M51: record zero-error rollout acceptance

The first real platform runs (2026-08-17, new instance runner) failed at
'Unable to resolve 0.30.0: reference not found' - the tag exists only as v0.30.0.
Where present, the trufflehog GitHub Action (which fails under the act runner) is
replaced by the pinned trufflehog binary in filesystem mode next to gitleaks.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.dockerignore

@@ -0,0 +1,21 @@
+# Backend image build context is the repository root (see compose.yaml); keep it small.
+.git
+.gitea
+.state
+.mypy_cache
+.ruff_cache
+.pytest_cache
+**/__pycache__
+**/.venv
+**/node_modules
+**/dist
+**/playwright-report
+**/test-results
+artifacts
+docs
+deploy
+n8n
+frontend
+*.tgz
+*.tar.gz
+.env

.env.example

@@ -2,24 +2,82 @@ COMPOSE_PROJECT_NAME=mobilityops
 MOBILITYOPS_ENV=development
 MOBILITYOPS_DEMO_MODE=true
 MOBILITYOPS_PUBLIC_URL=http://localhost:1228
-MOBILITYOPS_API_URL=http://localhost:8128
+# Build-time API origin baked into the web bundle. Leave empty: the SPA calls its own
+# origin and nginx proxies /api to the API (required by the CSP connect-src 'self').
+VITE_API_BASE_URL=
 DATABASE_URL=postgresql+psycopg://mobilityops:mobilityops@db:5432/mobilityops
 POSTGRES_DB=mobilityops
 POSTGRES_USER=mobilityops
 POSTGRES_PASSWORD=mobilityops
+# Signs session cookies. With MOBILITYOPS_ENV=production the API refuses to start while
+# this (or MOBILITYOPS_CALLBACK_TOKEN) still holds its placeholder value.
 APP_SECRET=replace-in-production
 TZ=Europe/Brussels
-# Session cookie Secure flag. Keep false for LAN/plain-HTTP deployments (including the
+# Session cookie Secure flag. Development on localhost may use false; production startup
-# current Unraid review environment); set true only once MobilityOps is served over HTTPS,
+# requires both an HTTPS public URL and this value set to true.
-# otherwise browsers will silently drop the cookie and no one can log in.
 SESSION_COOKIE_SECURE=false
 
+# Optional OpenID Connect login. Public demo role buttons remain available when enabled.
+OIDC_ENABLED=false
+OIDC_PROVIDER_NAME=Organisatieaccount
+OIDC_ISSUER_URL=
+OIDC_CLIENT_ID=
+OIDC_CLIENT_SECRET=
+OIDC_REDIRECT_URI=
+OIDC_ALLOWED_EMAIL_DOMAINS=
+OIDC_AUTO_PROVISION=true
+OIDC_DEFAULT_ROLE=rental_employee
+
+# Observability: JSON logs are always enabled. Set a token only if /metrics is exposed
+# outside the private Compose network; Prometheus can send it as a bearer token.
+LOG_LEVEL=INFO
+# Required when the observability profile is enabled. Keep private and high entropy.
+METRICS_BEARER_TOKEN=replace-me-private-metrics-token
+GRAFANA_ADMIN_USER=admin
+GRAFANA_ADMIN_PASSWORD=change-me-before-start
+# Alertmanager sends every firing/resolved alert and the continuous watchdog to this
+# owner-managed receiver. Production must route it to a channel that is actually watched.
+ALERTMANAGER_WEBHOOK_URL=https://n8n.itworx.tech/webhook/mobilityops-alerts
+
+# Verified scheduled PostgreSQL backups (Unraid override).
+BACKUP_INTERVAL_SECONDS=86400
+BACKUP_RETENTION_DAYS=30
+BACKUP_MINIMUM_COPIES=7
+# Restore the newest dump into a disposable database at least weekly. Backup health also
+# requires a successful drill within eight days.
+BACKUP_RESTORE_DRILL_INTERVAL_SECONDS=604800
+# Set both values to copy every verified backup to an independently mounted path.
+BACKUP_SECONDARY_DESTINATION=
+MOBILITYOPS_BACKUP_DIR=./backups/postgres
+MOBILITYOPS_BACKUP_SECONDARY_DIR=./backups/offsite
+# Optional real off-site copy through the official rclone OneDrive adapter. OAuth state
+# lives only in MOBILITYOPS_RCLONE_CONFIG_DIR and must never be committed.
+BACKUP_OFFSITE_INTERVAL_SECONDS=900
+RCLONE_ONEDRIVE_REMOTE=onedrive
+RCLONE_ONEDRIVE_PATH=FleetOps/backups
+MOBILITYOPS_RCLONE_CONFIG_DIR=./.secrets/rclone
+MOBILITYOPS_OFFSITE_VERIFY_DIR=./backups/offsite-verify
+
+# Privacy governance defaults.
+PRIVACY_MINIMUM_BOOKING_RETENTION_DAYS=30
+PRIVACY_AUDIT_RETENTION_DAYS=2555
+PRIVACY_AUDIT_EXPORT_MAX_ROWS=10000
+
 # Demo presentation (fictional org identity, badge/manifest, reset safety valve).
 # DEMO_ALLOW_RESET=false permanently disables POST /api/v1/demo/reset (403), independent
 # of role -- a safety valve for any environment where the dataset must not be rebuildable.
 DEMO_ORGANIZATION_NAME=Northstar Mobility
 DEMO_TIMEZONE=Europe/Brussels
 DEMO_ALLOW_RESET=true
+# Prevent public visitors from repeatedly rebuilding the shared dataset. Concurrent
+# resets are always rejected using both process and PostgreSQL advisory locks.
+DEMO_RESET_COOLDOWN_SECONDS=60
+
+# Operational mode: set MOBILITYOPS_DEMO_MODE=false and provide the first manager.
+# Keep these values in a secret store or an untracked production .env file.
+INITIAL_ADMIN_EMAIL=
+INITIAL_ADMIN_PASSWORD=
+INITIAL_ADMIN_DISPLAY_NAME=Operations Manager
 
