eq(senderId), eq(receiverId), any(LocalDate.class), any(LocalDate.class), eq(sort)))
-                .thenReturn(List.of());
-
-        documentService.getConversationFiltered(senderId, receiverId, null, null, sort);
-
-        ArgumentCaptor<LocalDate> fromCaptor = ArgumentCaptor.forClass(LocalDate.class);
-        verify(documentRepository).findConversation(eq(senderId), eq(receiverId), fromCaptor.capture(), any(LocalDate.class), eq(sort));
-        assertThat(fromCaptor.getValue()).isEqualTo(LocalDate.parse("0000-01-01"));
-    }
-
-    @Test
-    void getConversationFiltered_usesTodayForTo_whenToIsNull() {
-        UUID senderId = UUID.randomUUID();
-        UUID receiverId = UUID.randomUUID();
-        Sort sort = Sort.by(Sort.Direction.ASC, "documentDate");
-        when(documentRepository.findConversation(eq(senderId), eq(receiverId), any(LocalDate.class), any(LocalDate.class), eq(sort)))
-                .thenReturn(List.of());
-
-        documentService.getConversationFiltered(senderId, receiverId, null, null, sort);
-
-        ArgumentCaptor<LocalDate> toCaptor = ArgumentCaptor.forClass(LocalDate.class);
-        verify(documentRepository).findConversation(eq(senderId), eq(receiverId), any(LocalDate.class), toCaptor.capture(), eq(sort));
-        assertThat(toCaptor.getValue()).isEqualTo(LocalDate.now());
-    }
-
     // ─── updateDocumentTags — empty tag in list ───────────────────────────────
 
     @Test
@@ -1410,8 +1773,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of()));
 
-        documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                org.raddatz.familienarchiv.document.DocumentSort.DATE, "DESC", null, false, org.springframework.data.domain.PageRequest.of(1, 50));
+        documentService.searchDocuments(
+                noFilters(),
+                org.raddatz.familienarchiv.document.DocumentSort.DATE, "DESC", org.springframework.data.domain.PageRequest.of(1, 50));
 
         verify(documentRepository).findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class));
         verify(documentRepository, never()).findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Sort.class));
@@ -1423,8 +1787,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of()));
 
-        documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                org.raddatz.familienarchiv.document.DocumentSort.DATE, "DESC", null, false, org.springframework.data.domain.PageRequest.of(3, 25));
+        documentService.searchDocuments(
+                noFilters(),
+                org.raddatz.familienarchiv.document.DocumentSort.DATE, "DESC", org.springframework.data.domain.PageRequest.of(3, 25));
 
         verify(documentRepository).findAll(any(org.springframework.data.jpa.domain.Specification.class), captor.capture());
         assertThat(captor.getValue().getPageNumber()).isEqualTo(3);
@@ -1439,8 +1804,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of(d), org.springframework.data.domain.PageRequest.of(0, 50), 120L));
 
-        DocumentSearchResult result = documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                org.raddatz.familienarchiv.document.DocumentSort.DATE, "DESC", null, false, org.springframework.data.domain.PageRequest.of(0, 50));
+        DocumentSearchResult result = documentService.searchDocuments(
+                noFilters(),
+                org.raddatz.familienarchiv.document.DocumentSort.DATE, "DESC", org.springframework.data.domain.PageRequest.of(0, 50));
 
         assertThat(result.totalElements()).isEqualTo(120L);
         assertThat(result.pageNumber()).isZero();
@@ -1455,8 +1821,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of()));
 
-        documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "DESC", null, false, org.springframework.data.domain.PageRequest.of(0, 5));
+        documentService.searchDocuments(
+                noFilters(),
+                DocumentSort.DATE, "DESC", org.springframework.data.domain.PageRequest.of(0, 5));
 
         verify(documentRepository).findAll(any(org.springframework.data.jpa.domain.Specification.class), captor.capture());
         Sort.Order dateOrder = captor.getValue().getSort().getOrderFor("documentDate");
@@ -1478,8 +1845,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of()));
 
-        documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                DocumentSort.DATE, "ASC", null, false, org.springframework.data.domain.PageRequest.of(0, 5));
+        documentService.searchDocuments(
+                noFilters(),
+                DocumentSort.DATE, "ASC", org.springframework.data.domain.PageRequest.of(0, 5));
 
         verify(documentRepository).findAll(any(org.springframework.data.jpa.domain.Specification.class), captor.capture());
         Sort.Order dateOrder = captor.getValue().getSort().getOrderFor("documentDate");
@@ -1499,8 +1867,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of()));
 
-        documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                DocumentSort.UPDATED_AT, "DESC", null, false, org.springframework.data.domain.PageRequest.of(0, 5));
+        documentService.searchDocuments(
+                noFilters(),
+                DocumentSort.UPDATED_AT, "DESC", org.springframework.data.domain.PageRequest.of(0, 5));
 
         verify(documentRepository).findAll(any(org.springframework.data.jpa.domain.Specification.class), captor.capture());
         assertThat(captor.getValue().getSort())
@@ -1523,8 +1892,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class)))
                 .thenReturn(all);
 
-        DocumentSearchResult result = documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                org.raddatz.familienarchiv.document.DocumentSort.SENDER, "asc", null, false, org.springframework.data.domain.PageRequest.of(1, 50));
+        DocumentSearchResult result = documentService.searchDocuments(
+                noFilters(),
+                org.raddatz.familienarchiv.document.DocumentSort.SENDER, "asc", org.springframework.data.domain.PageRequest.of(1, 50));
 
         assertThat(result.totalElements()).isEqualTo(120L);
         assertThat(result.pageNumber()).isEqualTo(1);
@@ -1547,8 +1917,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class)))
                 .thenReturn(all);
 
-        DocumentSearchResult result = documentService.searchDocuments(null, null, null, null, null, null, null, null,
-                org.raddatz.familienarchiv.document.DocumentSort.SENDER, "asc", null, false, org.springframework.data.domain.PageRequest.of(10, 50));
+        DocumentSearchResult result = documentService.searchDocuments(
+                noFilters(),
+                org.raddatz.familienarchiv.document.DocumentSort.SENDER, "asc", org.springframework.data.domain.PageRequest.of(10, 50));
 
         assertThat(result.items()).isEmpty();
         assertThat(result.totalElements()).isEqualTo(30L);
@@ -1561,7 +1932,8 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of()));
 
-        documentService.searchDocuments(null, null, null, null, null, null, null, DocumentStatus.REVIEWED, null, null, null, false, UNPAGED);
+        documentService.searchDocuments(
+                new SearchFilters(null, null, null, null, null, null, null, DocumentStatus.REVIEWED, null, false), null, null, UNPAGED);
 
         verify(documentRepository).findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class));
     }
@@ -1571,7 +1943,8 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class)))
                 .thenReturn(new PageImpl<>(List.of()));
 
-        documentService.searchDocuments(null, null, null, null, null, null, null, null, null, null, null, false, UNPAGED);
+        documentService.searchDocuments(
+                noFilters(), null, null, UNPAGED);
 
         verify(documentRepository).findAll(any(org.springframework.data.jpa.domain.Specification.class), any(Pageable.class));
     }
@@ -1607,35 +1980,6 @@ class DocumentServiceTest {
                 .isEqualTo(Sort.by(Sort.Direction.DESC, "updatedAt"));
     }
 
-    // ─── getConversationFiltered (single-person mode) ─────────────────────────
-
-    @Test
-    void getConversationFiltered_callsSinglePersonQuery_whenReceiverIdIsNull() {
-        UUID personId = UUID.randomUUID();
-        Sort sort = Sort.by(Sort.Direction.DESC, "documentDate");
-        when(documentRepository.findSinglePersonCorrespondence(eq(personId), any(), any(), eq(sort)))
-                .thenReturn(List.of());
-
-        documentService.getConversationFiltered(personId, null, null, null, sort);
-
-        verify(documentRepository).findSinglePersonCorrespondence(eq(personId), any(), any(), eq(sort));
-        verify(documentRepository, never()).findConversation(any(), any(), any(), any(), any());
-    }
-
-    @Test
-    void getConversationFiltered_callsBilateralQuery_whenReceiverIdIsSet() {
-        UUID senderId = UUID.randomUUID();
-        UUID receiverId = UUID.randomUUID();
-        Sort sort = Sort.by(Sort.Direction.DESC, "documentDate");
-        when(documentRepository.findConversation(eq(senderId), eq(receiverId), any(), any(), eq(sort)))
-                .thenReturn(List.of());
-
-        documentService.getConversationFiltered(senderId, receiverId, null, null, sort);
-
-        verify(documentRepository).findConversation(eq(senderId), eq(receiverId), any(), any(), eq(sort));
-        verify(documentRepository, never()).findSinglePersonCorrespondence(any(), any(), any(), any());
-    }
-
     // ─── searchDocuments — SENDER sort includes documents with null sender ─────
 
     @Test
@@ -1649,7 +1993,8 @@ class DocumentServiceTest {
                 .thenReturn(List.of(withSender, noSender));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null, DocumentSort.SENDER, "asc", null, false, UNPAGED);
+                noFilters(),
+                DocumentSort.SENDER, "asc", UNPAGED);
 
         assertThat(result.items()).hasSize(2);
         assertThat(result.items()).extracting(DocumentListItem::title).containsExactly("Has Sender", "No Sender");
@@ -1669,7 +2014,8 @@ class DocumentServiceTest {
                 .thenReturn(List.of(noReceivers, withReceiver));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null, DocumentSort.RECEIVER, "asc", null, false, UNPAGED);
+                noFilters(),
+                DocumentSort.RECEIVER, "asc", UNPAGED);
 
         assertThat(result.items()).extracting(DocumentListItem::title)
                 .containsExactly("Has Receiver", "No Receivers");
@@ -1702,7 +2048,8 @@ class DocumentServiceTest {
                 .thenReturn(List.of(undatedBob, datedAnna, datedBob, undatedAnna));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null, DocumentSort.SENDER, "asc", null, false, UNPAGED);
+                noFilters(),
+                DocumentSort.SENDER, "asc", UNPAGED);
 
         // Bob's group precedes Anna's group (ASC by sender). The sort is stable, so
         // within each group the input order is preserved (undatedBob, datedBob for Bob;
@@ -1733,7 +2080,8 @@ class DocumentServiceTest {
                 .thenReturn(List.of(undatedBob, datedAnna, datedBob, undatedAnna));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null, DocumentSort.SENDER, "desc", null, false, UNPAGED);
+                noFilters(),
+                DocumentSort.SENDER, "desc", UNPAGED);
 
         // Anna's group precedes Bob's (DESC by sender); undated stays inside its group.
         assertThat(result.items()).extracting(DocumentListItem::title)
@@ -1756,7 +2104,8 @@ class DocumentServiceTest {
                 .thenReturn(List.of(undatedFromAlice));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null, DocumentSort.SENDER, "asc", null, true, UNPAGED);
+                noFilters().withUndated(true),
+                DocumentSort.SENDER, "asc", UNPAGED);
 
         // The in-memory path queried via a Specification (built by buildSearchSpec with
         // undatedOnly(true)) rather than skipping straight to a sorted findAll.
@@ -1772,8 +2121,9 @@ class DocumentServiceTest {
         when(documentRepository.findAll(any(org.springframework.data.jpa.domain.Specification.class)))
                 .thenReturn(List.of());
 
-        documentService.searchDocuments("brief", null, null, null, null, null, null, null,
-                DocumentSort.RELEVANCE, null, null, true, UNPAGED);
+        documentService.searchDocuments(
+                new SearchFilters("brief", null, null, null, null, null, null, null, null, true),
+                DocumentSort.RELEVANCE, null, UNPAGED);
 
         // The FTS-id path (buildSearchSpec) ran; the raw-page SQL shortcut did not.
         verify(documentRepository).findAllMatchingIdsByFts("brief");
@@ -1796,7 +2146,8 @@ class DocumentServiceTest {
                 .thenReturn(List.of(docNullName, docSmith));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null, DocumentSort.SENDER, "asc", null, false, UNPAGED);
+                noFilters(),
+                DocumentSort.SENDER, "asc", UNPAGED);
 
         // null lastName should sort to end (treated as empty), not before "smith" (as "null")
         assertThat(result.items()).extracting(DocumentListItem::title)
@@ -1819,7 +2170,8 @@ class DocumentServiceTest {
         when(documentRepository.findEnrichmentData(any(), eq("Brief"))).thenReturn(rows);
 
         DocumentSearchResult result = documentService.searchDocuments(
-                "Brief", null, null, null, null, null, null, null, DocumentSort.RELEVANCE, null, null, false, UNPAGED);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                DocumentSort.RELEVANCE, null, UNPAGED);
 
         assertThat(result.items()).hasSize(1);
         SearchMatchData md = result.items().get(0).matchData();
@@ -1833,7 +2185,8 @@ class DocumentServiceTest {
                 .thenReturn(new PageImpl<>(List.of()));
 
         DocumentSearchResult result = documentService.searchDocuments(
-                null, null, null, null, null, null, null, null, null, null, null, false, UNPAGED);
+                noFilters(),
+                null, null, UNPAGED);
 
         assertThat(result.items()).isEmpty();
     }
@@ -1853,7 +2206,8 @@ class DocumentServiceTest {
         when(documentRepository.findEnrichmentData(any(), eq("Brief"))).thenReturn(rows);
 
         DocumentSearchResult result = documentService.searchDocuments(
-                "Brief", null, null, null, null, null, null, null, DocumentSort.RELEVANCE, null, null, false, UNPAGED);
+                new SearchFilters("Brief", null, null, null, null, null, null, null, null, false),
+                DocumentSort.RELEVANCE, null, UNPAGED);
 
         SearchMatchData md = result.items().get(0).matchData();
         assertThat(md.transcriptionSnippet()).isEqualTo("Hier ist der Brief aus Berlin");
@@ -2370,7 +2724,7 @@ class DocumentServiceTest {
                 .thenReturn(List.of(d1, d2));
 
         List<UUID> result = documentService.findIdsForFilter(
-                null, null, null, null, null, null, null, null, null, false);
+                noFilters());
 
         assertThat(result).containsExactly(d1.getId(), d2.getId());
     }
@@ -2385,7 +2739,7 @@ class DocumentServiceTest {
         when(tagService.expandTagNamesToDescendantIdSets(any())).thenReturn(List.of());
 
         documentService.findIdsForFilter(
-                null, null, null, null, null, List.of("Brief"), null, null, TagOperator.OR, false);
+                new SearchFilters(null, null, null, null, null, List.of("Brief"), null, null, TagOperator.OR, false));
 
         // Spec built without throwing → OR branch was exercised. Coverage gain
         // is in not-throwing on the OR-specific code path; the actual SQL is
@@ -2398,7 +2752,7 @@ class DocumentServiceTest {
         when(documentRepository.findAllMatchingIdsByFts("xyz")).thenReturn(List.of());
 
         List<UUID> result = documentService.findIdsForFilter(
-                "xyz", null, null, null, null, null, null, null, null, false);
+                new SearchFilters("xyz", null, null, null, null, null, null, null, null, false));
 
         assertThat(result).isEmpty();
         verify(documentRepository, never()).findAll(any(org.springframework.data.jpa.domain.Specification.class));

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

@@ -0,0 +1,90 @@
+package org.raddatz.familienarchiv.document;
+
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.PostgresContainerConfig;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.boot.test.context.SpringBootTest;
+import org.springframework.context.annotation.Import;
+import org.springframework.test.context.ActiveProfiles;
+import org.springframework.test.context.bean.override.mockito.MockitoBean;
+import org.springframework.transaction.annotation.Transactional;
+import software.amazon.awssdk.services.s3.S3Client;
+
+import java.time.LocalDate;
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+/**
+ * End-to-end backfill against a real Postgres (#726, FR-003). H2 is unusable here — the
+ * {@code title} column is NOT NULL and the title-sync semantics depend on that — so this pins the
+ * behaviour on {@code postgres:16-alpine}: a stale auto-title is rewritten, the sweep is
+ * idempotent, prose is left alone, and the mechanical rename writes no {@code document_versions}
+ * rows. Permission enforcement (401/403) is covered faster by the {@code @WebMvcTest} slice in
+ * {@code AdminControllerTest}.
+ */
+@SpringBootTest(webEnvironment = SpringBootTest.WebEnvironment.NONE)
+@ActiveProfiles("test")
+@Import(PostgresContainerConfig.class)
+@Transactional
+class DocumentTitleBackfillIntegrationTest {
+
+    @MockitoBean S3Client s3Client;
+    @Autowired DocumentService documentService;
+    @Autowired DocumentRepository documentRepository;
+    @Autowired DocumentVersionRepository documentVersionRepository;
+
+    private Document persist(String index, String title, LocalDate date, DatePrecision precision, String location) {
+        return documentRepository.save(Document.builder()
+                .originalFilename(index)
+                .title(title)
+                .documentDate(date)
+                .metaDatePrecision(precision)
+                .location(location)
+                .status(DocumentStatus.PLACEHOLDER)
+                .build());
+    }
+
+    @Test
+    void backfill_rewritesStaleAutoTitle() {
+        Document stale = persist("C-0029", "C-0029 – 2028 – Berlin",
+                LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        int count = documentService.backfillTitles();
+
+        assertThat(count).isEqualTo(1); // exactly the one stale row seeded (clean test DB)
+        assertThat(documentRepository.findById(stale.getId()).orElseThrow().getTitle())
+                .isEqualTo("C-0029 – 1928 – Berlin");
+    }
+
+    @Test
+    void backfill_isIdempotent_secondRunChangesNothing() {
+        persist("C-0029", "C-0029 – 2028 – Berlin", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+
+        documentService.backfillTitles();
+        int secondRun = documentService.backfillTitles();
+
+        assertThat(secondRun).isZero();
+    }
+
+    @Test
+    void backfill_skipsProse() {
+        Document prose = persist("C-0030", "C-0030 – Brief an Mutter",
+                LocalDate.of(1928, 1, 1), DatePrecision.YEAR, null);
+
+        documentService.backfillTitles();
+
+        assertThat(documentRepository.findById(prose.getId()).orElseThrow().getTitle())
+                .isEqualTo("C-0030 – Brief an Mutter");
+    }
+
+    @Test
+    void backfill_addsNoDocumentVersionRows() {
+        persist("C-0029", "C-0029 – 2028 – Berlin", LocalDate.of(1928, 1, 1), DatePrecision.YEAR, "Berlin");
+        long versionsBefore = documentVersionRepository.count();
+
+        documentService.backfillTitles();
+
+        assertThat(documentVersionRepository.count()).isEqualTo(versionsBefore);
+    }
+}