 # n8n
 N8N_BASE_URL=http://n8n:5678
@@ -29,6 +87,11 @@ N8N_BASIC_AUTH_ACTIVE=true
 N8N_BASIC_AUTH_USER=admin
 N8N_BASIC_AUTH_PASSWORD=change-me
 MOBILITYOPS_CALLBACK_TOKEN=replace-me-n8n-callback-token
+# Sent as the X-Fleet-Ops-Trigger-Token header when Fleet Ops calls the n8n return-
+# processing webhook, so the webhook trigger can require Header Auth instead of being
+# publicly callable by anyone who discovers the URL. Must match the value stored in
+# n8n's "Fleet Ops Webhook Trigger Token" Header Auth credential.
+MOBILITYOPS_WEBHOOK_TRIGGER_TOKEN=replace-me-n8n-webhook-trigger-token
 
 # RAGcore integration
 KNOWLEDGE_PROVIDER=demo
@@ -37,9 +100,17 @@ RAGCORE_TENANT=northstar-mobility-demo
 RAGCORE_WORKSPACE=mobilityops
 RAGCORE_COLLECTION=internal-procedures
 RAGCORE_API_TOKEN=
+# UUID of the RAGcore knowledge space procedures were synced into (see workflow 3).
+RAGCORE_SPACE_ID=
+# Skip the slower generation endpoint temporarily after a timeout/non-2xx response and
+# use the still-grounded extractive search fallback immediately.
+RAGCORE_ANSWERS_CIRCUIT_BREAKER_SECONDS=60
 
-# ITWorx MCP Hub integration
+# ITWorx MCP Hub integration. Registration itself is catalog-driven on the Hub's own
+# side (it reconciles its catalog into the gateway; Fleet Ops never pushes a
+# registration call) -- MCP_HUB_BASE_URL is only used here for an honest reachability
+# health check surfaced on the integration status page.
 MCP_HUB_REGISTRATION_ENABLED=false
 MCP_HUB_BASE_URL=http://itworx-mcp-hub:8000
 MCP_HUB_SERVICE_TOKEN=replace-me-mcp-hub-token
-MCP_PROVIDER_ID=mobilityops
+MCP_PROVIDER_ID=fleet-ops

.gitea/workflows/ci.yml

@@ -0,0 +1,135 @@
+name: MobilityOps acceptance
+
+on:
+  push:
+    branches: [master]
+  pull_request:
+  schedule:
+    - cron: "17 3 * * 1"
+
+jobs:
+  backend:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - name: Secret scan
+        uses: trufflesecurity/trufflehog@b9dd330365132cd2d01dd5dc8a857a056a2544e1 # v3.79.0
+        with:
+          path: ./
+          extra_args: --only-verified
+      - name: Backend tests in isolated PostgreSQL stack
+        run: sh scripts/run-isolated-tests.sh
+      - name: Backend static checks
+        run: |
+          docker compose -p mobilityops-ci -f compose.yaml -f compose.test.yaml run --build --rm api ruff check app tests
+          docker compose -p mobilityops-ci -f compose.yaml -f compose.test.yaml run --rm api mypy app
+      - name: Contract drift gate
+        run: |
+          docker compose -p mobilityops-ci -f compose.yaml -f compose.test.yaml run --rm \
+            -v "$PWD:/repo:ro" api python /repo/scripts/check-contracts.py
+          python scripts/check-source-budgets.py
+      - name: Build production API image for vulnerability scan
+        run: |
+          docker build --target runtime --build-arg VCS_REF="$GITHUB_SHA" \
+            --tag mobilityops-api-ci --file backend/Dockerfile .
+      - name: Production API image vulnerability scan (HIGH/CRITICAL)
+        uses: aquasecurity/trivy-action@6c175e9c4083a92bbca2f9724c8a5e33bc2d97a5 # v0.30.0
+        with:
+          scan-type: image
+          image-ref: mobilityops-api-ci
+          format: table
+          severity: HIGH,CRITICAL
+          exit-code: "1"
+          ignore-unfixed: true
+      - name: Build production web image for vulnerability scan
+        run: |
+          docker build --build-arg VCS_REF="$GITHUB_SHA" \
+            --tag mobilityops-web-ci frontend
+      - name: Production web image vulnerability scan (HIGH/CRITICAL)
+        uses: aquasecurity/trivy-action@6c175e9c4083a92bbca2f9724c8a5e33bc2d97a5 # v0.30.0
+        with:
+          scan-type: image
+          image-ref: mobilityops-web-ci
+          format: table
+          severity: HIGH,CRITICAL
+          exit-code: "1"
+          ignore-unfixed: true
+      - name: Remove CI stack
+        if: always()
+        run: docker compose -p mobilityops-ci -f compose.yaml -f compose.test.yaml down -v --remove-orphans
+
+  frontend:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+        with:
+          node-version: 22
+          cache: npm
+          cache-dependency-path: frontend/package-lock.json
+      - name: Install locked dependencies
+        working-directory: frontend
+        run: npm ci --no-audit --no-fund
+      - name: Lint (tsc + ESLint with react-hooks and jsx-a11y)
+        working-directory: frontend
+        run: npm run lint
+      - name: Typecheck and production build
+        working-directory: frontend
+        run: npm run build && npm run budget
+      - name: Dependency audit
+        working-directory: frontend
+        run: npm audit --audit-level=high
+
+  e2e:
+    # The five-minute Playwright demo is part of the definition of done
+    # (docs/14-testing-and-acceptance.md); run it against the real Compose stack.
+    runs-on: ubuntu-latest
+    needs: [backend, frontend]
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+        with:
+          node-version: 22
+          cache: npm
+          cache-dependency-path: frontend/package-lock.json
+      - name: Start the demo stack
+        run: |
+          cp .env.example .env
+          docker compose -p mobilityops-e2e up --build -d db api web
+          for _attempt in $(seq 1 60); do
+            if curl -fsS http://localhost:1228/health/ready >/dev/null 2>&1; then break; fi
+            sleep 2
+          done
+          curl -fsS http://localhost:1228/health/ready
+          docker compose -p mobilityops-e2e exec -T api python -m app.cli seed --reset
+      - name: Install Playwright
+        working-directory: frontend
+        run: |
+          npm ci --no-audit --no-fund
+          npx playwright install --with-deps chromium firefox
+      - name: Run browser acceptance suite
+        working-directory: frontend
+        env:
+          MOBILITYOPS_PUBLIC_URL: http://localhost:1228
+        run: npx playwright test
+      - name: Run non-destructive Chromium and Firefox smoke suite
+        working-directory: frontend
+        env:
+          MOBILITYOPS_PUBLIC_URL: http://localhost:1228
+        run: npx playwright test --config=playwright.live.config.ts
+      - name: Run concurrent persisted-read smoke
+        run: python scripts/run-readonly-load-smoke.py --base-url http://localhost:1228
+      - name: Upload Playwright report
+        if: failure()
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        with:
+          name: playwright-report
+          path: |
+            frontend/playwright-report
+            frontend/playwright-live-report
+      - name: Stack logs on failure
+        if: failure()
+        run: docker compose -p mobilityops-e2e logs --tail=200 api web
+      - name: Remove e2e stack
+        if: always()
+        run: docker compose -p mobilityops-e2e down -v --remove-orphans

.gitea/workflows/live-canary.yml

@@ -0,0 +1,47 @@
+name: MobilityOps live canary
+
+on:
+  schedule:
+    - cron: "7 * * * *"
+  workflow_dispatch:
+
+jobs:
+  public-demo:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+        with:
+          node-version: 22
+          cache: npm
+          cache-dependency-path: frontend/package-lock.json
+      - name: Verify HTTPS readiness and certificate horizon
+        run: |
+          curl --fail --silent --show-error --retry 3 https://fleetops.itworx.tech/health/ready
+          openssl s_client -servername fleetops.itworx.tech -connect fleetops.itworx.tech:443 </dev/null 2>/dev/null \
+            | openssl x509 -checkend 1209600 -noout
+      - name: Install locked Playwright runtime
+        working-directory: frontend
+        run: |
+          npm ci --no-audit --no-fund
+          npx playwright install --with-deps chromium firefox
+      - name: Run non-destructive cross-browser production canary
+        working-directory: frontend
+        env:
+          MOBILITYOPS_PUBLIC_URL: https://fleetops.itworx.tech
+        run: npx playwright test --config=playwright.live.config.ts
+      - name: Report successful external heartbeat
+        env:
+          HEARTBEAT_URL: ${{ secrets.LIVE_CANARY_HEARTBEAT_URL }}
+        run: |
+          if [ -n "$HEARTBEAT_URL" ]; then
+            curl --fail --silent --show-error --retry 3 "$HEARTBEAT_URL"
+          fi
+      - name: Upload failure evidence
+        if: failure()
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        with:
+          name: live-canary-failure
+          path: |
+            frontend/playwright-live-report
+            frontend/test-results