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

@@ -0,0 +1,175 @@
+package org.raddatz.familienarchiv.document;
+
+import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.Timeout;
+
+import java.util.concurrent.TimeUnit;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+/**
+ * The backfill overwrite heuristic (FR-004) in isolation — every emittable date-label form is
+ * recognised, prose is left alone, and a regex-metacharacter index is matched literally without
+ * hanging. The exact label spellings mirror {@code docs/date-label-fixtures.json}.
+ */
+class DocumentTitleBackfillMatcherTest {
+
+    private static boolean overwritable(String title, String location) {
+        return DocumentTitleBackfillMatcher.isOverwritable(title, "C-0029", location);
+    }
+
+    // ─── each date-label form (index + form) is overwritable ──────────────────
+
+    @Test
+    void year_form() {
+        assertThat(overwritable("C-0029 – 1916", null)).isTrue();
+    }
+
+    @Test
+    void approx_form() {
+        assertThat(overwritable("C-0029 – ca. 1920", null)).isTrue();
+    }
+
+    @Test
+    void month_form() {
+        assertThat(overwritable("C-0029 – Juni 1916", null)).isTrue();
+    }
+
+    @Test
+    void day_form() {
+        assertThat(overwritable("C-0029 – 24. Dezember 1943", null)).isTrue();
+    }
+
+    @Test
+    void season_form() {
+        assertThat(overwritable("C-0029 – Sommer 1916", null)).isTrue();
+    }
+
+    @Test
+    void unknown_label_form() {
+        assertThat(overwritable("C-0029 – Datum unbekannt", null)).isTrue();
+    }
+
+    @Test
+    void range_same_month_form() {
+        assertThat(overwritable("C-0029 – 10.–11. Jan. 1917", null)).isTrue();
+    }
+
+    @Test
+    void range_cross_month_form() {
+        assertThat(overwritable("C-0029 – 30. Jan. – 2. Feb. 1917", null)).isTrue();
+    }
+
+    @Test
+    void range_cross_year_form() {
+        assertThat(overwritable("C-0029 – 30. Dez. 1916 – 2. Jan. 1917", null)).isTrue();
+    }
+
+    @Test
+    void range_single_day_form() {
+        assertThat(overwritable("C-0029 – 10. Jan. 1917", null)).isTrue();
+    }
+
+    @Test
+    void range_open_form() {
+        assertThat(overwritable("C-0029 – ab 10. Jan. 1917", null)).isTrue();
+    }
+
+    // ─── date label + trailing location (any location) ────────────────────────
+
+    @Test
+    void date_form_with_trailing_location() {
+        assertThat(overwritable("C-0029 – 1916 – Berlin", null)).isTrue();
+    }
+
+    @Test
+    void range_with_internal_separator_plus_trailing_location() {
+        // The range label itself contains " – "; the trailing " – Berlin" must still be peeled.
+        assertThat(overwritable("C-0029 – 30. Jan. – 2. Feb. 1917 – Berlin", null)).isTrue();
+    }
+
+    // ─── index-only and index+location cases ──────────────────────────────────
+
+    @Test
+    void exactly_index() {
+        assertThat(overwritable("C-0029", null)).isTrue();
+    }
+
+    @Test
+    void index_plus_location_equal_to_current() {
+        assertThat(overwritable("C-0029 – Berlin", "Berlin")).isTrue();
+    }
+
+    // ─── prose is left untouched ──────────────────────────────────────────────
+
+    @Test
+    void prose_segment_not_matching_location_is_skipped() {
+        assertThat(overwritable("C-0029 – Brief an Mutter", "Berlin")).isFalse();
+    }
+
+    @Test
+    void location_only_segment_is_skipped_when_no_current_location() {
+        // No date label, and the doc has no location to compare against → cannot prove machine.
+        assertThat(overwritable("C-0029 – Berlin", null)).isFalse();
+    }
+
+    @Test
+    void title_not_starting_with_index_is_skipped() {
+        assertThat(overwritable("Ganz anderer Titel", null)).isFalse();
+    }
+
+    // ─── near-miss: shapes that look almost machine-built but are not ──────────
+
+    @Test
+    void ascii_hyphen_instead_of_en_dash_separator_is_skipped() {
+        // The separator is " – " (en dash); a plain " - " is not the machine separator.
+        assertThat(overwritable("C-0029 - 1916", null)).isFalse();
+    }
+
+    @Test
+    void date_label_without_separator_before_trailing_text_is_skipped() {
+        // "1916 Berlin" is not a date label and is not joined by " – "; prose, not machine.
+        assertThat(overwritable("C-0029 – 1916 Berlin", null)).isFalse();
+    }
+
+    @Test
+    void year_with_trailing_letters_is_not_a_year_label() {
+        assertThat(overwritable("C-0029 – 1916er Brief", null)).isFalse();
+    }
+
+    @Test
+    void index_immediately_followed_by_text_without_separator_is_skipped() {
+        assertThat(overwritable("C-0029x – 1916", null)).isFalse();
+    }
+
+    // ─── fail-closed guards ───────────────────────────────────────────────────
+
+    @Test
+    void null_title_is_not_overwritable() {
+        assertThat(DocumentTitleBackfillMatcher.isOverwritable(null, "C-0029", null)).isFalse();
+    }
+
+    @Test
+    void null_index_is_not_overwritable() {
+        assertThat(DocumentTitleBackfillMatcher.isOverwritable("C-0029 – 1916", null, null)).isFalse();
+    }
+
+    @Test
+    void blank_index_is_not_overwritable() {
+        assertThat(DocumentTitleBackfillMatcher.isOverwritable("  – 1916", "   ", null)).isFalse();
+    }
+
+    // ─── ReDoS / regex-metacharacter index is matched literally and terminates ─
+
+    @Test
+    @Timeout(value = 5, unit = TimeUnit.SECONDS)
+    void index_with_regex_metacharacters_is_matched_literally_and_terminates() {
+        String hostileIndex = "C-0029(.*).pdf";
+        // Literal prefix → matches; trailing date label → overwritable. Must not hang.
+        assertThat(DocumentTitleBackfillMatcher.isOverwritable(
+                hostileIndex + " – 1916", hostileIndex, null)).isTrue();
+        // A title that does NOT start with the literal hostile index is skipped, also fast.
+        assertThat(DocumentTitleBackfillMatcher.isOverwritable(
+                "C-0029 – 1916", hostileIndex, null)).isFalse();
+    }
+}

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

@@ -0,0 +1,89 @@
+package org.raddatz.familienarchiv.document;
+
+import org.junit.jupiter.api.Test;
+
+import java.time.LocalDate;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+/**
+ * The auto-title composition {@code {index} – {dateLabel} – {location}} in isolation.
+ * The honest date-label forms themselves are pinned by {@link DocumentTitleFormatterTest}
+ * against the shared #666 fixture; here we assert only how the factory composes the
+ * three segments and which segments it omits.
+ */
+class DocumentTitleFactoryTest {
+
+    private final DocumentTitleFactory factory = new DocumentTitleFactory();
+
+    private static Document.DocumentBuilder doc(String index) {
+        return Document.builder()
+                .originalFilename(index)
+                .metaDatePrecision(DatePrecision.UNKNOWN);
+    }
+
+    @Test
+    void index_only_when_no_date_and_no_location() {
+        assertThat(factory.build(doc("C-0029").build())).isEqualTo("C-0029");
+    }
+
+    @Test
+    void index_and_year_date() {
+        Document d = doc("C-0029")
+                .documentDate(LocalDate.of(1928, 1, 15))
+                .metaDatePrecision(DatePrecision.YEAR)
+                .build();
+        assertThat(factory.build(d)).isEqualTo("C-0029 – 1928");
+    }
+
+    @Test
+    void index_date_and_location() {
+        Document d = doc("C-0029")
+                .documentDate(LocalDate.of(1928, 1, 15))
+                .metaDatePrecision(DatePrecision.YEAR)
+                .location("Berlin")
+                .build();
+        assertThat(factory.build(d)).isEqualTo("C-0029 – 1928 – Berlin");
+    }
+
+    @Test
+    void location_without_date_attaches_directly_to_index() {
+        Document d = doc("C-0029").location("Berlin").build();
+        assertThat(factory.build(d)).isEqualTo("C-0029 – Berlin");
+    }
+
+    @Test
+    void unknown_precision_omits_the_date_segment() {
+        Document d = doc("C-0029")
+                .documentDate(LocalDate.of(1928, 1, 15))
+                .metaDatePrecision(DatePrecision.UNKNOWN)
+                .build();
+        assertThat(factory.build(d)).isEqualTo("C-0029");
+    }
+
+    @Test
+    void blank_location_is_omitted() {
+        Document d = doc("C-0029")
+                .documentDate(LocalDate.of(1928, 1, 15))
+                .metaDatePrecision(DatePrecision.YEAR)
+                .location("   ")
+                .build();
+        assertThat(factory.build(d)).isEqualTo("C-0029 – 1928");
+    }
+
+    @Test
+    void bare_document_with_null_index_builds_empty_string_not_npe() {
+        // originalFilename is NOT NULL in production; the guard keeps a synthetic/partial entity
+        // from tripping StringBuilder(null) with an opaque NPE.
+        assertThat(factory.build(Document.builder().build())).isEqualTo("");
+    }
+
+    @Test
+    void day_precision_renders_the_full_german_label() {
+        Document d = doc("C-0029")
+                .documentDate(LocalDate.of(1928, 1, 15))
+                .metaDatePrecision(DatePrecision.DAY)
+                .build();
+        assertThat(factory.build(d)).isEqualTo("C-0029 – 15. Januar 1928");
+    }
+}

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

@@ -1,10 +1,9 @@
-package org.raddatz.familienarchiv.importing;
+package org.raddatz.familienarchiv.document;
 
 import com.fasterxml.jackson.databind.JsonNode;
 import com.fasterxml.jackson.databind.ObjectMapper;
 import org.junit.jupiter.api.DynamicTest;
 import org.junit.jupiter.api.TestFactory;
-import org.raddatz.familienarchiv.document.DatePrecision;
 
 import java.nio.file.Files;
 import java.nio.file.Path;

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

@@ -0,0 +1,17 @@
+package org.raddatz.familienarchiv.document;
+
+/** Test fixtures for {@link SearchFilters}. */
+final class SearchFiltersFixtures {
+
+    private SearchFiltersFixtures() {}
+
+    /**
+     * A {@link SearchFilters} with no predicate active — the common search-test
+     * baseline. Combine with {@code .withUndated(true)} for the undated-only case;
+     * construct {@code new SearchFilters(...)} directly when a test pins a specific
+     * field, so the intent stays visible at the call site.
+     */
+    static SearchFilters noFilters() {
+        return new SearchFilters(null, null, null, null, null, null, null, null, null, false);
+    }
+}

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

@@ -0,0 +1,123 @@
+package org.raddatz.familienarchiv.document;
+
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.PostgresContainerConfig;
+import org.raddatz.familienarchiv.tag.Tag;
+import org.raddatz.familienarchiv.tag.TagRepository;
+import org.raddatz.familienarchiv.tag.TagService;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.boot.test.context.SpringBootTest;
+import org.springframework.context.annotation.Import;
+import org.springframework.test.context.ActiveProfiles;
+import org.springframework.test.context.bean.override.mockito.MockitoBean;
+import org.springframework.transaction.annotation.Transactional;
+import software.amazon.awssdk.services.s3.S3Client;
+
+import java.time.LocalDate;
+import java.util.Comparator;
+import java.util.HashSet;
+import java.util.List;
+import java.util.Set;
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatCode;
+
+/**
+ * #730 — tag-name resolution against a real Postgres. A mocked repo can't prove the two things that
+ * actually break: that {@code findAllByNameIgnoreCase} folds case the way Postgres {@code LOWER()}
+ * does (critical for umlauts like {@code ü}), and that saving a document tagged with a case-colliding
+ * tag no longer throws {@code NonUniqueResultException}. H2 folds case differently, so this pins the
+ * behaviour on {@code postgres:16-alpine}. The four-branch resolution logic itself is covered faster
+ * by the mocked {@code TagServiceTest}.
+ */
+@SpringBootTest(webEnvironment = SpringBootTest.WebEnvironment.NONE)
+@ActiveProfiles("test")
+@Import(PostgresContainerConfig.class)
+@Transactional
+class TagCaseCollisionIntegrationTest {
+
+    @MockitoBean S3Client s3Client;
+    @Autowired DocumentService documentService;
+    @Autowired DocumentRepository documentRepository;
+    @Autowired TagRepository tagRepository;
+    @Autowired TagService tagService;
+
+    private Tag persistTag(String name, String sourceRef, UUID parentId) {
+        return tagRepository.save(Tag.builder().name(name).sourceRef(sourceRef).parentId(parentId).build());
+    }
+
+    private Document persistDocTaggedWith(Tag tag) {
+        return documentRepository.save(Document.builder()
+                .originalFilename("C-7301")
+                .title("Weihnachtsbrief")
+                .documentDate(LocalDate.of(1928, 1, 1))
+                .metaDatePrecision(DatePrecision.YEAR)
+                .status(DocumentStatus.UPLOADED)
+                .tags(new HashSet<>(Set.of(tag)))
+                .build());
+    }
+
+    @Test
+    void updateDocument_succeedsAndKeepsExactChildTag_whenTaggedWithCaseCollidingChild() throws Exception {
+        Tag parent = persistTag("Weihnachten", "Weihnachten", null);
+        Tag child = persistTag("weihnachten", "Weihnachten/weihnachten", parent.getId());
+        Document doc = persistDocTaggedWith(child);
+
+        DocumentUpdateDTO dto = new DocumentUpdateDTO();
+        dto.setTitle("Weihnachtsbrief");
+        dto.setDocumentDate(LocalDate.of(1930, 1, 1)); // change the date — the field that 500'd on staging
+        dto.setMetaDatePrecision(DatePrecision.YEAR);
+        dto.setTags("weihnachten"); // the edit form round-trips the stored child name
+
+        assertThatCode(() -> documentService.updateDocument(doc.getId(), dto, null, null))
+                .doesNotThrowAnyException();
+
+        Set<Tag> tags = documentRepository.findById(doc.getId()).orElseThrow().getTags();
+        assertThat(tags).hasSize(1);
+        assertThat(tags.iterator().next().getId()).isEqualTo(child.getId()); // child kept, not the parent
+    }
+
+    @Test
+    void findOrCreate_resolvesUmlautCollisionDeterministically_withoutThrow() {
+        // The regression catcher: a plain-ASCII pair would stay green even if Postgres folded ü wrongly.
+        Tag parent = persistTag("Glückwünsche", "Glückwünsche", null);
+        Tag child = persistTag("glückwünsche", "Glückwünsche/glückwünsche", parent.getId());
+
+        // Proof that real Postgres LOWER() folds the umlaut so both rows match case-insensitively.
+        // Query with the UPPERCASE form findOrCreate actually passes — folding LOWER('GLÜCKWÜNSCHE')
+        // against LOWER(name) is the exact step under test; a lowercase probe wouldn't exercise it.
+        assertThat(tagRepository.findAllByNameIgnoreCase("GLÜCKWÜNSCHE")).hasSize(2);
+
+        // No exact-case "GLÜCKWÜNSCHE" row exists → resolution falls through to the case-insensitive
+        // branch with two candidates and must pick the lowest id deterministically, never throwing.
+        UUID expected = List.of(parent, child).stream().min(Comparator.comparing(Tag::getId)).orElseThrow().getId();
+        Tag first = tagService.findOrCreate("GLÜCKWÜNSCHE");
+        Tag second = tagService.findOrCreate("GLÜCKWÜNSCHE");
+
+        assertThat(first.getId()).isEqualTo(expected);
+        assertThat(second.getId()).isEqualTo(first.getId());
+    }
+
+    @Test
+    void bulkEdit_resolvesCaseCollidingTagThroughFindOrCreate_withoutThrow() {
+        // Bulk-edit shares resolveTags → findOrCreate; this guards a future refactor that bypasses it.
+        Tag parent = persistTag("Weihnachten", "Weihnachten", null);
+        Tag child = persistTag("weihnachten", "Weihnachten/weihnachten", parent.getId());
+        Document doc = documentRepository.save(Document.builder()
+                .originalFilename("C-7302")
+                .title("Brief")
+                .status(DocumentStatus.UPLOADED)
+                .build());
+
+        DocumentBulkEditDTO dto = new DocumentBulkEditDTO();
+        dto.setTagNames(List.of("weihnachten"));
+
+        assertThatCode(() -> documentService.applyBulkEditToDocument(doc.getId(), dto, null))
+                .doesNotThrowAnyException();
+
+        Set<Tag> tags = documentRepository.findById(doc.getId()).orElseThrow().getTags();
+        assertThat(tags).hasSize(1);
+        assertThat(tags.iterator().next().getId()).isEqualTo(child.getId());
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/document/annotation/AnnotationControllerTest.java

@@ -83,6 +83,15 @@ class AnnotationControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void createAnnotation_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        mockMvc.perform(post("/api/documents/" + UUID.randomUUID() + "/annotations").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content(ANNOTATION_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void createAnnotation_returns201_whenHasWriteAllPermission() throws Exception {
@@ -190,6 +199,15 @@ class AnnotationControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void patchAnnotation_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        mockMvc.perform(patch("/api/documents/" + UUID.randomUUID() + "/annotations/" + UUID.randomUUID()).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content(PATCH_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void patchAnnotation_returns200_withWriteAllPermission() throws Exception {

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

@@ -94,6 +94,15 @@ class CommentControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void postBlockComment_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        UUID blockId = UUID.randomUUID();
+        mockMvc.perform(post("/api/documents/" + DOC_ID + "/transcription-blocks/" + blockId + "/comments").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON).content(COMMENT_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "ANNOTATE_ALL")
     void postBlockComment_returns201_whenHasAnnotatePermission() throws Exception {
@@ -142,6 +151,16 @@ class CommentControllerTest {
                 .andExpect(status().isUnauthorized());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void replyToBlockComment_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        UUID blockId = UUID.randomUUID();
+        mockMvc.perform(post("/api/documents/" + DOC_ID + "/transcription-blocks/" + blockId
+                                + "/comments/" + COMMENT_ID + "/replies").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON).content(COMMENT_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "ANNOTATE_ALL")
     void replyToBlockComment_returns201_whenHasPermission() throws Exception {
@@ -181,6 +200,14 @@ class CommentControllerTest {
                 .andExpect(status().isUnauthorized());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void editComment_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        mockMvc.perform(patch("/api/documents/" + DOC_ID + "/comments/" + COMMENT_ID).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON).content(COMMENT_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "ANNOTATE_ALL")
     void editComment_returns200_whenHasPermission() throws Exception {

backend/src/test/java/org/raddatz/familienarchiv/document/transcription/TranscriptionBlockControllerTest.java

@@ -159,6 +159,15 @@ class TranscriptionBlockControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void createBlock_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        mockMvc.perform(post(URL_BASE).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content(CREATE_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void createBlock_returns201_withSavedBlock_whenAuthorised() throws Exception {
@@ -233,6 +242,15 @@ class TranscriptionBlockControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void updateBlock_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        mockMvc.perform(put(URL_BLOCK).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content(UPDATE_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void updateBlock_returns200_withUpdatedBlock_whenAuthorised() throws Exception {
@@ -363,6 +381,15 @@ class TranscriptionBlockControllerTest {
                 .andExpect(status().isForbidden());
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void reorderBlocks_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        mockMvc.perform(put(URL_REORDER).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content(REORDER_JSON))
+                .andExpect(status().isForbidden());
+    }
+
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void reorderBlocks_returns200_withReorderedBlocks_whenAuthorised() throws Exception {
@@ -440,6 +467,14 @@ class TranscriptionBlockControllerTest {
                 .andExpect(jsonPath("$.reviewed").value(true));
     }
 
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void reviewBlock_returns403_whenUserHasOnlyReadAllPermission() throws Exception {
+        mockMvc.perform(put("/api/documents/{documentId}/transcription-blocks/{blockId}/review",
+                        DOC_ID, BLOCK_ID).with(csrf()))
+                .andExpect(status().isForbidden());
+    }
+
     // ─── PUT .../review-all ───────────────────────────────────────────────────
 
     private static final String URL_REVIEW_ALL = URL_BASE + "/review-all";

backend/src/test/java/org/raddatz/familienarchiv/document/transcription/TranscriptionBlockMentionsRepositoryTest.java

@@ -12,6 +12,8 @@ import org.raddatz.familienarchiv.document.annotation.DocumentAnnotation;
 import org.raddatz.familienarchiv.document.DocumentStatus;
 import org.raddatz.familienarchiv.document.transcription.PersonMention;
 import org.raddatz.familienarchiv.document.transcription.TranscriptionBlock;
+import org.raddatz.familienarchiv.person.Person;
+import org.raddatz.familienarchiv.person.PersonRepository;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.boot.jdbc.test.autoconfigure.AutoConfigureTestDatabase;
 import org.springframework.boot.data.jpa.test.autoconfigure.DataJpaTest;
@@ -30,6 +32,7 @@ class TranscriptionBlockMentionsRepositoryTest {
     @Autowired TranscriptionBlockRepository blockRepository;
     @Autowired DocumentRepository documentRepository;
     @Autowired AnnotationRepository annotationRepository;
+    @Autowired PersonRepository personRepository;
     @Autowired EntityManager em;
 
     private UUID documentId;
@@ -55,8 +58,9 @@ class TranscriptionBlockMentionsRepositoryTest {
 
     @Test
     void mentionedPersons_roundTripsTwoEntries() {
-        UUID auguste = UUID.randomUUID();
-        UUID hermann = UUID.randomUUID();
+        // person_id is a real FK since V71 — the mentioned persons must exist.
+        UUID auguste = personRepository.save(Person.builder().firstName("Auguste").lastName("Raddatz").build()).getId();
+        UUID hermann = personRepository.save(Person.builder().firstName("Hermann").lastName("Müller").build()).getId();
 
         TranscriptionBlock saved = blockRepository.saveAndFlush(TranscriptionBlock.builder()
                 .annotationId(annotationId)
@@ -97,8 +101,9 @@ class TranscriptionBlockMentionsRepositoryTest {
 
     @Test
     void findByPersonIdWithMentionsFetched_returnsOnlyBlocksReferencingPerson_withMentionsLoaded() {
-        UUID augusteId = UUID.randomUUID();
-        UUID hermannId = UUID.randomUUID();
+        // person_id is a real FK since V71 — the mentioned persons must exist.
+        UUID augusteId = personRepository.save(Person.builder().firstName("Auguste").lastName("Raddatz").build()).getId();
+        UUID hermannId = personRepository.save(Person.builder().firstName("Hermann").lastName("Müller").build()).getId();
 
         blockRepository.saveAndFlush(TranscriptionBlock.builder()
                 .annotationId(annotationId).documentId(documentId)

backend/src/test/java/org/raddatz/familienarchiv/document/transcription/TranscriptionBlockQueryServiceTest.java

@@ -0,0 +1,35 @@
+package org.raddatz.familienarchiv.document.transcription;
+
+import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.extension.ExtendWith;
+import org.mockito.InjectMocks;
+import org.mockito.Mock;
+import org.mockito.junit.jupiter.MockitoExtension;
+
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.mockito.Mockito.when;
+
+@ExtendWith(MockitoExtension.class)
+class TranscriptionBlockQueryServiceTest {
+
+    @Mock TranscriptionBlockRepository blockRepository;
+    @InjectMocks TranscriptionBlockQueryService queryService;
+
+    @Test
+    void hasBlocks_returns_true_when_a_block_exists() {
+        UUID documentId = UUID.randomUUID();
+        when(blockRepository.existsByDocumentId(documentId)).thenReturn(true);
+
+        assertThat(queryService.hasBlocks(documentId)).isTrue();
+    }
+
+    @Test
+    void hasBlocks_returns_false_when_no_block_exists() {
+        UUID documentId = UUID.randomUUID();
+        when(blockRepository.existsByDocumentId(documentId)).thenReturn(false);
+
+        assertThat(queryService.hasBlocks(documentId)).isFalse();
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/document/transcription/TranscriptionBlockRepositoryIntegrationTest.java

@@ -102,4 +102,22 @@ class TranscriptionBlockRepositoryIntegrationTest {
         assertThat(byDoc).containsEntry(DOC_A, 100);
         assertThat(byDoc).containsEntry(DOC_B, 0);
     }
+
+    @Test
+    @Sql(statements = {
+        "INSERT INTO documents (id, title, original_filename, status) VALUES ('aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa', 'Doc A', 'a.pdf', 'PLACEHOLDER')",
+        "INSERT INTO document_annotations (id, document_id, page_number, x, y, width, height, color) VALUES ('cccccccc-cccc-cccc-cccc-cccccccccccc', 'aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa', 1, 0, 0, 1, 1, '#fff')",
+        "INSERT INTO transcription_blocks (annotation_id, document_id, sort_order, reviewed) VALUES ('cccccccc-cccc-cccc-cccc-cccccccccccc', 'aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa', 0, false)"
+    })
+    void existsByDocumentId_returns_true_when_document_has_a_block() {
+        assertThat(repository.existsByDocumentId(DOC_A)).isTrue();
+    }
+
+    @Test
+    @Sql(statements = {
+        "INSERT INTO documents (id, title, original_filename, status) VALUES ('aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa', 'Doc A', 'a.pdf', 'PLACEHOLDER')"
+    })
+    void existsByDocumentId_returns_false_when_document_has_no_blocks() {
+        assertThat(repository.existsByDocumentId(DOC_A)).isFalse();
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/exception/GlobalExceptionHandlerTest.java

@@ -1,11 +1,18 @@
 package org.raddatz.familienarchiv.exception;
 
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
 import io.sentry.Sentry;
 import org.junit.jupiter.api.Test;
 import org.junit.jupiter.api.extension.ExtendWith;
 import org.mockito.InjectMocks;
 import org.mockito.MockedStatic;
 import org.mockito.junit.jupiter.MockitoExtension;
+import org.slf4j.LoggerFactory;
+import org.springframework.dao.DataIntegrityViolationException;
+import org.springframework.dao.IncorrectResultSizeDataAccessException;
 import org.springframework.http.ResponseEntity;
 
 import static org.assertj.core.api.Assertions.assertThat;
@@ -30,4 +37,108 @@ class GlobalExceptionHandlerTest {
             assertThat(response.getBody().code()).isEqualTo(ErrorCode.INTERNAL_ERROR);
         }
     }
+
+    @Test
+    void handleGeneric_incorrectResultSize_staysOpaque_noHibernateOrRowCountLeak() {
+        // #731: before the fix, a case-colliding alias/name made Hibernate throw
+        // NonUniqueResultException → IncorrectResultSizeDataAccessException, which has no
+        // dedicated handler and falls through to handleGeneric. The fix removes the throw, but
+        // this pins the handler: a stray one must stay opaque — no Hibernate class name, no SQL,
+        // no "2 results were returned" row count reaching the client (CWE-209).
+        IncorrectResultSizeDataAccessException ex = new IncorrectResultSizeDataAccessException(
+                "query did not return a unique result: 2 results were returned", 1, 2);
+
+        try (MockedStatic<Sentry> sentryMock = mockStatic(Sentry.class)) {
+            ResponseEntity<GlobalExceptionHandler.ErrorResponse> response = handler.handleGeneric(ex);
+
+            assertThat(response.getStatusCode().value()).isEqualTo(500);
+            assertThat(response.getBody()).isNotNull();
+            assertThat(response.getBody().code()).isEqualTo(ErrorCode.INTERNAL_ERROR);
+            assertThat(response.getBody().message())
+                    .isEqualTo("An unexpected error occurred")
+                    .doesNotContain("results were returned")
+                    .doesNotContain("NonUnique")
+                    .doesNotContain("IncorrectResultSize");
+        }
+    }
+
+    @Test
+    void handleDataIntegrityViolation_returns400_withoutLeakingConstraint_orSentry() {
+        // A DataIntegrityViolationException carries the constraint name + SQL in its message;
+        // the response and logs must never echo it (CWE-209). It must become a clean 400, not a 500.
+        DataIntegrityViolationException ex = new DataIntegrityViolationException(
+                "could not execute statement; constraint [chk_meta_date_end_after_start]; "
+                        + "column meta_date_end of relation documents");
+
+        Logger handlerLogger = (Logger) LoggerFactory.getLogger(GlobalExceptionHandler.class);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        handlerLogger.addAppender(appender);
+
+        try (MockedStatic<Sentry> sentryMock = mockStatic(Sentry.class)) {
+            ResponseEntity<GlobalExceptionHandler.ErrorResponse> response =
+                    handler.handleDataIntegrityViolation(ex);
+
+            assertThat(response.getStatusCode().value()).isEqualTo(400);
+            assertThat(response.getBody()).isNotNull();
+            assertThat(response.getBody().code()).isEqualTo(ErrorCode.VALIDATION_ERROR);
+            assertThat(response.getBody().message())
+                    .doesNotContain("chk_")
+                    .doesNotContain("meta_date");
+
+            // Defense-in-depth: an unanticipated integrity violation is not a system fault,
+            // so it must NOT fabricate a Sentry alert.
+            sentryMock.verifyNoInteractions();
+        } finally {
+            handlerLogger.detachAppender(appender);
+        }
+
+        assertThat(appender.list)
+                .as("logs a WARN line")
+                .anySatisfy(e -> assertThat(e.getLevel()).isEqualTo(Level.WARN));
+        assertThat(appender.list)
+                .as("never logs the SQL statement / values (would re-leak to Loki)")
+                .noneSatisfy(e -> {
+                    assertThat(e.getFormattedMessage()).contains("could not execute statement");
+                });
+    }
+
+    @Test
+    void handleDataIntegrityViolation_logsConstraintName_butNotTheSql() {
+        // Debuggability (DevOps): the WARN must name *which* constraint fired so an
+        // unanticipated violation isn't a silent mystery — but it must carry the name only,
+        // never the SQL statement or the offending values that the SQLException message holds.
+        java.sql.SQLException sql = new java.sql.SQLException(
+                "ERROR: violates check constraint; could not execute statement; values (1917-01-10)");
+        org.hibernate.exception.ConstraintViolationException cve =
+                new org.hibernate.exception.ConstraintViolationException(
+                        "constraint violation", sql, "chk_meta_date_end_after_start");
+        DataIntegrityViolationException ex = new DataIntegrityViolationException("wrapper", cve);
+
+        Logger handlerLogger = (Logger) LoggerFactory.getLogger(GlobalExceptionHandler.class);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        handlerLogger.addAppender(appender);
+
+        try (MockedStatic<Sentry> sentryMock = mockStatic(Sentry.class)) {
+            ResponseEntity<GlobalExceptionHandler.ErrorResponse> response =
+                    handler.handleDataIntegrityViolation(ex);
+
+            // Response stays generic and leak-free (CWE-209) regardless of what we log.
+            assertThat(response.getStatusCode().value()).isEqualTo(400);
+            assertThat(response.getBody().message())
+                    .doesNotContain("chk_")
+                    .doesNotContain("meta_date");
+            sentryMock.verifyNoInteractions();
+        } finally {
+            handlerLogger.detachAppender(appender);
+        }
+
+        assertThat(appender.list)
+                .as("WARN names the constraint for debuggability")
+                .anySatisfy(e -> assertThat(e.getFormattedMessage()).contains("chk_meta_date_end_after_start"));
+        assertThat(appender.list)
+                .as("but never the SQL statement or values")
+                .noneSatisfy(e -> assertThat(e.getFormattedMessage()).contains("could not execute statement"));
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/importing/CanonicalImportOrchestratorTest.java

@@ -7,12 +7,18 @@ import org.mockito.InOrder;
 import org.mockito.Mock;
 import org.mockito.junit.jupiter.MockitoExtension;
 import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.person.relationship.RelationType;
+import org.raddatz.familienarchiv.person.relationship.RelationshipService;
+import org.raddatz.familienarchiv.person.relationship.dto.NetworkDTO;
+import org.raddatz.familienarchiv.person.relationship.dto.PersonNodeDTO;
+import org.raddatz.familienarchiv.person.relationship.dto.RelationshipDTO;
 import org.springframework.test.util.ReflectionTestUtils;
 
 import java.io.File;
 import java.nio.file.Files;
 import java.nio.file.Path;
 import java.util.List;
+import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
 import static org.assertj.core.api.Assertions.assertThatThrownBy;
@@ -29,10 +35,12 @@ class CanonicalImportOrchestratorTest {
     @Mock PersonRegisterImporter personRegisterImporter;
     @Mock PersonTreeImporter personTreeImporter;
     @Mock DocumentImporter documentImporter;
+    @Mock RelationshipService relationshipService;
 
     private CanonicalImportOrchestrator orchestrator(Path dir) {
         CanonicalImportOrchestrator o = new CanonicalImportOrchestrator(
-                tagTreeImporter, personRegisterImporter, personTreeImporter, documentImporter);
+                tagTreeImporter, personRegisterImporter, personTreeImporter, documentImporter,
+                relationshipService);
         ReflectionTestUtils.setField(o, "canonicalDir", dir.toString());
         return o;
     }
@@ -53,6 +61,7 @@ class CanonicalImportOrchestratorTest {
     void runImport_loadsTagsAndPersonsBeforeDocuments(@TempDir Path dir) throws Exception {
         writeAllArtifacts(dir);
         when(documentImporter.load(any())).thenReturn(new DocumentImporter.LoadResult(0, List.of()));
+        when(relationshipService.getFamilyNetwork()).thenReturn(new NetworkDTO(List.of(), List.of()));
         CanonicalImportOrchestrator o = orchestrator(dir);
 
         o.runImport();
@@ -68,6 +77,7 @@ class CanonicalImportOrchestratorTest {
     void runImport_setsStatusDone_onSuccess(@TempDir Path dir) throws Exception {
         writeAllArtifacts(dir);
         when(documentImporter.load(any())).thenReturn(new DocumentImporter.LoadResult(3, List.of()));
+        when(relationshipService.getFamilyNetwork()).thenReturn(new NetworkDTO(List.of(), List.of()));
         CanonicalImportOrchestrator o = orchestrator(dir);
 
         o.runImport();
@@ -118,6 +128,7 @@ class CanonicalImportOrchestratorTest {
         writeAllArtifacts(dir);
         when(documentImporter.load(any())).thenReturn(new DocumentImporter.LoadResult(1,
                 List.of(new ImportStatus.SkippedFile("fake.pdf", ImportStatus.SkipReason.INVALID_PDF_SIGNATURE))));
+        when(relationshipService.getFamilyNetwork()).thenReturn(new NetworkDTO(List.of(), List.of()));
         CanonicalImportOrchestrator o = orchestrator(dir);
 
         o.runImport();
@@ -127,4 +138,46 @@ class CanonicalImportOrchestratorTest {
                 .extracting(ImportStatus.SkippedFile::filename)
                 .containsExactly("fake.pdf");
     }
+
+    // ─── generation monotonicity soft-check (#689) ─────────────────────────────
+
+    @Test
+    void runImport_invokesGetFamilyNetwork_afterPersonLoaders_beforeDocuments(@TempDir Path dir) throws Exception {
+        writeAllArtifacts(dir);
+        when(documentImporter.load(any())).thenReturn(new DocumentImporter.LoadResult(0, List.of()));
+        when(relationshipService.getFamilyNetwork()).thenReturn(new NetworkDTO(List.of(), List.of()));
+        CanonicalImportOrchestrator o = orchestrator(dir);
+
+        o.runImport();
+
+        InOrder order = inOrder(personRegisterImporter, personTreeImporter, relationshipService, documentImporter);
+        order.verify(personRegisterImporter).load(any());
+        order.verify(personTreeImporter).load(any());
+        order.verify(relationshipService).getFamilyNetwork();
+        order.verify(documentImporter).load(any());
+    }
+
+    @Test
+    void runImport_completes_evenWhenMonotonicityViolatingEdgePresent(@TempDir Path dir) throws Exception {
+        // child.generation (2) <= parent.generation (3) — monotonicity violation.
+        // The orchestrator must WARN and continue; it must not abort or fail-closed.
+        writeAllArtifacts(dir);
+        UUID parentId = UUID.randomUUID();
+        UUID childId = UUID.randomUUID();
+        PersonNodeDTO parent = new PersonNodeDTO(parentId, "Parent", null, null, 3, true);
+        PersonNodeDTO child = new PersonNodeDTO(childId, "Child", null, null, 2, true);
+        RelationshipDTO edge = new RelationshipDTO(
+                UUID.randomUUID(), parentId, childId,
+                "Parent", null, null, "Child", null, null,
+                RelationType.PARENT_OF, null, null, null);
+        when(relationshipService.getFamilyNetwork())
+                .thenReturn(new NetworkDTO(List.of(parent, child), List.of(edge)));
+        when(documentImporter.load(any())).thenReturn(new DocumentImporter.LoadResult(0, List.of()));
+        CanonicalImportOrchestrator o = orchestrator(dir);
+
+        o.runImport();
+
+        assertThat(o.getStatus().state()).isEqualTo(ImportStatus.State.DONE);
+        verify(documentImporter).load(any());
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/importing/DocumentImporterTest.java

@@ -11,6 +11,7 @@ import org.mockito.Mock;
 import org.mockito.junit.jupiter.MockitoExtension;
 import org.raddatz.familienarchiv.document.Document;
 import org.raddatz.familienarchiv.document.DocumentService;
+import org.raddatz.familienarchiv.document.DocumentTitleFactory;
 import org.raddatz.familienarchiv.document.DocumentStatus;
 import org.raddatz.familienarchiv.document.ThumbnailAsyncRunner;
 import org.raddatz.familienarchiv.person.Person;
@@ -49,12 +50,20 @@ class DocumentImporterTest {
     @Mock TagService tagService;
     @Mock S3Client s3Client;
     @Mock ThumbnailAsyncRunner thumbnailAsyncRunner;
+    @Mock FileStreamOpener fileStreamOpener;
 
     DocumentImporter importer;
 