.gitea/workflows/release.yml

@@ -0,0 +1,62 @@
+name: MobilityOps release evidence
+
+on:
+  push:
+    tags: ["v*"]
+
+jobs:
+  release-evidence:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - name: Build commit-labelled release images
+        run: |
+          docker build --target runtime --build-arg VCS_REF="$GITHUB_SHA" --tag mobilityops-api-release --file backend/Dockerfile .
+          docker build --build-arg VCS_REF="$GITHUB_SHA" --tag mobilityops-web-release frontend
+          docker build --build-arg VCS_REF="$GITHUB_SHA" --tag mobilityops-backup-tools-release --file deploy/unraid/Dockerfile.backup-tools .
+      - name: Scan all release images
+        run: |
+          for image in mobilityops-api-release mobilityops-web-release mobilityops-backup-tools-release; do
+            docker run --rm -v /var/run/docker.sock:/var/run/docker.sock \
+              -v mobilityops-release-trivy:/root/.cache/ \
+              aquasec/trivy:0.74.0@sha256:62b1e65e8869bc4b4c6aa4fa2b21595256c7c2f6018a9d9ad61caf87187c1969 \
+              image --scanners vuln --severity HIGH,CRITICAL \
+              --ignore-unfixed --exit-code 1 "$image"
+          done
+      - name: Generate API CycloneDX SBOM
+        uses: aquasecurity/trivy-action@6c175e9c4083a92bbca2f9724c8a5e33bc2d97a5 # v0.30.0
+        with:
+          scan-type: image
+          image-ref: mobilityops-api-release
+          format: cyclonedx
+          output: mobilityops-api-sbom.cdx.json
+      - name: Generate web CycloneDX SBOM
+        uses: aquasecurity/trivy-action@6c175e9c4083a92bbca2f9724c8a5e33bc2d97a5 # v0.30.0
+        with:
+          scan-type: image
+          image-ref: mobilityops-web-release
+          format: cyclonedx
+          output: mobilityops-web-sbom.cdx.json
+      - name: Generate backup-tools CycloneDX SBOM
+        uses: aquasecurity/trivy-action@6c175e9c4083a92bbca2f9724c8a5e33bc2d97a5 # v0.30.0
+        with:
+          scan-type: image
+          image-ref: mobilityops-backup-tools-release
+          format: cyclonedx
+          output: mobilityops-backup-tools-sbom.cdx.json
+      - name: Record immutable image metadata
+        run: |
+          docker image inspect mobilityops-api-release > mobilityops-api-image.json
+          docker image inspect mobilityops-web-release > mobilityops-web-image.json
+          docker image inspect mobilityops-backup-tools-release > mobilityops-backup-tools-image.json
+          python scripts/generate-release-provenance.py
+          sha256sum mobilityops-*-sbom.cdx.json mobilityops-*-image.json release-provenance.json > SHA256SUMS
+      - name: Upload release evidence
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        with:
+          name: mobilityops-${{ github.ref_name }}-evidence
+          path: |
+            mobilityops-*-sbom.cdx.json
+            mobilityops-*-image.json
+            release-provenance.json
+            SHA256SUMS

.gitignore

@@ -8,9 +8,18 @@ node_modules/
 dist/
 coverage/
 playwright-report/
+playwright-live-report/
 test-results/
 *.pyc
 .DS_Store
 .idea/
 .vscode/
 *.tsbuildinfo
+*.zip
+*.tar.gz
+
+# Local Claude/Codex per-user settings and scratch archives
+.claude/settings.local.json
+*.tgz
+*.dump
+backups/

CHANGELOG.md

@@ -0,0 +1,18 @@
+# Changelog
+
+All notable changes are documented here. The project follows semantic release tags for
+the deployable PoC; detailed validation evidence remains in `PROJECT_STATE.md`.
+
+## [Unreleased]
+
+- Added contract-drift, accessibility, Firefox smoke and frontend asset-budget gates.
+- Added immutable commit-labelled deployment with automatic application rollback.
+- Added real PostgreSQL restore drills and routed Alertmanager notifications.
+- Added RAGcore generation circuit breaking and retrieval telemetry.
+- Added scheduled dependency maintenance, dual-image vulnerability scans and release SBOMs.
+
+## [1.0.0-poc] - 2026-08-21
+
+- Completed the locked Fleet Ops proof of concept: operational core, transactional returns,
+  data quality, n8n orchestration, grounded RAGcore knowledge, read-only MCP integration,
+  privacy governance, observability, backup/recovery and full browser acceptance.

FILE_INDEX.md

@@ -1,25 +1,147 @@
 # File index
 
+Tracked source, contract, documentation and configuration files. Release evidence and
+screenshots live under `artifacts/<release>/` and are omitted here for brevity.
+Regenerate with `git ls-files` when the tree changes.
+
 - `.env.example`
+- `.gitattributes`
+- `.gitea/workflows/ci.yml`
 - `.gitignore`
+- `AGENTS.md`
 - `CLAUDE.md`
+- `FILE_INDEX.md`
 - `MASTER_BUILD_PROMPT.md`
 - `Makefile`
 - `PROJECT_STATE.md`
 - `README.md`
 - `START_HERE.md`
 - `backend/Dockerfile`
+- `backend/alembic.ini`
+- `backend/alembic/env.py`
+- `backend/alembic/script.py.mako`
+- `backend/alembic/versions/0a4c1d2e3f5b_idempotency_request_fingerprint.py`
+- `backend/alembic/versions/799d8800e241_outbox_last_error_code.py`
+- `backend/alembic/versions/a81d0ce9f662_oidc_identity.py`
+- `backend/alembic/versions/b7c7b536df85_operational_user_credentials.py`
+- `backend/alembic/versions/b913a72e8c14_customer_privacy_state.py`
+- `backend/alembic/versions/c24f6a9d013e_domain_constraints_indexes.py`
+- `backend/alembic/versions/c9498525abb5_initial_schema.py`
+- `backend/alembic/versions/d1f83bc64170_revoked_sessions.py`
+- `backend/alembic/versions/e7b08389f47f_idempotency_records.py`
+- `backend/alembic/versions/f43d829ab610_quality_work_queue.py`
 - `backend/app/__init__.py`
+- `backend/app/api/__init__.py`
+- `backend/app/api/deps.py`
+- `backend/app/api/routers/__init__.py`
+- `backend/app/api/routers/audit.py`
+- `backend/app/api/routers/auth.py`
+- `backend/app/api/routers/bookings.py`
+- `backend/app/api/routers/customers.py`
+- `backend/app/api/routers/dashboard.py`
+- `backend/app/api/routers/data_quality.py`
+- `backend/app/api/routers/demo.py`
+- `backend/app/api/routers/integration_status.py`
+- `backend/app/api/routers/integrations.py`
+- `backend/app/api/routers/knowledge.py`
+- `backend/app/api/routers/mcp_integrations.py`
+- `backend/app/api/routers/observability.py`
+- `backend/app/api/routers/privacy.py`
+- `backend/app/api/routers/search.py`
+- `backend/app/api/routers/users.py`
+- `backend/app/api/routers/vehicles.py`
+- `backend/app/api/routers/workflows.py`
+- `backend/app/cli.py`
 - `backend/app/core/__init__.py`
 - `backend/app/core/config.py`
+- `backend/app/core/db.py`
+- `backend/app/core/errors.py`
+- `backend/app/core/observability.py`
+- `backend/app/core/ratelimit.py`
+- `backend/app/core/security.py`
 - `backend/app/main.py`
+- `backend/app/models/__init__.py`
+- `backend/app/models/audit.py`
+- `backend/app/models/booking.py`
+- `backend/app/models/customer.py`
+- `backend/app/models/data_quality.py`
+- `backend/app/models/idempotency.py`
+- `backend/app/models/inspection.py`
+- `backend/app/models/maintenance.py`
+- `backend/app/models/mixins.py`
+- `backend/app/models/outbox.py`
+- `backend/app/models/revoked_session.py`
+- `backend/app/models/user.py`
+- `backend/app/models/vehicle.py`
+- `backend/app/schemas.py`
+- `backend/app/seed_loader.py`
+- `backend/app/services/__init__.py`
+- `backend/app/services/audit.py`
+- `backend/app/services/data_quality.py`
+- `backend/app/services/demo_manifest.py`
+- `backend/app/services/dispatcher.py`
+- `backend/app/services/integration_status.py`
+- `backend/app/services/knowledge/__init__.py`
+- `backend/app/services/knowledge/demo.py`
+- `backend/app/services/knowledge/procedures.py`
+- `backend/app/services/knowledge/ragcore.py`
+- `backend/app/services/operations.py`
+- `backend/app/services/returns.py`
+- `backend/app/services/sessions.py`
+- `backend/app/services/vehicle_status.py`