     @BeforeEach
-    void setUp() {
-        importer = new DocumentImporter(documentService, personService, tagService, s3Client, thumbnailAsyncRunner);
+    void setUp() throws java.io.IOException {
+        // Default opener delegates to FileInputStream — tests that need to force an IOException
+        // override this stub locally (load_skipsFile_whenMagicByteCheckThrowsIoException).
+        lenient().when(fileStreamOpener.open(any(File.class)))
+                .thenAnswer(inv -> new java.io.FileInputStream(inv.getArgument(0, File.class)));
+        // Real factory (pure, dependency-free) so the title-content assertions below exercise
+        // the shared composition rather than a stub — the #726 single source of truth.
+        importer = new DocumentImporter(documentService, new DocumentTitleFactory(), personService,
+                tagService, s3Client, thumbnailAsyncRunner, fileStreamOpener);
         ReflectionTestUtils.setField(importer, "bucketName", "test-bucket");
     }
 
@@ -305,11 +314,11 @@ class DocumentImporterTest {
         lenient().when(documentService.findByOriginalFilename(any())).thenReturn(Optional.empty());
         Path xlsx = writeDocs(tempDir, docRow("W-0001", "", "", "", "", "", "", "", ""));
 
-        DocumentImporter spyImporter = org.mockito.Mockito.spy(importer);
-        org.mockito.Mockito.doThrow(new java.io.IOException("read error"))
-                .when(spyImporter).openFileStream(any(File.class));
+        // FileStreamOpener is injected — stub it to throw, no spy on the importer needed.
+        org.mockito.Mockito.when(fileStreamOpener.open(any(File.class)))
+                .thenThrow(new java.io.IOException("read error"));
 
-        DocumentImporter.LoadResult result = spyImporter.load(xlsx.toFile());
+        DocumentImporter.LoadResult result = importer.load(xlsx.toFile());
 
         assertThat(result.skippedFiles())
                 .extracting(ImportStatus.SkippedFile::reason)
@@ -424,6 +433,60 @@ class DocumentImporterTest {
                         && "Herbert Cram|Clara".equals(d.getReceiverText())));
     }
 
+    @Test
+    void load_provisionalReceiverUsesHumanNameFromReceiverNames_notSlug(@TempDir Path tempDir) throws Exception {
+        // Regression: resolveReceivers used to pass the slug as both `sourceRef` AND `lastName`,
+        // so an unresolved receiver "smith-john" became a provisional Person with
+        // lastName="smith-john". The fix consumes the parallel `receiver_names` column.
+        ReflectionTestUtils.setField(importer, "importDir", tempDir.toString());
+        Person provisional = Person.builder().id(UUID.randomUUID()).sourceRef("smith-john")
+                .lastName("John Smith").provisional(true).build();
+        when(documentService.findByOriginalFilename("W-0050")).thenReturn(Optional.empty());
+        when(documentService.save(any())).thenAnswer(inv -> inv.getArgument(0));
+        when(personService.findBySourceRef("smith-john")).thenReturn(Optional.empty());
+        when(personService.upsertBySourceRef(any())).thenReturn(provisional);
+        Path xlsx = writeDocs(tempDir, docRow("W-0050", "", "",
+                "smith-john", "John Smith", "", "", "", ""));
+
+        importer.load(xlsx.toFile());
+
+        org.mockito.ArgumentCaptor<PersonUpsertCommand> captor =
+                org.mockito.ArgumentCaptor.forClass(PersonUpsertCommand.class);
+        verify(personService).upsertBySourceRef(captor.capture());
+        assertThat(captor.getValue().sourceRef()).isEqualTo("smith-john");
+        assertThat(captor.getValue().lastName()).isEqualTo("John Smith");
+        assertThat(captor.getValue().provisional()).isTrue();
+    }
+
+    @Test
+    void load_provisionalReceiverFallsBackToSlug_whenNamesListShorterThanSlugs(@TempDir Path tempDir) throws Exception {
+        // Parallel-list zip: if the names list is shorter than the slugs list, slugs without a
+        // matching name fall back to slug as the display name. This is the "missing name" case
+        // (rare in canonical data but the contract must define it).
+        ReflectionTestUtils.setField(importer, "importDir", tempDir.toString());
+        Person alice = Person.builder().id(UUID.randomUUID()).sourceRef("alice-jones")
+                .lastName("Alice Jones").provisional(true).build();
+        Person bob = Person.builder().id(UUID.randomUUID()).sourceRef("bob-roe")
+                .lastName("bob-roe").provisional(true).build();
+        when(documentService.findByOriginalFilename("W-0051")).thenReturn(Optional.empty());
+        when(documentService.save(any())).thenAnswer(inv -> inv.getArgument(0));
+        when(personService.findBySourceRef("alice-jones")).thenReturn(Optional.empty());
+        when(personService.findBySourceRef("bob-roe")).thenReturn(Optional.empty());
+        when(personService.upsertBySourceRef(any())).thenReturn(alice).thenReturn(bob);
+        Path xlsx = writeDocs(tempDir, docRow("W-0051", "", "",
+                "alice-jones|bob-roe", "Alice Jones", "", "", "", ""));
+
+        importer.load(xlsx.toFile());
+
+        org.mockito.ArgumentCaptor<PersonUpsertCommand> captor =
+                org.mockito.ArgumentCaptor.forClass(PersonUpsertCommand.class);
+        verify(personService, org.mockito.Mockito.times(2)).upsertBySourceRef(captor.capture());
+        assertThat(captor.getAllValues()).extracting(PersonUpsertCommand::sourceRef)
+                .containsExactly("alice-jones", "bob-roe");
+        assertThat(captor.getAllValues()).extracting(PersonUpsertCommand::lastName)
+                .containsExactly("Alice Jones", "bob-roe");
+    }
+
     // ─── clean date values parse without semantic logic ──────────────────────────────
 
     @Test

backend/src/test/java/org/raddatz/familienarchiv/importing/PersonRegisterImporterTest.java

@@ -6,6 +6,8 @@ import org.apache.poi.xssf.usermodel.XSSFWorkbook;
 import org.junit.jupiter.api.Test;
 import org.junit.jupiter.api.extension.ExtendWith;
 import org.junit.jupiter.api.io.TempDir;
+import org.junit.jupiter.params.ParameterizedTest;
+import org.junit.jupiter.params.provider.CsvSource;
 import org.mockito.ArgumentCaptor;
 import org.mockito.junit.jupiter.MockitoExtension;
 import org.raddatz.familienarchiv.person.Person;
@@ -87,6 +89,50 @@ class PersonRegisterImporterTest {
         assertThat(processed).isEqualTo(2);
     }
 
+    // ─── generation parsing (#689) ─────────────────────────────────────────────
+
+    @ParameterizedTest
+    @CsvSource(value = {
+            "'G 3', 3",
+            "'G3', 3",
+            "'G  3', 3",
+            "'3', 3",
+            "'  3  ', 3",
+            "'G 2  de Gruyter', 2",
+            "'', null",
+            "'garbage', null",
+            "'G 99', null",
+            "'G -1', null"
+    }, nullValues = "null")
+    void load_parsesGeneration_perRegex(String raw, Integer expected, @TempDir Path tempDir) throws Exception {
+        PersonService personService = mock(PersonService.class);
+        when(personService.upsertBySourceRef(any())).thenAnswer(inv -> personOf(inv.getArgument(0)));
+        Path xlsx = writePersonsWithGeneration(tempDir,
+                rowWithGeneration("herbert-cram", "Cram", "Herbert", "", "", "False", raw));
+
+        new PersonRegisterImporter(personService).load(xlsx.toFile());
+
+        ArgumentCaptor<PersonUpsertCommand> captor = ArgumentCaptor.forClass(PersonUpsertCommand.class);
+        verify(personService).upsertBySourceRef(captor.capture());
+        assertThat(captor.getValue().generation()).isEqualTo(expected);
+    }
+
+    @Test
+    void load_succeeds_andLeavesGenerationNull_whenArtifactHasNoGenerationColumn(@TempDir Path tempDir) throws Exception {
+        // REQ-IMP-001: older artifacts without the `generation` column must still
+        // import. REQUIRED_HEADERS is intentionally not extended.
+        PersonService personService = mock(PersonService.class);
+        when(personService.upsertBySourceRef(any())).thenAnswer(inv -> personOf(inv.getArgument(0)));
+        Path xlsx = writePersons(tempDir, row(
+                "old-artifact", "Mueller", "Hans", "", "", "False"));
+
+        new PersonRegisterImporter(personService).load(xlsx.toFile());
+
+        ArgumentCaptor<PersonUpsertCommand> captor = ArgumentCaptor.forClass(PersonUpsertCommand.class);
+        verify(personService).upsertBySourceRef(captor.capture());
+        assertThat(captor.getValue().generation()).isNull();
+    }
+
     private static Person personOf(PersonUpsertCommand cmd) {
         return Person.builder().id(UUID.randomUUID()).sourceRef(cmd.sourceRef())
                 .firstName(cmd.firstName()).lastName(cmd.lastName())
@@ -127,4 +173,36 @@ class PersonRegisterImporterTest {
         }
         return xlsx;
     }
+
+    private Map<String, String> rowWithGeneration(String personId, String lastName, String firstName,
+                                                  String maidenName, String notes, String provisional,
+                                                  String generation) {
+        Map<String, String> r = row(personId, lastName, firstName, maidenName, notes, provisional);
+        r.put("generation", generation);
+        return r;
+    }
+
+    @SafeVarargs
+    private Path writePersonsWithGeneration(Path dir, Map<String, String>... rows) throws Exception {
+        Path xlsx = dir.resolve("canonical-persons.xlsx");
+        List<String> headers = List.of(
+                "person_id", "last_name", "first_name", "maiden_name", "notes", "provisional", "generation");
+        try (XSSFWorkbook wb = new XSSFWorkbook()) {
+            Sheet sheet = wb.createSheet("Sheet1");
+            Row header = sheet.createRow(0);
+            for (int i = 0; i < headers.size(); i++) {
+                header.createCell(i).setCellValue(headers.get(i));
+            }
+            for (int r = 0; r < rows.length; r++) {
+                Row row = sheet.createRow(r + 1);
+                for (int c = 0; c < headers.size(); c++) {
+                    row.createCell(c).setCellValue(rows[r].getOrDefault(headers.get(c), ""));
+                }
+            }
+            try (OutputStream out = Files.newOutputStream(xlsx)) {
+                wb.write(out);
+            }
+        }
+        return xlsx;
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/importing/PersonTreeImporterTest.java

@@ -151,6 +151,65 @@ class PersonTreeImporterTest {
         verify(relationshipService, org.mockito.Mockito.never()).addRelationship(any(), any());
     }
 
+    // ─── generation (#689) ─────────────────────────────────────────────────────
+
+    @Test
+    void load_passesGenerationFromJson(@TempDir Path tempDir) throws Exception {
+        PersonService personService = mock(PersonService.class);
+        RelationshipService relationshipService = mock(RelationshipService.class);
+        when(personService.upsertBySourceRef(any())).thenAnswer(inv -> personOf(inv.getArgument(0)));
+        Path json = write(tempDir, """
+                {"persons":[
+                  {"rowId":"row_a","lastName":"Cram","firstName":"Herbert","familyMember":true,
+                   "personId":"herbert-cram","generation":3}
+                ],"relationships":[]}
+                """);
+
+        new PersonTreeImporter(personService, relationshipService).load(json.toFile());
+
+        ArgumentCaptor<PersonUpsertCommand> captor = ArgumentCaptor.forClass(PersonUpsertCommand.class);
+        verify(personService).upsertBySourceRef(captor.capture());
+        assertThat(captor.getValue().generation()).isEqualTo(3);
+    }
+
+    @Test
+    void load_returnsNullGeneration_whenAbsentFromJson(@TempDir Path tempDir) throws Exception {
+        PersonService personService = mock(PersonService.class);
+        RelationshipService relationshipService = mock(RelationshipService.class);
+        when(personService.upsertBySourceRef(any())).thenAnswer(inv -> personOf(inv.getArgument(0)));
+        Path json = write(tempDir, """
+                {"persons":[
+                  {"rowId":"row_a","lastName":"Cram","firstName":"Herbert","familyMember":true,
+                   "personId":"herbert-cram"}
+                ],"relationships":[]}
+                """);
+
+        new PersonTreeImporter(personService, relationshipService).load(json.toFile());
+
+        ArgumentCaptor<PersonUpsertCommand> captor = ArgumentCaptor.forClass(PersonUpsertCommand.class);
+        verify(personService).upsertBySourceRef(captor.capture());
+        assertThat(captor.getValue().generation()).isNull();
+    }
+
+    @Test
+    void load_skipsOutOfRangeGeneration_logsWarn_neverAborts(@TempDir Path tempDir) throws Exception {
+        PersonService personService = mock(PersonService.class);
+        RelationshipService relationshipService = mock(RelationshipService.class);
+        when(personService.upsertBySourceRef(any())).thenAnswer(inv -> personOf(inv.getArgument(0)));
+        Path json = write(tempDir, """
+                {"persons":[
+                  {"rowId":"row_a","lastName":"Cram","firstName":"Herbert","familyMember":true,
+                   "personId":"herbert-cram","generation":99}
+                ],"relationships":[]}
+                """);
+
+        new PersonTreeImporter(personService, relationshipService).load(json.toFile());
+
+        ArgumentCaptor<PersonUpsertCommand> captor = ArgumentCaptor.forClass(PersonUpsertCommand.class);
+        verify(personService).upsertBySourceRef(captor.capture());
+        assertThat(captor.getValue().generation()).isNull();
+    }
+
     private static Person personOf(PersonUpsertCommand cmd) {
         return Person.builder().id(UUID.randomUUID()).sourceRef(cmd.sourceRef()).lastName(cmd.lastName()).build();
     }

backend/src/test/java/org/raddatz/familienarchiv/person/PersonControllerTest.java

@@ -718,4 +718,74 @@ class PersonControllerTest {
                         .content("{\"lastName\":\"de Gruyter\"}"))
                 .andExpect(status().isBadRequest());
     }
+
+    // ─── generation field validation (#689) ────────────────────────────────────
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void updatePerson_returns400_whenGenerationAboveRange() throws Exception {
+        mockMvc.perform(put("/api/persons/{id}", UUID.randomUUID()).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"firstName\":\"Hans\",\"lastName\":\"Müller\","
+                                + "\"personType\":\"PERSON\",\"generation\":11}"))
+                .andExpect(status().isBadRequest())
+                .andExpect(jsonPath("$.code").value(ErrorCode.VALIDATION_ERROR.name()));
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void updatePerson_returns400_whenGenerationBelowRange() throws Exception {
+        mockMvc.perform(put("/api/persons/{id}", UUID.randomUUID()).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"firstName\":\"Hans\",\"lastName\":\"Müller\","
+                                + "\"personType\":\"PERSON\",\"generation\":-1}"))
+                .andExpect(status().isBadRequest())
+                .andExpect(jsonPath("$.code").value(ErrorCode.VALIDATION_ERROR.name()));
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void updatePerson_returns200_whenGenerationNull() throws Exception {
+        // Symmetric body assertion: the response must echo generation as null (not
+        // absent), so the frontend re-hydrates the "(none)" option after a clear.
+        // Without this, the in-range test below would be the only end-to-end proof
+        // that the field flows through the controller.
+        Person saved = Person.builder().id(UUID.randomUUID()).firstName("Hans").lastName("Müller").build();
+        when(personService.updatePerson(any(), any())).thenReturn(saved);
+
+        mockMvc.perform(put("/api/persons/{id}", UUID.randomUUID()).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"firstName\":\"Hans\",\"lastName\":\"Müller\","
+                                + "\"personType\":\"PERSON\",\"generation\":null}"))
+                .andExpect(status().isOk())
+                .andExpect(jsonPath("$.generation").value(org.hamcrest.Matchers.nullValue()));
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void updatePerson_returns200_whenGenerationInRange() throws Exception {
+        Person saved = Person.builder().id(UUID.randomUUID()).firstName("Hans").lastName("Müller").generation(3).build();
+        when(personService.updatePerson(any(), any())).thenReturn(saved);
+
+        mockMvc.perform(put("/api/persons/{id}", UUID.randomUUID()).with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"firstName\":\"Hans\",\"lastName\":\"Müller\","
+                                + "\"personType\":\"PERSON\",\"generation\":3}"))
+                .andExpect(status().isOk())
+                .andExpect(jsonPath("$.generation").value(3));
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void createPerson_returns200_whenGenerationInRange() throws Exception {
+        Person saved = Person.builder().id(UUID.randomUUID()).firstName("Hans").lastName("Müller").generation(3).build();
+        when(personService.createPerson(any(org.raddatz.familienarchiv.person.PersonUpdateDTO.class))).thenReturn(saved);
+
+        mockMvc.perform(post("/api/persons").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"firstName\":\"Hans\",\"lastName\":\"Müller\","
+                                + "\"personType\":\"PERSON\",\"generation\":3}"))
+                .andExpect(status().isOk())
+                .andExpect(jsonPath("$.generation").value(3));
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/person/PersonImportUpsertTest.java

@@ -148,4 +148,55 @@ class PersonImportUpsertTest {
 
         assertThat(result.isProvisional()).isTrue();
     }
+
+    // ─── generation (#689) ─────────────────────────────────────────────────────
+
+    @Test
+    void upsertBySourceRef_writesGeneration_onFirstImport() {
+        when(personRepository.findBySourceRef("herbert-cram")).thenReturn(Optional.empty());
+        when(personRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        PersonUpsertCommand cmd = PersonUpsertCommand.builder()
+                .sourceRef("herbert-cram").firstName("Herbert").lastName("Cram")
+                .generation(3).personType(PersonType.PERSON).provisional(false).build();
+
+        Person result = personService.upsertBySourceRef(cmd);
+
+        assertThat(result.getGeneration()).isEqualTo(3);
+    }
+
+    @Test
+    void upsertBySourceRef_preservesHumanEditedGeneration_onReimport() {
+        Person humanEdited = Person.builder()
+                .id(UUID.randomUUID()).sourceRef("herbert-cram")
+                .firstName("Herbert").lastName("Cram").generation(4).build();
+        when(personRepository.findBySourceRef("herbert-cram")).thenReturn(Optional.of(humanEdited));
+        when(personRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        PersonUpsertCommand cmd = PersonUpsertCommand.builder()
+                .sourceRef("herbert-cram").firstName("Herbert").lastName("Cram")
+                .generation(2).personType(PersonType.PERSON).provisional(false).build();
+
+        Person result = personService.upsertBySourceRef(cmd);
+
+        assertThat(result.getGeneration()).isEqualTo(4);
+    }
+
+    @Test
+    void mergeCanonical_overwrites_human_null_with_canonical_value_documenting_known_limitation() {
+        // If preferHuman gains explicit-null-vs-unset semantics, delete this test (see issue #689).
+        Person existing = Person.builder()
+                .id(UUID.randomUUID()).sourceRef("herbert-cram")
+                .firstName("Herbert").lastName("Cram").generation(null).build();
+        when(personRepository.findBySourceRef("herbert-cram")).thenReturn(Optional.of(existing));
+        when(personRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        PersonUpsertCommand cmd = PersonUpsertCommand.builder()
+                .sourceRef("herbert-cram").firstName("Herbert").lastName("Cram")
+                .generation(3).personType(PersonType.PERSON).provisional(false).build();
+
+        Person result = personService.upsertBySourceRef(cmd);
+
+        assertThat(result.getGeneration()).isEqualTo(3);
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/person/PersonRepositoryTest.java