+- `backend/entrypoint.sh`
 - `backend/pyproject.toml`
+- `backend/requirements.lock`
+- `backend/scripts/generate_openapi.py`
+- `backend/tests/conftest.py`
+- `backend/tests/test_audit.py`
+- `backend/tests/test_auth.py`
+- `backend/tests/test_bookings.py`
+- `backend/tests/test_dashboard.py`
+- `backend/tests/test_data_quality.py`
+- `backend/tests/test_database_constraints.py`
+- `backend/tests/test_demo_manifest.py`
+- `backend/tests/test_dispatcher.py`
+- `backend/tests/test_hardening.py`
 - `backend/tests/test_health.py`
+- `backend/tests/test_integration_status.py`
+- `backend/tests/test_integrations.py`
+- `backend/tests/test_knowledge.py`
+- `backend/tests/test_mcp_integrations.py`
+- `backend/tests/test_migrations.py`
+- `backend/tests/test_observability.py`
+- `backend/tests/test_operational_auth.py`
+- `backend/tests/test_privacy.py`
+- `backend/tests/test_return.py`
+- `backend/tests/test_search.py`
+- `backend/tests/test_seed.py`
+- `backend/tests/test_users.py`
+- `backend/tests/test_vehicle_status.py`
+- `backend/tests/test_vehicles.py`
+- `backend/tests/test_workflows.py`
+- `compose.observability.yaml`
+- `compose.test.yaml`
+- `compose.unraid.yaml`
 - `compose.yaml`
 - `contracts/events.schema.json`
 - `contracts/mcp-tools.json`
 - `contracts/openapi.yaml`
 - `contracts/ragcore-contract-assumptions.md`
+- `deploy/observability/alerts.yml`
+- `deploy/observability/grafana/dashboards/mobilityops-overview.json`
+- `deploy/observability/grafana/provisioning/dashboards/mobilityops.yml`
+- `deploy/observability/grafana/provisioning/datasources/prometheus.yml`
+- `deploy/observability/prometheus.yml`
+- `deploy/unraid/README.md`
+- `deploy/unraid/backup-postgres.sh`
+- `deploy/unraid/configure-env.sh`
+- `deploy/unraid/prune-postgres-backups.sh`
+- `deploy/unraid/restore-postgres.sh`
+- `deploy/unraid/scheduled-backup.sh`
+- `deploy/unraid/setup-existing-n8n.sh`
+- `deploy/unraid/setup-n8n.sh`
+- `deploy/unraid/setup-scheduled-scan.sh`
+- `deploy/unraid/verify-postgres-backups.sh`
 - `docs/00-product-brief.md`
 - `docs/01-scope-and-non-goals.md`
 - `docs/02-user-stories.md`
@@ -38,29 +160,203 @@
 - `docs/15-build-plan.md`
 - `docs/16-portfolio-case-study.md`
 - `docs/17-runbook.md`
+- `docs/18-privacy-governance.md`
+- `docs/19-visual-product-roadmap.md`
 - `docs/deferred.md`
+- `docs/demo-release/current-demo-gap-audit.md`
+- `docs/demo-release/demo-concept.md`
+- `docs/demo-release/demo-data.md`
+- `docs/demo-release/demo-guide.md`
+- `docs/demo-release/demo-runbook.md`
+- `docs/demo-release/demo-scenarios.md`
+- `docs/design/current-ux-audit.md`
+- `docs/design/design-directions.md`
+- `docs/design/design-system.md`
+- `docs/design/implementation-validation.md`
+- `docs/design/stitch-manifest.md`
+- `docs/final-integrations/ai-operations-brief-runbook.md`
+- `docs/final-integrations/current-state-audit.md`
+- `docs/final-product-polish/audit.md`
+- `docs/fleet-ops-correction/current-gap-audit.md`
+- `docs/fleet-ops-correction/i18n-inventory.md`
+- `docs/fleet-ops-correction/vehicle-status-decision-table.md`
+- `docs/fleet-ops-final-localization/audit.md`
+- `docs/functional-completion/current-functional-audit.md`
+- `docs/functional-completion/server-baseline.md`
+- `docs/live-ai-integration/n8n-current-state.md`
 - `frontend/Dockerfile`
+- `frontend/e2e/_capture-demo-screenshots.spec.ts`
+- `frontend/e2e/_capture-recruiter-screenshots.spec.ts`
+- `frontend/e2e/_capture-screenshots.spec.ts`
+- `frontend/e2e/clickable-rows.spec.ts`
+- `frontend/e2e/demo-accessibility.spec.ts`