@@ -21,6 +21,7 @@ import jakarta.persistence.PersistenceContext;
 import java.util.List;
 import java.util.Optional;
 import java.util.Set;
+import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
 
@@ -120,37 +121,60 @@ class PersonRepositoryTest {
                 .containsExactly("Anna", "Clara");
     }
 
-    // ─── findByAliasIgnoreCase ────────────────────────────────────────────────
+    // ─── findByAlias (exact) / findAllByAliasIgnoreCase (case-folding siblings) ───
 
     @Test
-    void findByAliasIgnoreCase_returnsMatchingPerson() {
+    void findByAlias_returnsExactCaseMatchOnly() {
         personRepository.save(Person.builder()
                 .firstName("Karl").lastName("Brandt").alias("Opa Karl").build());
 
-        Optional<Person> found = personRepository.findByAliasIgnoreCase("opa karl");
-
-        assertThat(found).isPresent();
-        assertThat(found.get().getFirstName()).isEqualTo("Karl");
+        assertThat(personRepository.findByAlias("Opa Karl")).isPresent();
+        assertThat(personRepository.findByAlias("opa karl")).isEmpty(); // exact-case: a folded form does NOT match
     }
 
     @Test
-    void findByAliasIgnoreCase_returnsEmpty_whenAliasDoesNotMatch() {
-        Optional<Person> found = personRepository.findByAliasIgnoreCase("nobody");
-
-        assertThat(found).isEmpty();
+    void findAllByAliasIgnoreCase_returnsEmpty_whenAliasDoesNotMatch() {
+        assertThat(personRepository.findAllByAliasIgnoreCase("nobody")).isEmpty();
     }
 
-    // ─── findByFirstNameIgnoreCaseAndLastNameIgnoreCase ───────────────────────
+    @Test
+    void findAllByAliasIgnoreCase_foldsUmlautCase_inRealPostgres() {
+        // Proves Postgres LOWER() folds ü the same way for both rows — a plain-ASCII probe would
+        // stay green even if umlaut folding regressed. Both case-colliding aliases must match.
+        personRepository.save(Person.builder().lastName("Müller").alias("Müller").build());
+        personRepository.save(Person.builder().lastName("müller").alias("müller").build());
+
+        assertThat(personRepository.findAllByAliasIgnoreCase("MÜLLER")).hasSize(2);
+    }
+
+    // ─── findByFirstNameAndLastName (exact) / findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase ───
 
     @Test
-    void findByFirstNameIgnoreCaseAndLastNameIgnoreCase_returnsMatch() {
+    void findByFirstNameAndLastName_returnsExactCaseMatchOnly() {
         personRepository.save(Person.builder().firstName("Maria").lastName("Raddatz").build());
 
-        Optional<Person> found = personRepository.findByFirstNameIgnoreCaseAndLastNameIgnoreCase(
-                "maria", "raddatz");
+        assertThat(personRepository.findByFirstNameAndLastName("Maria", "Raddatz")).isPresent();
+        assertThat(personRepository.findByFirstNameAndLastName("maria", "raddatz")).isEmpty(); // exact-case only
+    }
 
-        assertThat(found).isPresent();
-        assertThat(found.get().getFirstName()).isEqualTo("Maria");
+    @Test
+    void findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase_foldsUmlautCase_inRealPostgres() {
+        personRepository.save(Person.builder().firstName("Hans").lastName("Müller").build());
+        personRepository.save(Person.builder().firstName("hans").lastName("müller").build());
+
+        assertThat(personRepository.findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase("HANS", "MÜLLER"))
+                .hasSize(2);
+    }
+
+    @Test
+    void findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase_nullFirstName_foldsToNoMatch() {
+        // Fail-closed: a last-name-only filename (null first name) must NOT widen to first_name IS
+        // NULL and pull in the institution/last-name-only row as a "sender". Proven on real
+        // Postgres because a mocked unit test cannot catch the IS NULL vs `= NULL` semantics.
+        personRepository.save(Person.builder().lastName("Müller").build()); // first_name NULL
+
+        assertThat(personRepository.findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase(null, "Müller"))
+                .isEmpty();
     }
 
     // ─── findCorrespondents ───────────────────────────────────────────────────
@@ -366,30 +390,6 @@ class PersonRepositoryTest {
         assertThat(result).hasSize(1);
     }
 
-    // ─── deleteReceiverReferences ─────────────────────────────────────────────
-
-    @Test
-    void deleteReceiverReferences_removesPersonFromAllDocumentReceivers() {
-        Person toDelete = personRepository.save(Person.builder().firstName("Weg").lastName("Person").build());
-        Person sender   = personRepository.save(Person.builder().firstName("Send").lastName("Er").build());
-
-        Document doc1 = documentRepository.save(Document.builder()
-                .title("Brief 1").originalFilename("b1.pdf")
-                .status(DocumentStatus.UPLOADED)
-                .sender(sender).receivers(Set.of(toDelete)).build());
-        Document doc2 = documentRepository.save(Document.builder()
-                .title("Brief 2").originalFilename("b2.pdf")
-                .status(DocumentStatus.UPLOADED)
-                .sender(sender).receivers(Set.of(toDelete)).build());
-
-        personRepository.deleteReceiverReferences(toDelete.getId());
-        entityManager.flush();
-        entityManager.clear();
-
-        assertThat(documentRepository.findById(doc1.getId()).orElseThrow().getReceivers()).isEmpty();
-        assertThat(documentRepository.findById(doc2.getId()).orElseThrow().getReceivers()).isEmpty();
-    }
-
     // ─── searchByName with aliases ───────────────────────────────────────────
 
     @Test
@@ -672,4 +672,181 @@ class PersonRepositoryTest {
 
         assertThat(slice.get(0).getDocumentCount()).isEqualTo(1);
     }
+
+    // ─── generation column (#689) ──────────────────────────────────────────────
+
+    @Test
+    void save_persistsGeneration_andFindByIdReturnsSameGeneration() {
+        Person person = Person.builder()
+                .firstName("Walter")
+                .lastName("Raddatz")
+                .generation(3)
+                .build();
+
+        Person saved = personRepository.save(person);
+        entityManager.flush();
+        entityManager.clear();
+
+        Optional<Person> found = personRepository.findById(saved.getId());
+        assertThat(found).isPresent();
+        assertThat(found.get().getGeneration()).isEqualTo(3);
+    }
+
+    @Test
+    void save_allowsNullGeneration_existingRowsRemainNull() {
+        Person person = Person.builder()
+                .firstName("Anonym")
+                .lastName("Person")
+                .build();
+
+        Person saved = personRepository.save(person);
+        entityManager.flush();
+        entityManager.clear();
+
+        Optional<Person> found = personRepository.findById(saved.getId());
+        assertThat(found).isPresent();
+        assertThat(found.get().getGeneration()).isNull();
+    }
+
+    // ─── #684: ON DELETE integrity enforced at the database layer ──────────────
+    // A raw deleteById (bypassing PersonService) must keep referential integrity:
+    // documents.sender_id → SET NULL, document_receivers.person_id → CASCADE, and the
+    // transcription_block_mentioned_persons soft reference → CASCADE. These run against
+    // real Postgres because the FK ON DELETE behaviour never fires on H2.
+
+    @Test
+    void deleteById_personSenderOfAReceiverOfB_nullsSender_dropsReceiverRow_bothDocumentsSurvive() {
+        Person target = personRepository.save(Person.builder().firstName("Weg").lastName("Person").build());
+        Person bystander = personRepository.save(Person.builder().firstName("Bleibt").lastName("Hier").build());
+
+        Document sent = documentRepository.save(Document.builder()
+                .title("Gesendet").originalFilename("sent.pdf")
+                .status(DocumentStatus.UPLOADED).sender(target).build());
+        Document received = documentRepository.save(Document.builder()
+                .title("Empfangen").originalFilename("received.pdf")
+                .status(DocumentStatus.UPLOADED).sender(bystander)
+                .receivers(Set.of(target)).build());
+        entityManager.flush();
+        entityManager.clear();
+
+        personRepository.deleteById(target.getId());
+        entityManager.flush();
+        entityManager.clear();
+
+        assertThat(personRepository.findById(target.getId())).isEmpty();
+
+        Document reloadedSent = documentRepository.findById(sent.getId()).orElseThrow();
+        assertThat(reloadedSent.getSender()).isNull(); // AC-1: SET NULL
+
+        Document reloadedReceived = documentRepository.findById(received.getId()).orElseThrow();
+        assertThat(reloadedReceived.getReceivers())
+                .noneMatch(p -> p.getId().equals(target.getId())); // AC-2: CASCADE drops the join row
+
+        // Cascade-boundary guard (Nora, non-negotiable): the cascade stops at the join/reference
+        // layer — both documents themselves survive. Guards against a future migration turning
+        // documents.sender_id SET NULL into CASCADE and destroying historical letters.
+        assertThat(documentRepository.findById(sent.getId())).isPresent();
+        assertThat(documentRepository.findById(received.getId())).isPresent();
+    }
+
+    @Test
+    void deleteById_receiverWithCoReceiver_dropsOnlyDeletedPersonsJoinRow() {
+        Person target = personRepository.save(Person.builder().firstName("Weg").lastName("Person").build());
+        Person coReceiver = personRepository.save(Person.builder().firstName("Mit").lastName("Empfänger").build());
+        Person sender = personRepository.save(Person.builder().firstName("Send").lastName("Er").build());
+
+        Document doc = documentRepository.save(Document.builder()
+                .title("Brief").originalFilename("brief.pdf")
+                .status(DocumentStatus.UPLOADED).sender(sender)
+                .receivers(Set.of(target, coReceiver)).build());
+        entityManager.flush();
+        entityManager.clear();
+
+        personRepository.deleteById(target.getId());
+        entityManager.flush();
+        entityManager.clear();
+
+        Document reloaded = documentRepository.findById(doc.getId()).orElseThrow();
+        assertThat(reloaded.getReceivers()).extracting(Person::getId)
+                .containsExactly(coReceiver.getId()); // co-receiver untouched
+    }
+
+    @Test
+    void deleteById_personIsSenderAndReceiverOfSameDocument_documentSurvives_senderNull_receiverDropped() {
+        // AC-8: the trickier same-document interaction the cross-document cases don't exercise.
+        Person target = personRepository.save(Person.builder().firstName("Beides").lastName("Person").build());
+        Person coReceiver = personRepository.save(Person.builder().firstName("Mit").lastName("Empfänger").build());
+
+        Document doc = documentRepository.save(Document.builder()
+                .title("Selbstbrief").originalFilename("self.pdf")
+                .status(DocumentStatus.UPLOADED).sender(target)
+                .receivers(Set.of(target, coReceiver)).build());
+        entityManager.flush();
+        entityManager.clear();
+
+        personRepository.deleteById(target.getId());
+        entityManager.flush();
+        entityManager.clear();
+
+        Document reloaded = documentRepository.findById(doc.getId()).orElseThrow();
+        assertThat(reloaded.getSender()).isNull();
+        assertThat(reloaded.getReceivers()).extracting(Person::getId)
+                .containsExactly(coReceiver.getId());
+    }
+
+    @Test
+    void deleteById_mentionedPerson_dropsMentionRow_blockTextSurvives() {
+        // AC-3: the @-mention sidecar is a CASCADE soft reference, but the literal "@Name" lives
+        // in transcription_blocks.text and must stay visible as plain text after the person goes.
+        Person mentioned = personRepository.save(Person.builder().firstName("Auguste").lastName("Raddatz").build());
+        Person survivor = personRepository.save(Person.builder().firstName("Clara").lastName("Cram").build());
+        Document doc = documentRepository.save(Document.builder()
+                .title("Brief").originalFilename("brief.pdf")
+                .status(DocumentStatus.UPLOADED).build());
+        entityManager.flush();
+
+        UUID annotationId = UUID.randomUUID();
+        UUID blockId = UUID.randomUUID();
+        entityManager.createNativeQuery(
+                "INSERT INTO document_annotations (id, document_id, page_number, x, y, width, height, color) "
+                        + "VALUES (?1, ?2, 1, 0.1, 0.2, 0.3, 0.1, '#fff')")
+                .setParameter(1, annotationId).setParameter(2, doc.getId()).executeUpdate();
+        entityManager.createNativeQuery(
+                "INSERT INTO transcription_blocks (id, annotation_id, document_id, text) VALUES (?1, ?2, ?3, ?4)")
+                .setParameter(1, blockId).setParameter(2, annotationId).setParameter(3, doc.getId())
+                .setParameter(4, "Brief an @Auguste Raddatz und @Clara Cram").executeUpdate();
+        // Two mention rows on the same block: the deleted person and an innocent bystander.
+        entityManager.createNativeQuery(
+                "INSERT INTO transcription_block_mentioned_persons (block_id, person_id, display_name) "
+                        + "VALUES (?1, ?2, ?3)")
+                .setParameter(1, blockId).setParameter(2, mentioned.getId())
+                .setParameter(3, "Auguste Raddatz").executeUpdate();
+        entityManager.createNativeQuery(
+                "INSERT INTO transcription_block_mentioned_persons (block_id, person_id, display_name) "
+                        + "VALUES (?1, ?2, ?3)")
+                .setParameter(1, blockId).setParameter(2, survivor.getId())
+                .setParameter(3, "Clara Cram").executeUpdate();
+        entityManager.flush();
+        entityManager.clear();
+
+        personRepository.deleteById(mentioned.getId());
+        entityManager.flush();
+        entityManager.clear();
+
+        Number mentionRows = (Number) entityManager.createNativeQuery(
+                "SELECT count(*) FROM transcription_block_mentioned_persons WHERE person_id = ?1")
+                .setParameter(1, mentioned.getId()).getSingleResult();
+        assertThat(mentionRows.longValue()).isZero();
+
+        // The cascade is scoped to the deleted person — the bystander's mention row is untouched.
+        Number survivorRows = (Number) entityManager.createNativeQuery(
+                "SELECT count(*) FROM transcription_block_mentioned_persons WHERE person_id = ?1")
+                .setParameter(1, survivor.getId()).getSingleResult();
+        assertThat(survivorRows.longValue()).isEqualTo(1);
+
+        String text = (String) entityManager.createNativeQuery(
+                "SELECT text FROM transcription_blocks WHERE id = ?1")
+                .setParameter(1, blockId).getSingleResult();
+        assertThat(text).isEqualTo("Brief an @Auguste Raddatz und @Clara Cram");
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/person/PersonServiceIntegrationTest.java

@@ -4,6 +4,7 @@ import org.junit.jupiter.api.Test;
 import org.raddatz.familienarchiv.PostgresContainerConfig;
 import org.raddatz.familienarchiv.document.Document;
 import org.raddatz.familienarchiv.document.DocumentRepository;
+import org.raddatz.familienarchiv.document.DocumentService;
 import org.raddatz.familienarchiv.document.DocumentStatus;
 import org.raddatz.familienarchiv.person.Person;
 import org.raddatz.familienarchiv.person.PersonType;
@@ -16,10 +17,13 @@ import org.springframework.test.context.bean.override.mockito.MockitoBean;
 import org.springframework.transaction.annotation.Transactional;
 import software.amazon.awssdk.services.s3.S3Client;
 
+import org.springframework.mock.web.MockMultipartFile;
+
 import jakarta.persistence.EntityManager;
 import jakarta.persistence.PersistenceContext;
 
 import java.util.Set;
+import java.util.UUID;
 
 import static org.assertj.core.api.Assertions.assertThat;
 
@@ -33,6 +37,7 @@ class PersonServiceIntegrationTest {
     @Autowired PersonService personService;
     @Autowired PersonRepository personRepository;
     @Autowired DocumentRepository documentRepository;
+    @Autowired DocumentService documentService;
 
     @PersistenceContext EntityManager entityManager;
 
@@ -75,6 +80,93 @@ class PersonServiceIntegrationTest {
         assertThat(result.getLastName()).isEqualTo("Cram");
     }
 
+    // ─── #731: case-colliding alias resolution against real Postgres ───────────
+    // The umlaut pair is mandatory — only the real DB proves Postgres LOWER() folds ü; a
+    // plain-ASCII test would stay green while umlaut aliases regressed.
+
+    @Test
+    void findOrCreateByAlias_resolvesUmlautAliasCollision_toLowestId_withoutThrow() {
+        Person muller = personRepository.save(Person.builder().lastName("Müller").alias("Müller").build());
+        Person mullerLower = personRepository.save(Person.builder().lastName("müller").alias("müller").build());
+        UUID expected = muller.getId().compareTo(mullerLower.getId()) <= 0 ? muller.getId() : mullerLower.getId();
+
+        // No exact-case "MÜLLER" row → falls through to the case-insensitive branch with two
+        // candidates and must pick the lowest id, never throwing NonUniqueResultException.
+        Person resolved = personService.findOrCreateByAlias("MÜLLER");
+
+        assertThat(resolved.getId()).isEqualTo(expected);
+    }
+
+    @Test
+    void findOrCreateByAlias_umlautAliasCollision_isDeterministicAcrossCalls() {
+        personRepository.save(Person.builder().lastName("Müller").alias("Müller").build());
+        personRepository.save(Person.builder().lastName("müller").alias("müller").build());
+
+        Person first = personService.findOrCreateByAlias("MÜLLER");
+        Person second = personService.findOrCreateByAlias("MÜLLER");
+
+        assertThat(second.getId()).isEqualTo(first.getId());
+    }
+
+    // ─── #731: filename-based sender resolution against real Postgres ──────────
+
+    @Test
+    void storeDocument_resolvesSender_whenFilenameNameIsUnique() throws Exception {
+        Person hans = personRepository.save(Person.builder().firstName("Hans").lastName("Müller").build());
+
+        Document doc = uploadNamed("1965-03-12_Müller_Hans.pdf").document();
+
+        assertThat(doc.getSender()).isNotNull();
+        assertThat(doc.getSender().getId()).isEqualTo(hans.getId());
+    }
+
+    @Test
+    void storeDocument_resolvesSender_onSingleCaseInsensitiveMatch() throws Exception {
+        Person hans = personRepository.save(Person.builder().firstName("Hans").lastName("Müller").build());
+
+        // Filename folds to "hans müller"; the only stored person is "Hans Müller".
+        Document doc = uploadNamed("1965-03-12_müller_hans.pdf").document();
+
+        assertThat(doc.getSender()).isNotNull();
+        assertThat(doc.getSender().getId()).isEqualTo(hans.getId());
+    }
+
+    @Test
+    void storeDocument_leavesSenderUnset_whenFilenameNameIsAmbiguous() throws Exception {
+        // Two persons collide case-insensitively; the filename casing ("HANS"/"MÜLLER") matches
+        // neither exactly → no exact-case winner → bail to null (never an arbitrary guess), no 500.
+        personRepository.save(Person.builder().firstName("Hans").lastName("Müller").build());
+        personRepository.save(Person.builder().firstName("hans").lastName("müller").build());
+
+        Document doc = uploadNamed("1965-03-12_MÜLLER_HANS.pdf").document();
+
+        assertThat(doc.getSender()).isNull();
+    }
+
+    @Test
+    void storeDocument_leavesSenderUnset_whenFilenameHasNoFirstName() throws Exception {
+        // A last-name-only filename never resolves to a sender (the parser yields no parsed name).
+        personRepository.save(Person.builder().lastName("Müller").build());
+
+        Document doc = uploadNamed("1965-03-12_Müller.pdf").document();
+
+        assertThat(doc.getSender()).isNull();
+    }
+
+    @Test
+    void findByName_nullFirstName_resolvesToEmpty_inRealPostgres() {
+        // Fail-closed against the real DB: a null first name must NOT widen to first_name IS NULL
+        // and pick up the last-name-only row.
+        personRepository.save(Person.builder().lastName("Müller").build()); // first_name NULL
+
+        assertThat(personService.findByName(null, "Müller")).isEmpty();
+    }
+
+    private DocumentService.StoreResult uploadNamed(String filename) throws Exception {
+        MockMultipartFile file = new MockMultipartFile("file", filename, "application/pdf", new byte[]{1, 2, 3});
+        return documentService.storeDocument(file, null);
+    }
+
     // ─── #667: confirm round-trip + reader-default semantics ──────────────────
 
     @Test
@@ -124,12 +216,65 @@ class PersonServiceIntegrationTest {
         assertThat(personRepository.findById(target.getId())).isEmpty();
     }
 
+    // ─── generation full-stack round-trip (#689) ──────────────────────────────
+
+    @Test
+    void updatePerson_clearGenerationToNull_readsBackNullFromDb() {
+        // Sara's QA concern: pin the full PUT→DB→GET round-trip for the
+        // null-clear path. Without this we only have the WebMvcTest mocked
+        // boundary; nothing proved the JPA flush actually wrote SQL NULL.
+        Person seeded = personRepository.save(Person.builder()
+                .firstName("Hans").lastName("Raddatz")
+                .personType(PersonType.PERSON).generation(3).build());
+        entityManager.flush();
+        entityManager.clear();
+
+        PersonUpdateDTO dto = new PersonUpdateDTO();
+        dto.setPersonType(PersonType.PERSON);
+        dto.setFirstName("Hans");
+        dto.setLastName("Raddatz");
+        dto.setGeneration(null);
+
+        personService.updatePerson(seeded.getId(), dto);
+        entityManager.flush();
+        entityManager.clear();
+
+        Person reloaded = personRepository.findById(seeded.getId()).orElseThrow();
+        assertThat(reloaded.getGeneration()).isNull();
+    }
+
+    @Test
+    void updatePerson_setGenerationToZero_readsBackZeroFromDb() {
+        // Pin the G 0 case end-to-end. The form-action spec covers that 0
+        // doesn't get spread-dropped at the SvelteKit boundary; this test
+        // covers that the controller + service + JPA chain preserves the
+        // primitive zero (not coerced to null somewhere along the way).
+        Person seeded = personRepository.save(Person.builder()
+                .firstName("Walter").lastName("Raddatz")
+                .personType(PersonType.PERSON).build());
+        entityManager.flush();
+        entityManager.clear();
+
+        PersonUpdateDTO dto = new PersonUpdateDTO();
+        dto.setPersonType(PersonType.PERSON);
+        dto.setFirstName("Walter");
+        dto.setLastName("Raddatz");
+        dto.setGeneration(0);
+
+        personService.updatePerson(seeded.getId(), dto);
+        entityManager.flush();
+        entityManager.clear();
+
+        Person reloaded = personRepository.findById(seeded.getId()).orElseThrow();
+        assertThat(reloaded.getGeneration()).isEqualTo(0);
+    }
+
     @Test
     void deletePerson_detachesSentAndReceivedReferences_beforeDelete_noOrphan() {
         // A person referenced as BOTH a document sender and a document receiver must delete
-        // cleanly: deletePerson nulls the sender_id FK and removes the receiver join row first
-        // (reassignSenderToNull → deleteReceiverReferences → deleteById), so no FK orphan and
-        // the documents themselves survive.
+        // cleanly via the service path: deletePerson just calls deleteById, and V71's ON DELETE
+        // constraints null the sender_id FK and drop the receiver join row, so there is no FK
+        // orphan and the documents themselves survive.
         Person target = personRepository.save(Person.builder()
                 .firstName("Weg").lastName("Person").provisional(true).build());
         Person bystander = personRepository.save(Person.builder()
@@ -143,16 +288,16 @@ class PersonServiceIntegrationTest {
                 .status(DocumentStatus.UPLOADED).sender(bystander)
                 .receivers(new java.util.HashSet<>(Set.of(target))).build());
 
-        // Persist the fixture and detach everything so the native @Modifying deletes operate on
-        // the database directly without the persistence context holding stale references that
-        // would re-flush a now-deleted person as a transient association.
+        // Persist the fixture and detach everything so the delete operates on the database
+        // directly without the persistence context holding stale references.
         entityManager.flush();
         entityManager.clear();
 
         personService.deletePerson(target.getId());
 
-        // Native @Modifying queries bypass the persistence context — clear it so the asserting
-        // reads observe the post-delete database state, not stale managed entities.
+        // The ON DELETE cascade fires beneath Hibernate — flush the delete and clear the L1
+        // cache so the asserting reads observe the post-delete database state, not stale
+        // managed entities still holding the dropped sender/receiver associations.
         entityManager.flush();
         entityManager.clear();
 
@@ -167,4 +312,38 @@ class PersonServiceIntegrationTest {
         // The other person and the documents themselves survive the delete.
         assertThat(personRepository.findById(bystander.getId())).isPresent();
     }
+
+    @Test
+    void mergePersons_targetInheritsReferences_sourceJoinRowCascadeDrops_noFkError() {
+        // AC-7: merging a source who is sender of A and receiver of B into a target leaves the
+        // target as sender of A and receiver of B, drops the source's leftover receiver row via
+        // V71's ON DELETE CASCADE (no explicit delete, no FK error), and co-receivers are intact.
+        Person source = personRepository.save(Person.builder().firstName("Anna").lastName("Alt").build());
+        Person target = personRepository.save(Person.builder().firstName("Anna").lastName("Neu").build());
+        Person coReceiver = personRepository.save(Person.builder().firstName("Mit").lastName("Empfänger").build());
+        Person sender = personRepository.save(Person.builder().firstName("Send").lastName("Er").build());
+
+        Document docA = documentRepository.save(Document.builder()
+                .title("Von Anna").originalFilename("a.pdf")
+                .status(DocumentStatus.UPLOADED).sender(source).build());
+        Document docB = documentRepository.save(Document.builder()
+                .title("An Anna").originalFilename("b.pdf")
+                .status(DocumentStatus.UPLOADED).sender(sender)
+                .receivers(new java.util.HashSet<>(Set.of(source, coReceiver))).build());
+        entityManager.flush();
+        entityManager.clear();
+
+        personService.mergePersons(source.getId(), target.getId());
+        entityManager.flush();
+        entityManager.clear();
+
+        assertThat(personRepository.findById(source.getId())).isEmpty();
+
+        Document reloadedA = documentRepository.findById(docA.getId()).orElseThrow();
+        assertThat(reloadedA.getSender().getId()).isEqualTo(target.getId());
+
+        Document reloadedB = documentRepository.findById(docB.getId()).orElseThrow();
+        assertThat(reloadedB.getReceivers()).extracting(Person::getId)
+                .containsExactlyInAnyOrder(target.getId(), coReceiver.getId());
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/person/PersonServiceTest.java

@@ -27,6 +27,7 @@ import static org.mockito.ArgumentMatchers.any;
 import static org.mockito.ArgumentMatchers.argThat;
 import static org.mockito.Mockito.never;
 import static org.mockito.Mockito.verify;
+import static org.mockito.Mockito.verifyNoMoreInteractions;
 import static org.mockito.Mockito.when;
 
 @ExtendWith(MockitoExtension.class)
@@ -147,9 +148,11 @@ class PersonServiceTest {
 
         personService.deletePerson(id);
 
-        verify(personRepository).reassignSenderToNull(id);
-        verify(personRepository).deleteReceiverReferences(id);
+        // Integrity is enforced by V71's ON DELETE constraints — the service only checks
+        // existence then deletes; it no longer detaches sender/receiver references itself.
+        verify(personRepository).findById(id);
         verify(personRepository).deleteById(id);
+        verifyNoMoreInteractions(personRepository);
     }
 
     @Test
@@ -261,6 +264,54 @@ class PersonServiceTest {
                 .isEqualTo(400);
     }
 
+    @Test
+    void createPerson_dto_persistsGeneration() {
+        when(personRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        PersonUpdateDTO dto = new PersonUpdateDTO();
+        dto.setFirstName("Hans"); dto.setLastName("Raddatz");
+        dto.setPersonType(PersonType.PERSON); dto.setGeneration(3);
+
+        Person result = personService.createPerson(dto);
+
+        assertThat(result.getGeneration()).isEqualTo(3);
+    }
+
+    @Test
+    void updatePerson_writesGeneration_includingExplicitNullClear() {
+        // The form path is the only place a human can clear generation back to null.
+        UUID id = UUID.randomUUID();
+        Person existing = Person.builder().id(id).firstName("Hans").lastName("Raddatz")
+                .personType(PersonType.PERSON).generation(3).build();
+        when(personRepository.findById(id)).thenReturn(Optional.of(existing));
+        when(personRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        PersonUpdateDTO dto = new PersonUpdateDTO();
+        dto.setFirstName("Hans"); dto.setLastName("Raddatz");
+        dto.setPersonType(PersonType.PERSON); dto.setGeneration(null);
+
+        Person result = personService.updatePerson(id, dto);
+
+        assertThat(result.getGeneration()).isNull();
+    }
+
+    @Test
+    void updatePerson_writesGeneration_whenSet() {
+        UUID id = UUID.randomUUID();
+        Person existing = Person.builder().id(id).firstName("Hans").lastName("Raddatz")
+                .personType(PersonType.PERSON).build();
+        when(personRepository.findById(id)).thenReturn(Optional.of(existing));
+        when(personRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        PersonUpdateDTO dto = new PersonUpdateDTO();
+        dto.setFirstName("Hans"); dto.setLastName("Raddatz");
+        dto.setPersonType(PersonType.PERSON); dto.setGeneration(2);
+
+        Person result = personService.updatePerson(id, dto);
+
+        assertThat(result.getGeneration()).isEqualTo(2);
+    }
+
     // ─── updatePerson (personType) ───────────────────────────────────────────
 
     @Test
@@ -324,14 +375,57 @@ class PersonServiceTest {
     // ─── findOrCreateByAlias ─────────────────────────────────────────────────
 
     @Test
-    void findOrCreateByAlias_returnsExisting_whenAliasFound() {
-        String alias = "Walter de Gruyter";
-        Person existing = Person.builder().id(UUID.randomUUID()).alias(alias).build();
-        when(personRepository.findByAliasIgnoreCase(alias)).thenReturn(Optional.of(existing));
+    void findOrCreateByAlias_returnsExactCaseMatch_overCaseInsensitiveSibling() {
+        String alias = "müller";
+        Person exact = Person.builder().id(UUID.randomUUID()).alias("müller").build();
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.of(exact));
 
         Person result = personService.findOrCreateByAlias(alias);
 
-        assertThat(result).isEqualTo(existing);
+        assertThat(result).isEqualTo(exact);
+        verify(personRepository, never()).findAllByAliasIgnoreCase(any());
+        verify(personRepository, never()).save(any());
+    }
+
+    @Test
+    void findOrCreateByAlias_returnsExactCaseMatch_evenWhenMultipleSiblingsCollide() {
+        String alias = "Müller";
+        Person exact = Person.builder().id(UUID.randomUUID()).alias("Müller").build();
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.of(exact));
+
+        Person result = personService.findOrCreateByAlias(alias);
+
+        assertThat(result).isEqualTo(exact);
+        // exact-case short-circuits — the case-insensitive siblings are never consulted.
+        verify(personRepository, never()).findAllByAliasIgnoreCase(any());
+    }
+
+    @Test
+    void findOrCreateByAlias_usesSingleCaseInsensitiveMatch_whenNoExactCase() {
+        String alias = "müller";
+        Person only = Person.builder().id(UUID.randomUUID()).alias("Müller").build();
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.empty());
+        when(personRepository.findAllByAliasIgnoreCase(alias)).thenReturn(List.of(only));
+
+        Person result = personService.findOrCreateByAlias(alias);
+
+        assertThat(result).isEqualTo(only);
+        verify(personRepository, never()).save(any());
+    }
+
+    @Test
+    void findOrCreateByAlias_returnsLowestIdDeterministically_whenMultipleCaseInsensitiveMatches() {
+        String alias = "müller";
+        Person lower = Person.builder().id(UUID.fromString("00000000-0000-0000-0000-000000000001")).alias("Müller").build();
+        Person higher = Person.builder().id(UUID.fromString("00000000-0000-0000-0000-000000000002")).alias("müller").build();
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.empty());
+        when(personRepository.findAllByAliasIgnoreCase(alias)).thenReturn(List.of(higher, lower)); // unordered
+
+        Person first = personService.findOrCreateByAlias(alias);
+        Person second = personService.findOrCreateByAlias(alias);
+
+        assertThat(first.getId()).isEqualTo(lower.getId());      // lowest id wins
+        assertThat(second.getId()).isEqualTo(first.getId());     // same result every call — never throws
         verify(personRepository, never()).save(any());
     }
 
@@ -339,7 +433,8 @@ class PersonServiceTest {
     void findOrCreateByAlias_createsNew_whenAliasNotFound() {
         String alias = "Clara Cram";
         Person saved = Person.builder().id(UUID.randomUUID()).alias(alias).firstName("Clara").lastName("Cram").build();
-        when(personRepository.findByAliasIgnoreCase(alias)).thenReturn(Optional.empty());
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.empty());
+        when(personRepository.findAllByAliasIgnoreCase(alias)).thenReturn(List.of());
         when(personRepository.save(any())).thenReturn(saved);
 
         Person result = personService.findOrCreateByAlias(alias);
@@ -352,7 +447,8 @@ class PersonServiceTest {
     void findOrCreateByAlias_createsMaidenNameAlias_whenGebPresent() {
         String alias = "Clara Cram geb. de Gruyter";
         Person saved = Person.builder().id(UUID.randomUUID()).alias(alias).firstName("Clara").lastName("Cram").build();
-        when(personRepository.findByAliasIgnoreCase(alias)).thenReturn(Optional.empty());
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.empty());
+        when(personRepository.findAllByAliasIgnoreCase(alias)).thenReturn(List.of());
         when(personRepository.save(any())).thenReturn(saved);
         when(aliasRepository.findMaxSortOrder(saved.getId())).thenReturn(0);
         when(aliasRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
@@ -374,7 +470,8 @@ class PersonServiceTest {
     @Test
     void findOrCreateByAlias_setsInstitutionType_withFullNameInLastName() {
         String alias = "Arthur Collignon GmbH";
-        when(personRepository.findByAliasIgnoreCase(alias)).thenReturn(Optional.empty());
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.empty());
+        when(personRepository.findAllByAliasIgnoreCase(alias)).thenReturn(List.of());
         when(personRepository.save(any())).thenAnswer(inv -> {
             Person p = inv.getArgument(0);
             p.setId(UUID.randomUUID());
@@ -391,7 +488,8 @@ class PersonServiceTest {
     @Test
     void findOrCreateByAlias_setsGroupType_withFullNameInLastName() {
         String alias = "Geschwister de Gruyter";
-        when(personRepository.findByAliasIgnoreCase(alias)).thenReturn(Optional.empty());
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.empty());
+        when(personRepository.findAllByAliasIgnoreCase(alias)).thenReturn(List.of());
         when(personRepository.save(any())).thenAnswer(inv -> {
             Person p = inv.getArgument(0);
             p.setId(UUID.randomUUID());
@@ -409,7 +507,8 @@ class PersonServiceTest {
     void findOrCreateByAlias_noAlias_whenNoGeb() {
         String alias = "Clara Cram";
         Person saved = Person.builder().id(UUID.randomUUID()).alias(alias).firstName("Clara").lastName("Cram").build();
-        when(personRepository.findByAliasIgnoreCase(alias)).thenReturn(Optional.empty());
+        when(personRepository.findByAlias(alias)).thenReturn(Optional.empty());
+        when(personRepository.findAllByAliasIgnoreCase(alias)).thenReturn(List.of());
         when(personRepository.save(any())).thenReturn(saved);
 
         personService.findOrCreateByAlias(alias);
@@ -421,11 +520,54 @@ class PersonServiceTest {
     void findOrCreateByAlias_trimsInput() {
         String alias = "  Clara Cram  ";
         Person saved = Person.builder().id(UUID.randomUUID()).alias("Clara Cram").build();
-        when(personRepository.findByAliasIgnoreCase("Clara Cram")).thenReturn(Optional.of(saved));
+        when(personRepository.findByAlias("Clara Cram")).thenReturn(Optional.of(saved));
 
         personService.findOrCreateByAlias(alias);
 
-        verify(personRepository).findByAliasIgnoreCase("Clara Cram");
+        verify(personRepository).findByAlias("Clara Cram");
+    }
+
+    // ─── findByName (filename-based sender resolution) ────────────────────────
+
+    @Test
+    void findByName_returnsExactCaseMatch_overCaseInsensitiveSibling() {
+        Person exact = Person.builder().id(UUID.randomUUID()).firstName("Hans").lastName("Müller").build();
+        when(personRepository.findByFirstNameAndLastName("Hans", "Müller")).thenReturn(Optional.of(exact));
+
+        assertThat(personService.findByName("Hans", "Müller")).contains(exact);
+        verify(personRepository, never()).findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase(any(), any());
+    }
+
+    @Test
+    void findByName_usesSingleCaseInsensitiveMatch_whenNoExactCase() {
+        Person only = Person.builder().id(UUID.randomUUID()).firstName("Hans").lastName("Müller").build();
+        when(personRepository.findByFirstNameAndLastName("hans", "müller")).thenReturn(Optional.empty());
+        when(personRepository.findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase("hans", "müller"))
+                .thenReturn(List.of(only));
+
+        assertThat(personService.findByName("hans", "müller")).contains(only);
+    }
+
+    @Test
+    void findByName_bailsToEmpty_whenTwoOrMoreCaseInsensitiveMatches() {
+        Person a = Person.builder().id(UUID.randomUUID()).firstName("Hans").lastName("Müller").build();
+        Person b = Person.builder().id(UUID.randomUUID()).firstName("hans").lastName("müller").build();
+        when(personRepository.findByFirstNameAndLastName("hans", "müller")).thenReturn(Optional.empty());
+        when(personRepository.findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase("hans", "müller"))
+                .thenReturn(List.of(a, b));
+
+        // Ambiguous sender → unset, never an arbitrary guess (provenance correctness over a
+        // confidently-wrong pre-fill). This is the deliberate divergence from the alias path.
+        assertThat(personService.findByName("hans", "müller")).isEmpty();
+    }
+
+    @Test
+    void findByName_returnsEmpty_whenFirstNameNullFoldsToNoMatch() {
+        when(personRepository.findByFirstNameAndLastName(null, "Müller")).thenReturn(Optional.empty());
+        when(personRepository.findAllByFirstNameIgnoreCaseAndLastNameIgnoreCase(null, "Müller"))
+                .thenReturn(List.of());
+
+        assertThat(personService.findByName(null, "Müller")).isEmpty();
     }
 