+- `frontend/e2e/demo-entry.spec.ts`
+- `frontend/e2e/demo-guide.spec.ts`
+- `frontend/e2e/demo-legibility.spec.ts`
+- `frontend/e2e/demo.spec.ts`
+- `frontend/e2e/error-messages.spec.ts`
+- `frontend/e2e/fleet-ops-correction.spec.ts`
+- `frontend/e2e/greeting-live.spec.ts`
+- `frontend/e2e/greeting.spec.ts`
+- `frontend/e2e/guided-demo-full.spec.ts`
+- `frontend/e2e/i18n-coverage.spec.ts`
+- `frontend/e2e/interactive-elements.spec.ts`
+- `frontend/e2e/operational-workflows.spec.ts`
+- `frontend/e2e/privacy.spec.ts`
+- `frontend/e2e/recruiter-polish.spec.ts`
+- `frontend/e2e/responsive-i18n.spec.ts`
+- `frontend/e2e/roadmap-regression.spec.ts`
+- `frontend/e2e/ui-redesign.spec.ts`
 - `frontend/index.html`
 - `frontend/nginx.conf`
+- `frontend/package-lock.json`
 - `frontend/package.json`
+- `frontend/playwright.config.ts`
+- `frontend/public/favicon.svg`
+- `frontend/public/og-fleet-ops.svg`
 - `frontend/src/App.tsx`
+- `frontend/src/api/apiError.ts`
+- `frontend/src/api/client.ts`
+- `frontend/src/api/errorMessages.ts`
+- `frontend/src/api/types.ts`
+- `frontend/src/components/Badge.tsx`
+- `frontend/src/components/CheckoutForm.tsx`
+- `frontend/src/components/DemoBadge.tsx`
+- `frontend/src/components/DemoGuide.tsx`
+- `frontend/src/components/Icons.tsx`
+- `frontend/src/components/LanguageSwitcher.tsx`
+- `frontend/src/components/Layout.tsx`
+- `frontend/src/components/PageChrome.tsx`
+- `frontend/src/components/Pagination.tsx`
+- `frontend/src/components/RequireAuth.tsx`
+- `frontend/src/components/ReturnForm.tsx`
+- `frontend/src/components/VehicleMaintenanceActions.tsx`
+- `frontend/src/context/AuthContext.tsx`
+- `frontend/src/context/DemoGuideContext.tsx`
+- `frontend/src/context/DemoManifestContext.tsx`
+- `frontend/src/data/demoGuideSteps.ts`
+- `frontend/src/data/evidenceSignals.ts`
+- `frontend/src/data/integrationLabels.ts`
+- `frontend/src/hooks/useViewportTier.ts`
+- `frontend/src/i18n/brusselsDateTime.ts`
+- `frontend/src/i18n/config.ts`
+- `frontend/src/i18n/format.ts`
+- `frontend/src/i18n/greeting.ts`
+- `frontend/src/i18n/locales/en-GB/accessibility.json`
+- `frontend/src/i18n/locales/en-GB/audit.json`
+- `frontend/src/i18n/locales/en-GB/auth.json`
+- `frontend/src/i18n/locales/en-GB/bookings.json`
+- `frontend/src/i18n/locales/en-GB/common.json`
+- `frontend/src/i18n/locales/en-GB/dashboard.json`
+- `frontend/src/i18n/locales/en-GB/demo.json`
+- `frontend/src/i18n/locales/en-GB/errors.json`
+- `frontend/src/i18n/locales/en-GB/fleet.json`
+- `frontend/src/i18n/locales/en-GB/integrations.json`
+- `frontend/src/i18n/locales/en-GB/knowledge.json`
+- `frontend/src/i18n/locales/en-GB/navigation.json`
+- `frontend/src/i18n/locales/en-GB/operations.json`
+- `frontend/src/i18n/locales/en-GB/privacy.json`
+- `frontend/src/i18n/locales/en-GB/quality.json`
+- `frontend/src/i18n/locales/en-GB/returns.json`
+- `frontend/src/i18n/locales/fr-BE/accessibility.json`
+- `frontend/src/i18n/locales/fr-BE/audit.json`
+- `frontend/src/i18n/locales/fr-BE/auth.json`
+- `frontend/src/i18n/locales/fr-BE/bookings.json`
+- `frontend/src/i18n/locales/fr-BE/common.json`
+- `frontend/src/i18n/locales/fr-BE/dashboard.json`
+- `frontend/src/i18n/locales/fr-BE/demo.json`
+- `frontend/src/i18n/locales/fr-BE/errors.json`
+- `frontend/src/i18n/locales/fr-BE/fleet.json`
+- `frontend/src/i18n/locales/fr-BE/integrations.json`
+- `frontend/src/i18n/locales/fr-BE/knowledge.json`
+- `frontend/src/i18n/locales/fr-BE/navigation.json`
+- `frontend/src/i18n/locales/fr-BE/operations.json`