     // ─── updatePerson (notes) ────────────────────────────────────────────────
@@ -652,10 +794,14 @@ class PersonServiceTest {
 
         personService.mergePersons(sourceId, targetId);
 
+        verify(personRepository).findById(sourceId);
+        verify(personRepository).findById(targetId);
         verify(personRepository).reassignSender(sourceId, targetId);
         verify(personRepository).insertMissingReceiverReference(sourceId, targetId);
-        verify(personRepository).deleteReceiverReferences(sourceId);
         verify(personRepository).deleteById(sourceId);
+        // The source's leftover receiver rows cascade-drop via V71's ON DELETE CASCADE on
+        // deleteById — merge no longer deletes them explicitly.
+        verifyNoMoreInteractions(personRepository);
     }
 
     // ─── getAliases ─────────────────────────────────────────────────────────
@@ -752,4 +898,15 @@ class PersonServiceTest {
                 .extracting(e -> ((DomainException) e).getStatus().value())
                 .isEqualTo(403);
     }
+
+    @Test
+    void findByDisplayNameContaining_delegatesToSearchByName() {
+        Person walter = Person.builder().id(UUID.randomUUID()).firstName("Walter").lastName("Müller").build();
+        when(personRepository.searchByName("Walter")).thenReturn(List.of(walter));
+
+        List<Person> result = personService.findByDisplayNameContaining("Walter");
+
+        assertThat(result).containsExactly(walter);
+        verify(personRepository).searchByName("Walter");
+    }
 }

backend/src/test/java/org/raddatz/familienarchiv/person/relationship/RelationshipControllerTest.java

@@ -93,7 +93,7 @@ class RelationshipControllerTest {
     @Test
     @WithMockUser(username = "testuser", authorities = {"READ_ALL"})
     void getNetwork_returns200_with_NetworkDTO_for_authenticated_user() throws Exception {
-        PersonNodeDTO node = new PersonNodeDTO(PERSON_ID, "Alice Müller", 1900, 1980, true);
+        PersonNodeDTO node = new PersonNodeDTO(PERSON_ID, "Alice Müller", 1900, 1980, null, true);
         RelationshipDTO edge = new RelationshipDTO(
                 UUID.randomUUID(), PERSON_ID, OTHER_ID,
                 "Alice Müller", 1900, 1980,
@@ -111,7 +111,7 @@ class RelationshipControllerTest {
     @Test
     @WithMockUser(username = "testuser", authorities = {"READ_ALL"})
     void getInferredRelationships_returns200_with_list_for_authenticated_user() throws Exception {
-        PersonNodeDTO relative = new PersonNodeDTO(OTHER_ID, "Bob Müller", 1930, null, true);
+        PersonNodeDTO relative = new PersonNodeDTO(OTHER_ID, "Bob Müller", 1930, null, null, true);
         InferredRelationshipWithPersonDTO inferred =
                 new InferredRelationshipWithPersonDTO(relative, "Großvater", 2);
         when(relationshipService.getInferredRelationships(PERSON_ID))

backend/src/test/java/org/raddatz/familienarchiv/person/relationship/RelationshipServiceIntegrationTest.java

@@ -144,10 +144,12 @@ class RelationshipServiceIntegrationTest {
 
     @Test
     void setFamilyMember_true_makes_person_appear_in_network() {
-        // charlie starts with familyMember = false. Add a PARENT_OF edge alice→charlie
-        // so the edge exists, then flip charlie's flag and verify he appears in nodes.
+        // addRelationship side-effects family_member=true on both endpoints for family-graph
+        // edges (PARENT_OF/SPOUSE_OF/SIBLING_OF). Reset charlie so the explicit
+        // setFamilyMember(true) call below is the thing under test, not the auto-flip.
         relationshipService.addRelationship(alice.getId(),
                 new CreateRelationshipRequest(charlie.getId(), RelationType.PARENT_OF, null, null, null));
+        relationshipService.setFamilyMember(charlie.getId(), false);
 
         NetworkDTO before = relationshipService.getFamilyNetwork();
         assertThat(before.nodes()).extracting("id").doesNotContain(charlie.getId());

backend/src/test/java/org/raddatz/familienarchiv/person/relationship/RelationshipServiceTest.java

@@ -23,6 +23,8 @@ import java.util.UUID;
 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.anyBoolean;
+import static org.mockito.ArgumentMatchers.eq;
 import static org.mockito.Mockito.never;
 import static org.mockito.Mockito.verify;
 import static org.mockito.Mockito.when;
@@ -148,6 +150,50 @@ class RelationshipServiceTest {
         assertThat(result.notes()).isEqualTo("first born");
     }
 
+    @Test
+    void addRelationship_marks_both_endpoints_as_family_member_when_type_is_family() {
+        // Creating a family-graph edge (PARENT_OF / SPOUSE_OF / SIBLING_OF) must mark both
+        // endpoints as family members so they appear in findAllFamilyMembers and the network.
+        // This is what makes the canonical importer's relationships actually show up in the UI.
+        when(personService.getById(alice.getId())).thenReturn(alice);
+        when(personService.getById(bob.getId())).thenReturn(bob);
+        when(relationshipRepository.existsByPersonIdAndRelatedPersonIdAndRelationType(
+                bob.getId(), alice.getId(), RelationType.PARENT_OF)).thenReturn(false);
+        when(relationshipRepository.saveAndFlush(any())).thenAnswer(inv -> {
+            PersonRelationship r = inv.getArgument(0);
+            r.setId(UUID.randomUUID());
+            r.setCreatedAt(Instant.now());
+            return r;
+        });
+
+        var dto = new CreateRelationshipRequest(bob.getId(), RelationType.PARENT_OF, null, null, null);
+        service.addRelationship(alice.getId(), dto);
+
+        verify(personService).setFamilyMember(alice.getId(), true);
+        verify(personService).setFamilyMember(bob.getId(), true);
+    }
+
+    @Test
+    void addRelationship_does_not_flip_family_member_for_non_family_type() {
+        // FRIEND / COLLEAGUE / EMPLOYER / DOCTOR / NEIGHBOR / OTHER are NOT family-graph
+        // edges (see getFamilyNetwork's filter), so addRelationship must leave family_member
+        // alone — a doctor of the family is not a family member.
+        when(personService.getById(alice.getId())).thenReturn(alice);
+        when(personService.getById(bob.getId())).thenReturn(bob);
+        when(relationshipRepository.saveAndFlush(any())).thenAnswer(inv -> {
+            PersonRelationship r = inv.getArgument(0);
+            r.setId(UUID.randomUUID());
+            r.setCreatedAt(Instant.now());
+            return r;
+        });
+
+        var dto = new CreateRelationshipRequest(bob.getId(), RelationType.FRIEND, null, null, null);
+        service.addRelationship(alice.getId(), dto);
+
+        verify(personService, never()).setFamilyMember(eq(alice.getId()), anyBoolean());
+        verify(personService, never()).setFamilyMember(eq(bob.getId()), anyBoolean());
+    }
+
     @Test
     void deleteRelationship_succeeds_when_viewpoint_is_object() {
         UUID relId = UUID.randomUUID();
@@ -191,6 +237,22 @@ class RelationshipServiceTest {
         assertThat(result.edges().get(0).relatedPersonId()).isEqualTo(bob.getId());
     }
 
+    @Test
+    void getFamilyNetwork_populates_generation_on_PersonNodeDTO() {
+        Person walter = Person.builder().id(UUID.randomUUID()).lastName("Raddatz")
+                .familyMember(true).generation(2).build();
+        Person clara = Person.builder().id(UUID.randomUUID()).lastName("Raddatz")
+                .familyMember(true).generation(3).build();
+        when(personService.findAllFamilyMembers()).thenReturn(List.of(walter, clara));
+        when(relationshipRepository.findAllByRelationTypeIn(any())).thenReturn(List.of());
+
+        NetworkDTO result = service.getFamilyNetwork();
+
+        assertThat(result.nodes()).hasSize(2);
+        assertThat(result.nodes().stream().map(n -> n.generation()).toList())
+                .containsExactlyInAnyOrder(2, 3);
+    }
+
     // --- helpers ---
 
     private static Person person(String name) {

backend/src/test/java/org/raddatz/familienarchiv/search/NlQueryParserServiceTest.java

@@ -0,0 +1,440 @@
+package org.raddatz.familienarchiv.search;
+
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.mockito.ArgumentCaptor;
+import org.mockito.Mock;
+import org.mockito.MockitoAnnotations;
+import org.raddatz.familienarchiv.document.DocumentSearchResult;
+import org.raddatz.familienarchiv.document.DocumentService;
+import org.raddatz.familienarchiv.document.DocumentSort;
+import org.raddatz.familienarchiv.document.SearchFilters;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.raddatz.familienarchiv.person.Person;
+import org.raddatz.familienarchiv.person.PersonService;
+import org.raddatz.familienarchiv.tag.TagOperator;
+import org.springframework.data.domain.PageRequest;
+import org.springframework.data.domain.Pageable;
+
+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.assertj.core.api.Assertions.assertThatThrownBy;
+import static org.mockito.ArgumentMatchers.*;
+import static org.mockito.Mockito.*;
+
+class NlQueryParserServiceTest {
+
+    @Mock OllamaClient ollamaClient;
+    @Mock PersonService personService;
+    @Mock DocumentService documentService;
+
+    NlQueryParserService service;
+
+    static final Pageable PAGE = PageRequest.of(0, 20);
+
+    @BeforeEach
+    void setUp() {
+        MockitoAnnotations.openMocks(this);
+        service = new NlQueryParserService(ollamaClient, personService, documentService);
+        when(documentService.searchDocuments(any(), any(), any(), any()))
+                .thenReturn(DocumentSearchResult.of(List.of()));
+        when(documentService.searchDocumentsByPersonId(any(), any(), any(), any()))
+                .thenReturn(DocumentSearchResult.of(List.of()));
+    }
+
+    // --- Factory helpers ---
+
+    private OllamaExtraction extraction(List<String> names, String role, LocalDate from, LocalDate to,
+                                         List<String> keywords) {
+        String raw = names.isEmpty() ? "test query" : String.join(" ", names);
+        return new OllamaExtraction(names, role, from, to, keywords, raw);
+    }
+
+    private Person person(UUID id, String firstName, String lastName) {
+        return Person.builder().id(id).firstName(firstName).lastName(lastName).build();
+    }
+
+    private static final UUID P1 = UUID.fromString("00000000-0000-0000-0000-000000000001");
+    private static final UUID P2 = UUID.fromString("00000000-0000-0000-0000-000000000002");
+    private static final UUID P3 = UUID.fromString("00000000-0000-0000-0000-000000000003");
+
+    // --- 1. Single resolved name + personRole=sender ---
+
+    @Test
+    void search_resolvesSingleName_asSender() {
+        Person walter = person(P1, "Walter", "Raddatz");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Walter"), "sender", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(List.of(walter));
+
+        NlSearchResponse resp = service.search("Was hat Walter geschrieben?", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), eq(DocumentSort.DATE), eq("desc"), eq(PAGE));
+        assertThat(cap.getValue().sender()).isEqualTo(P1);
+        assertThat(cap.getValue().receiver()).isNull();
+        assertThat(resp.interpretation().resolvedPersons()).hasSize(1);
+        assertThat(resp.interpretation().resolvedPersons().get(0).id()).isEqualTo(P1);
+        assertThat(resp.interpretation().ambiguousPersons()).isEmpty();
+    }
+
+    // --- 2. Multi-match name → ambiguous, search NOT executed ---
+
+    @Test
+    void search_multiMatchName_populatesAmbiguous_andSkipsSearch() {
+        Person a = person(UUID.randomUUID(), "Walter", "Braun");
+        Person b = person(UUID.randomUUID(), "Walter", "Schmidt");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Walter"), "sender", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(List.of(a, b));
+
+        NlSearchResponse resp = service.search("Briefe von Walter", PAGE);
+
+        verify(documentService, never()).searchDocuments(any(), any(), any(), any());
+        verify(documentService, never()).searchDocumentsByPersonId(any(), any(), any(), any());
+        assertThat(resp.interpretation().ambiguousPersons()).hasSize(2);
+        assertThat(resp.interpretation().resolvedPersons()).isEmpty();
+    }
+
+    // --- 3. Multi-match + personRole=any → still ambiguous, search NOT executed ---
+
+    @Test
+    void search_multiMatchName_withPersonRoleAny_stillSkipsSearch() {
+        Person a = person(UUID.randomUUID(), "Emma", "Braun");
+        Person b = person(UUID.randomUUID(), "Emma", "Raddatz");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Emma"), "any", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Emma")).thenReturn(List.of(a, b));
+
+        NlSearchResponse resp = service.search("Briefe an Emma", PAGE);
+
+        verify(documentService, never()).searchDocuments(any(), any(), any(), any());
+        verify(documentService, never()).searchDocumentsByPersonId(any(), any(), any(), any());
+        assertThat(resp.interpretation().ambiguousPersons()).hasSize(2);
+    }
+
+    // --- 4. No-match name → folded into text ---
+
+    @Test
+    void search_noMatchName_isFoldedIntoText() {
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Karl"), "any", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Karl")).thenReturn(List.of());
+
+        service.search("Briefe von Karl", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().text()).contains("Karl");
+        assertThat(cap.getValue().sender()).isNull();
+        assertThat(cap.getValue().receiver()).isNull();
+    }
+
+    // --- 5. personRole=any + 1 resolved → searchDocumentsByPersonId called ---
+
+    @Test
+    void search_personRoleAny_singleMatch_callsSearchDocumentsByPersonId() {
+        Person walter = person(P1, "Walter", "Raddatz");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Walter"), "any", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(List.of(walter));
+
+        NlSearchResponse resp = service.search("Briefe von Walter", PAGE);
+
+        verify(documentService).searchDocumentsByPersonId(eq(P1), isNull(), isNull(), eq(PAGE));
+        verify(documentService, never()).searchDocuments(any(), any(), any(), any());
+        assertThat(resp.interpretation().keywordsApplied()).isFalse();
+    }
+
+    // --- 6. 2 names both resolve → sender=person1, receiver=person2 ---
+
+    @Test
+    void search_twoNamesResolve_assignsSenderAndReceiver() {
+        Person walter = person(P1, "Walter", "Raddatz");
+        Person emma = person(P2, "Emma", "Raddatz");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Walter", "Emma"), "any", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(List.of(walter));
+        when(personService.findByDisplayNameContaining("Emma")).thenReturn(List.of(emma));
+
+        NlSearchResponse resp = service.search("Briefe von Walter an Emma", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), eq(DocumentSort.DATE), eq("desc"), eq(PAGE));
+        assertThat(cap.getValue().sender()).isEqualTo(P1);
+        assertThat(cap.getValue().receiver()).isEqualTo(P2);
+        assertThat(resp.interpretation().resolvedPersons().get(0).id()).isEqualTo(P1);
+        assertThat(resp.interpretation().resolvedPersons().get(1).id()).isEqualTo(P2);
+    }
+
+    // --- 7. 2 names, first resolves, second ambiguous → search NOT executed ---
+
+    @Test
+    void search_twoNames_secondAmbiguous_skipsSearch() {
+        Person walter = person(P1, "Walter", "Raddatz");
+        Person emma1 = person(P2, "Emma", "Braun");
+        Person emma2 = person(P3, "Emma", "Schmidt");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Walter", "Emma"), "sender", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(List.of(walter));
+        when(personService.findByDisplayNameContaining("Emma")).thenReturn(List.of(emma1, emma2));
+
+        NlSearchResponse resp = service.search("Briefe von Walter an Emma", PAGE);
+
+        verify(documentService, never()).searchDocuments(any(), any(), any(), any());
+        assertThat(resp.interpretation().ambiguousPersons()).hasSize(2);
+    }
+
+    // --- 8. 2 names, first no match → folded into text, second used as single person ---
+
+    @Test
+    void search_twoNames_firstNoMatch_secondResolved_foldFirstIntoText() {
+        Person emma = person(P2, "Emma", "Raddatz");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Karl", "Emma"), "sender", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Karl")).thenReturn(List.of());
+        when(personService.findByDisplayNameContaining("Emma")).thenReturn(List.of(emma));
+
+        service.search("Briefe von Karl an Emma", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().text()).contains("Karl");
+        assertThat(cap.getValue().sender()).isEqualTo(P2);
+    }
+
+    // --- 9. 3+ names all resolve → first two as sender/receiver, third folded into text ---
+
+    @Test
+    void search_threeNamesResolve_extraFoldedIntoText() {
+        Person walter = person(P1, "Walter", "Raddatz");
+        Person emma = person(P2, "Emma", "Raddatz");
+        Person heinrich = person(P3, "Heinrich", "Braun");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Walter", "Emma", "Heinrich"), "any", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(List.of(walter));
+        when(personService.findByDisplayNameContaining("Emma")).thenReturn(List.of(emma));
+        when(personService.findByDisplayNameContaining("Heinrich")).thenReturn(List.of(heinrich));
+
+        service.search("Briefe von Walter an Emma über Heinrich", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().sender()).isEqualTo(P1);
+        assertThat(cap.getValue().receiver()).isEqualTo(P2);
+        assertThat(cap.getValue().text()).contains("Heinrich");
+    }
+
+    // --- 10. Keywords space-joined into text ---
+
+    @Test
+    void search_keywords_areJoinedIntoText() {
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of(), "any", null, null, List.of("Krieg", "Walter")));
+
+        service.search("Dokumente über den Krieg Walter", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().text()).isEqualTo("Krieg Walter");
+    }
+
+    // --- 11. Date range passed through ---
+
+    @Test
+    void search_dateRange_passedIntoSearchFilters() {
+        LocalDate from = LocalDate.of(1914, 1, 1);
+        LocalDate to = LocalDate.of(1914, 12, 31);
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of(), "any", from, to, List.of()));
+
+        service.search("Briefe aus dem Jahr 1914", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().from()).isEqualTo(from);
+        assertThat(cap.getValue().to()).isEqualTo(to);
+    }
+
+    // --- 12. Null dates → null in SearchFilters (not an error) ---
+
+    @Test
+    void search_nullDates_passedAsNullIntoFilters() {
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of(), "any", null, null, List.of("Hochzeit")));
+
+        service.search("Hochzeitsbriefe", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().from()).isNull();
+        assertThat(cap.getValue().to()).isNull();
+    }
+
+    // --- 13. Query under 3 chars → VALIDATION_ERROR before Ollama call ---
+
+    @Test
+    void search_queryTooShort_throwsValidationError() {
+        assertThatThrownBy(() -> service.search("ab", PAGE))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.VALIDATION_ERROR);
+
+        verify(ollamaClient, never()).parse(anyString());
+    }
+
+    // --- 14. Query over 500 chars → VALIDATION_ERROR ---
+
+    @Test
+    void search_queryTooLong_throwsValidationError() {
+        String longQuery = "a".repeat(501);
+        assertThatThrownBy(() -> service.search(longQuery, PAGE))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.VALIDATION_ERROR);
+
+        verify(ollamaClient, never()).parse(anyString());
+    }
+
+    // --- 15. Ollama returns empty names/keywords → raw query used as keyword fallback ---
+
+    @Test
+    void search_ollamaReturnsEmpty_usesRawQueryAsTextFallback() {
+        String raw = "Briefe aus dem Krieg";
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(new OllamaExtraction(List.of(), "any", null, null, List.of(), raw));
+
+        service.search(raw, PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().text()).isEqualTo(raw);
+    }
+
+    // --- 16. Null personNames/keywords from Ollama → no NPE ---
+
+    @Test
+    void search_nullPersonNamesAndKeywords_handledWithoutNpe() {
+        OllamaExtraction ext = new OllamaExtraction(null, "any", null, null, null, "test query");
+        when(ollamaClient.parse(anyString())).thenReturn(ext);
+
+        NlSearchResponse resp = service.search("test query", PAGE);
+
+        assertThat(resp).isNotNull();
+        verify(documentService).searchDocuments(any(), any(), any(), any());
+    }
+
+    // --- 17. Unrecognized personRole → defaults to any-like behavior (no crash) ---
+
+    @Test
+    void search_unrecognizedPersonRole_treatedLikeAny_withSingleResolvedPerson() {
+        Person walter = person(P1, "Walter", "Raddatz");
+        // OllamaClient defensive parsing returns "any" for unknown roles,
+        // but NlQueryParserService must also be safe if something unexpected arrives.
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(new OllamaExtraction(List.of("Walter"), "unknown_role", null, null, List.of(), "query"));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(List.of(walter));
+
+        NlSearchResponse resp = service.search("Briefe von Walter", PAGE);
+
+        // Should not crash; "unknown_role" treated as fallback (neither sender nor receiver → any)
+        assertThat(resp).isNotNull();
+    }
+
+    // --- 18. Ollama throws SMART_SEARCH_UNAVAILABLE → propagates to caller ---
+
+    @Test
+    void search_ollamaThrowsUnavailable_propagates() {
+        when(ollamaClient.parse(anyString()))
+                .thenThrow(DomainException.tooManyRequests(ErrorCode.SMART_SEARCH_UNAVAILABLE, "offline"));
+
+        assertThatThrownBy(() -> service.search("Was hat Walter geschrieben?", PAGE))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.SMART_SEARCH_UNAVAILABLE);
+    }
+
+    // --- 19. LLM-extracted name > 200 chars → skipped, PersonService never called ---
+
+    @Test
+    void search_nameLongerThan200Chars_isSkippedBeforePersonServiceCall() {
+        String longName = "A".repeat(201);
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of(longName), "sender", null, null, List.of()));
+
+        service.search("Briefe von sehr langem Namen", PAGE);
+
+        verify(personService, never()).findByDisplayNameContaining(anyString());
+    }
+
+    // --- 20. Max 10 candidates cap: 11 persons returned → only first 10 in ambiguousPersons ---
+
+    @Test
+    void search_elevenCandidates_capsAtTen() {
+        List<Person> eleven = new ArrayList<>();
+        for (int i = 0; i < 11; i++) {
+            eleven.add(person(UUID.randomUUID(), "Walter", "Person" + i));
+        }
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Walter"), "sender", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Walter")).thenReturn(eleven);
+
+        NlSearchResponse resp = service.search("Briefe von Walter", PAGE);
+
+        assertThat(resp.interpretation().ambiguousPersons()).hasSize(10);
+        verify(documentService, never()).searchDocuments(any(), any(), any(), any());
+    }
+
+    // --- 21. SearchFilters defaults: tagOperator=AND, status=null, undated=false, tags=empty ---
+
+    @Test
+    void search_searchFiltersDefaults_areCorrect() {
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of(), "any", null, null, List.of("Krieg")));
+
+        service.search("Dokumente über den Krieg", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), eq(DocumentSort.DATE), eq("desc"), eq(PAGE));
+        SearchFilters f = cap.getValue();
+        assertThat(f.tagOperator()).isEqualTo(TagOperator.AND);
+        assertThat(f.status()).isNull();
+        assertThat(f.undated()).isFalse();
+        assertThat(f.tags()).isEmpty();
+        assertThat(f.tagQ()).isNull();
+    }
+
+    // --- 22. personRole=receiver + 1 resolved → receiver UUID set ---
+
+    @Test
+    void search_personRoleReceiver_singleMatch_setsReceiver() {
+        Person emma = person(P2, "Emma", "Raddatz");
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of("Emma"), "receiver", null, null, List.of()));
+        when(personService.findByDisplayNameContaining("Emma")).thenReturn(List.of(emma));
+
+        service.search("Briefe an Emma", PAGE);
+
+        ArgumentCaptor<SearchFilters> cap = ArgumentCaptor.forClass(SearchFilters.class);
+        verify(documentService).searchDocuments(cap.capture(), any(), any(), any());
+        assertThat(cap.getValue().receiver()).isEqualTo(P2);
+        assertThat(cap.getValue().sender()).isNull();
+    }
+
+    // --- 23. keywordsApplied=true when text is non-blank ---
+
+    @Test
+    void search_keywordsApplied_trueWhenTextNonBlank() {
+        when(ollamaClient.parse(anyString()))
+                .thenReturn(extraction(List.of(), "any", null, null, List.of("Feldpost")));
+
+        NlSearchResponse resp = service.search("Feldpost aus dem Krieg", PAGE);
+
+        assertThat(resp.interpretation().keywordsApplied()).isTrue();
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/search/NlSearchControllerTest.java

@@ -0,0 +1,161 @@
+package org.raddatz.familienarchiv.search;
+
+import tools.jackson.databind.ObjectMapper;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.document.DocumentSearchResult;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.raddatz.familienarchiv.security.SecurityConfig;
+import org.raddatz.familienarchiv.security.PermissionAspect;
+import org.raddatz.familienarchiv.user.CustomUserDetailsService;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.boot.autoconfigure.aop.AopAutoConfiguration;
+import org.springframework.boot.webmvc.test.autoconfigure.WebMvcTest;
+import org.springframework.context.annotation.Import;
+import org.springframework.http.MediaType;
+import org.springframework.security.test.context.support.WithMockUser;
+import org.springframework.test.context.bean.override.mockito.MockitoBean;
+import org.springframework.test.web.servlet.MockMvc;
+
+import java.util.List;
+import java.util.UUID;
+
+import static org.mockito.ArgumentMatchers.*;
+import static org.mockito.Mockito.when;
+import static org.springframework.security.test.web.servlet.request.SecurityMockMvcRequestPostProcessors.csrf;
+import static org.springframework.test.web.servlet.request.MockMvcRequestBuilders.post;
+import static org.springframework.test.web.servlet.result.MockMvcResultMatchers.*;
+
+@WebMvcTest(NlSearchController.class)
+@Import({SecurityConfig.class, PermissionAspect.class, AopAutoConfiguration.class,
+        NlSearchRateLimiter.class, NlSearchRateLimitProperties.class})
+class NlSearchControllerTest {
+
+    @Autowired MockMvc mockMvc;
+    private final ObjectMapper objectMapper = new ObjectMapper();
+
+    @MockitoBean NlQueryParserService nlQueryParserService;
+    @MockitoBean CustomUserDetailsService customUserDetailsService;
+    @Autowired NlSearchRateLimiter rateLimiter;
+
+    @BeforeEach
+    void resetRateLimiter() {
+        rateLimiter.resetForTest();
+    }
+
+    private NlSearchResponse makeResponse() {
+        PersonHint hint = new PersonHint(UUID.randomUUID(), "Walter Raddatz");
+        NlQueryInterpretation interp = new NlQueryInterpretation(
+                List.of(hint), List.of(), null, null,
+                List.of("Krieg"), "Briefe von Walter im Krieg", true);
+        return new NlSearchResponse(DocumentSearchResult.of(List.of()), interp);
+    }
+
+    // --- 1. Happy path ---
+
+    @Test
+    @WithMockUser(username = "user@test.com", authorities = {"READ_ALL"})
+    void search_returns200_withNlSearchResponse() throws Exception {
+        when(nlQueryParserService.search(anyString(), any())).thenReturn(makeResponse());
+
+        mockMvc.perform(post("/api/search/nl").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"query\":\"Briefe von Walter im Krieg\"}"))
+                .andExpect(status().isOk())
+                .andExpect(jsonPath("$.interpretation.rawQuery").value("Briefe von Walter im Krieg"))
+                .andExpect(jsonPath("$.interpretation.resolvedPersons[0].displayName").value("Walter Raddatz"))
+                .andExpect(jsonPath("$.interpretation.keywordsApplied").value(true));
+    }
+
+    // --- 2. ambiguousPersons in response shape ---
+
+    @Test
+    @WithMockUser(username = "user@test.com", authorities = {"READ_ALL"})
+    void search_returns200_withAmbiguousPersons() throws Exception {
+        PersonHint a = new PersonHint(UUID.randomUUID(), "Walter Braun");
+        PersonHint b = new PersonHint(UUID.randomUUID(), "Walter Schmidt");
+        NlQueryInterpretation interp = new NlQueryInterpretation(
+                List.of(), List.of(a, b), null, null,
+                List.of(), "Briefe von Walter", false);
+        NlSearchResponse resp = new NlSearchResponse(DocumentSearchResult.of(List.of()), interp);
+        when(nlQueryParserService.search(anyString(), any())).thenReturn(resp);
+
+        mockMvc.perform(post("/api/search/nl").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"query\":\"Briefe von Walter\"}"))
+                .andExpect(status().isOk())
+                .andExpect(jsonPath("$.interpretation.ambiguousPersons").isArray())
+                .andExpect(jsonPath("$.interpretation.ambiguousPersons[0].displayName").value("Walter Braun"))
+                .andExpect(jsonPath("$.interpretation.ambiguousPersons[1].id").isNotEmpty());
+    }
+
+    // --- 3. Unauthenticated → 401 ---
+
+    @Test
+    void search_returns401_whenUnauthenticated() throws Exception {
+        mockMvc.perform(post("/api/search/nl").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"query\":\"Briefe von Walter\"}"))
+                .andExpect(status().isUnauthorized());
+    }
+
+    // --- 4. Query < 3 chars → 400 ---
+
+    @Test
+    @WithMockUser(username = "user@test.com", authorities = {"READ_ALL"})
+    void search_returns400_whenQueryTooShort() throws Exception {
+        mockMvc.perform(post("/api/search/nl").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"query\":\"ab\"}"))
+                .andExpect(status().isBadRequest());
+    }
+
+    // --- 5. Query > 500 chars → 400 ---
+
+    @Test
+    @WithMockUser(username = "user@test.com", authorities = {"READ_ALL"})
+    void search_returns400_whenQueryTooLong() throws Exception {
+        String longQuery = "a".repeat(501);
+        mockMvc.perform(post("/api/search/nl").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"query\":\"" + longQuery + "\"}"))
+                .andExpect(status().isBadRequest());
+    }
+
+    // --- 6. Ollama unavailable → 503 ---
+
+    @Test
+    @WithMockUser(username = "user@test.com", authorities = {"READ_ALL"})
+    void search_returns503_whenOllamaUnavailable() throws Exception {
+        when(nlQueryParserService.search(anyString(), any()))
+                .thenThrow(DomainException.serviceUnavailable(ErrorCode.SMART_SEARCH_UNAVAILABLE, "Ollama offline"));
+
+        mockMvc.perform(post("/api/search/nl").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"query\":\"Briefe von Walter\"}"))
+                .andExpect(status().isServiceUnavailable())
+                .andExpect(jsonPath("$.code").value("SMART_SEARCH_UNAVAILABLE"));
+    }
+
+    // --- 7. 6th request in 1 minute → 429 ---
+
+    @Test
+    @WithMockUser(username = "user@test.com", authorities = {"READ_ALL"})
+    void search_returns429_onSixthRequestWithinRateLimit() throws Exception {
+        when(nlQueryParserService.search(anyString(), any())).thenReturn(makeResponse());
+
+        for (int i = 0; i < 5; i++) {
+            mockMvc.perform(post("/api/search/nl").with(csrf())
+                            .contentType(MediaType.APPLICATION_JSON)
+                            .content("{\"query\":\"Briefe von Walter\"}"))
+                    .andExpect(status().isOk());
+        }
+
+        mockMvc.perform(post("/api/search/nl").with(csrf())
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"query\":\"Briefe von Walter\"}"))
+                .andExpect(status().isTooManyRequests())
+                .andExpect(jsonPath("$.code").value("SMART_SEARCH_RATE_LIMITED"));
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/search/NlSearchRateLimiterTest.java

@@ -0,0 +1,62 @@
+package org.raddatz.familienarchiv.search;
+
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+
+import static org.assertj.core.api.Assertions.assertThatCode;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
+
+class NlSearchRateLimiterTest {
+
+    private NlSearchRateLimiter rateLimiter;
+
+    @BeforeEach
+    void setUp() {
+        NlSearchRateLimitProperties props = new NlSearchRateLimitProperties();
+        props.setMaxRequestsPerMinute(5);
+        rateLimiter = new NlSearchRateLimiter(props);
+    }
+
+    @Test
+    void checkAndConsume_allowsRequestsWithinLimit() {
+        for (int i = 0; i < 5; i++) {
+            assertThatCode(() -> rateLimiter.checkAndConsume("user@example.com"))
+                    .doesNotThrowAnyException();
+        }
+    }
+
+    @Test
+    void checkAndConsume_throwsRateLimited_onSixthRequest() {
+        for (int i = 0; i < 5; i++) {
+            rateLimiter.checkAndConsume("user@example.com");
+        }
+
+        assertThatThrownBy(() -> rateLimiter.checkAndConsume("user@example.com"))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.SMART_SEARCH_RATE_LIMITED);
+    }
+
+    @Test
+    void checkAndConsume_limitsAreIndependentPerUser() {
+        for (int i = 0; i < 5; i++) {
+            rateLimiter.checkAndConsume("alice@example.com");
+        }
+        assertThatCode(() -> rateLimiter.checkAndConsume("bob@example.com"))
+                .doesNotThrowAnyException();
+    }
+
+    @Test
+    void resetForTest_clearsAllBuckets() {
+        for (int i = 0; i < 5; i++) {
+            rateLimiter.checkAndConsume("user@example.com");
+        }
+
+        rateLimiter.resetForTest();
+
+        assertThatCode(() -> rateLimiter.checkAndConsume("user@example.com"))
+                .doesNotThrowAnyException();
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/search/RestClientOllamaClientTest.java

@@ -0,0 +1,113 @@
+package org.raddatz.familienarchiv.search;
+
+import com.github.tomakehurst.wiremock.WireMockServer;
+import com.github.tomakehurst.wiremock.core.WireMockConfiguration;
+import org.junit.jupiter.api.AfterEach;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+
+import static com.github.tomakehurst.wiremock.client.WireMock.*;
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
+
+class RestClientOllamaClientTest {
+
+    private WireMockServer wireMock;
+    private RestClientOllamaClient client;
+
+    @BeforeEach
+    void setUp() {
+        wireMock = new WireMockServer(WireMockConfiguration.wireMockConfig().dynamicPort());
+        wireMock.start();
+
+        OllamaProperties props = new OllamaProperties();
+        props.setBaseUrl("http://localhost:" + wireMock.port());
+        props.setModel("qwen2.5:7b-instruct-q4_K_M");
+        props.setTimeoutSeconds(5);
+        props.setHealthCheckTimeoutSeconds(2);
+
+        client = new RestClientOllamaClient(props);
+    }
+
+    @AfterEach
+    void tearDown() {
+        wireMock.stop();
+    }
+
+    // --- Factory helpers ---
+
+    private String makeOllamaResponseJson(String personNamesJson, String personRole,
+                                           String dateFrom, String dateTo, String keywordsJson) {
+        String inner = String.format(
+                "{\"personNames\":%s,\"personRole\":\"%s\",\"dateFrom\":%s,\"dateTo\":%s,\"keywords\":%s}",
+                personNamesJson, personRole,
+                dateFrom == null ? "null" : "\"" + dateFrom + "\"",
+                dateTo == null ? "null" : "\"" + dateTo + "\"",
+                keywordsJson
+        );
+        return String.format("{\"model\":\"qwen2.5:7b-instruct-q4_K_M\",\"response\":\"%s\",\"done\":true}",
+                inner.replace("\"", "\\\""));
+    }
+
+    // --- Test cases ---
+
+    @Test
+    void parse_returnsExtraction_whenOllamaReturnsValidJson() {
+        String body = makeOllamaResponseJson("[\"Walter\"]", "sender", "1914-01-01", "1914-12-31", "[\"Krieg\"]");
+        wireMock.stubFor(post(urlEqualTo("/api/generate"))
+                .willReturn(aResponse()
+                        .withStatus(200)
+                        .withHeader("Content-Type", "application/json")
+                        .withBody(body)));
+
+        OllamaExtraction result = client.parse("Was hat Walter im Krieg geschrieben?");
+
+        assertThat(result.personNames()).containsExactly("Walter");
+        assertThat(result.personRole()).isEqualTo("sender");
+        assertThat(result.keywords()).containsExactly("Krieg");
+        assertThat(result.dateFrom()).isNotNull();
+        assertThat(result.dateTo()).isNotNull();
+    }
+
+    @Test
+    void parse_throwsSmartSearchUnavailable_whenOllamaReturns500() {
+        wireMock.stubFor(post(urlEqualTo("/api/generate"))
+                .willReturn(aResponse().withStatus(500)));
+
+        assertThatThrownBy(() -> client.parse("some query"))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.SMART_SEARCH_UNAVAILABLE);
+    }
+
+    @Test
+    void parse_throwsSmartSearchUnavailable_whenOllamaExceedsTimeout() {
+        wireMock.stubFor(post(urlEqualTo("/api/generate"))
+                .willReturn(aResponse()
+                        .withStatus(200)
+                        .withHeader("Content-Type", "application/json")
+                        .withFixedDelay(6000)
+                        .withBody("{\"response\":\"{}\",\"done\":true}")));
+
+        assertThatThrownBy(() -> client.parse("some query"))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.SMART_SEARCH_UNAVAILABLE);
+    }
+
+    @Test
+    void parse_throwsSmartSearchUnavailable_whenOllamaReturnsMalformedJson() {
+        wireMock.stubFor(post(urlEqualTo("/api/generate"))
+                .willReturn(aResponse()
+                        .withStatus(200)
+                        .withHeader("Content-Type", "application/json")
+                        .withBody("{\"response\":\"not-json-at-all\",\"done\":true}")));
+
+        assertThatThrownBy(() -> client.parse("some query"))
+                .isInstanceOf(DomainException.class)
+                .extracting(e -> ((DomainException) e).getCode())
+                .isEqualTo(ErrorCode.SMART_SEARCH_UNAVAILABLE);
+    }
+}