+- `frontend/src/i18n/locales/fr-BE/privacy.json`
+- `frontend/src/i18n/locales/fr-BE/quality.json`
+- `frontend/src/i18n/locales/fr-BE/returns.json`
+- `frontend/src/i18n/locales/nl-BE/accessibility.json`
+- `frontend/src/i18n/locales/nl-BE/audit.json`
+- `frontend/src/i18n/locales/nl-BE/auth.json`
+- `frontend/src/i18n/locales/nl-BE/bookings.json`
+- `frontend/src/i18n/locales/nl-BE/common.json`
+- `frontend/src/i18n/locales/nl-BE/dashboard.json`
+- `frontend/src/i18n/locales/nl-BE/demo.json`
+- `frontend/src/i18n/locales/nl-BE/errors.json`
+- `frontend/src/i18n/locales/nl-BE/fleet.json`
+- `frontend/src/i18n/locales/nl-BE/integrations.json`
+- `frontend/src/i18n/locales/nl-BE/knowledge.json`
+- `frontend/src/i18n/locales/nl-BE/navigation.json`
+- `frontend/src/i18n/locales/nl-BE/operations.json`
+- `frontend/src/i18n/locales/nl-BE/privacy.json`
+- `frontend/src/i18n/locales/nl-BE/quality.json`
+- `frontend/src/i18n/locales/nl-BE/returns.json`
+- `frontend/src/i18n/useGreetingPeriod.ts`
 - `frontend/src/main.tsx`
+- `frontend/src/pages/AboutDemo.tsx`
+- `frontend/src/pages/Audit.tsx`
+- `frontend/src/pages/Automation.tsx`
+- `frontend/src/pages/BookingCreate.tsx`
+- `frontend/src/pages/BookingDetail.tsx`
+- `frontend/src/pages/Bookings.tsx`
+- `frontend/src/pages/Dashboard.tsx`
+- `frontend/src/pages/DataQuality.tsx`
+- `frontend/src/pages/DataQualityIssueDetail.tsx`
+- `frontend/src/pages/Highlights.tsx`
+- `frontend/src/pages/Knowledge.tsx`
+- `frontend/src/pages/Login.tsx`
+- `frontend/src/pages/Privacy.tsx`
+- `frontend/src/pages/Scenarios.tsx`
+- `frontend/src/pages/Users.tsx`
+- `frontend/src/pages/VehicleDetail.tsx`
+- `frontend/src/pages/Vehicles.tsx`
+- `frontend/src/product.ts`
 - `frontend/src/styles.css`
+- `frontend/src/vite-env.d.ts`
 - `frontend/tsconfig.json`
 - `frontend/vite.config.ts`
 - `knowledge/manifest.json`
-- `knowledge/procedures/01-vehicle-checkout.md`
+- `knowledge/procedures/en-GB/01-vehicle-checkout.md`
-- `knowledge/procedures/02-vehicle-return.md`
+- `knowledge/procedures/en-GB/02-vehicle-return.md`
-- `knowledge/procedures/03-damage-handling.md`
+- `knowledge/procedures/en-GB/03-damage-handling.md`
-- `knowledge/procedures/04-odometer-anomalies.md`
+- `knowledge/procedures/en-GB/04-odometer-anomalies.md`
-- `knowledge/procedures/05-cleaning-checklist.md`
+- `knowledge/procedures/en-GB/05-cleaning-checklist.md`
-- `knowledge/procedures/06-maintenance-escalation.md`
+- `knowledge/procedures/en-GB/06-maintenance-escalation.md`
-- `knowledge/procedures/07-customer-documents.md`
+- `knowledge/procedures/en-GB/07-customer-documents.md`
-- `knowledge/procedures/08-privacy.md`
+- `knowledge/procedures/en-GB/08-privacy.md`
-- `knowledge/procedures/09-booking-conflicts.md`
+- `knowledge/procedures/en-GB/09-booking-conflicts.md`
-- `knowledge/procedures/10-roles-and-escalation.md`
+- `knowledge/procedures/en-GB/10-roles-and-escalation.md`
+- `knowledge/procedures/en-GB/11-vehicle-availability.md`
+- `knowledge/procedures/fr-BE/01-vehicle-checkout.md`
+- `knowledge/procedures/fr-BE/02-vehicle-return.md`
+- `knowledge/procedures/fr-BE/03-damage-handling.md`
+- `knowledge/procedures/fr-BE/04-odometer-anomalies.md`
+- `knowledge/procedures/fr-BE/05-cleaning-checklist.md`
+- `knowledge/procedures/fr-BE/06-maintenance-escalation.md`
+- `knowledge/procedures/fr-BE/07-customer-documents.md`
+- `knowledge/procedures/fr-BE/08-privacy.md`
+- `knowledge/procedures/fr-BE/09-booking-conflicts.md`
+- `knowledge/procedures/fr-BE/10-roles-and-escalation.md`
+- `knowledge/procedures/fr-BE/11-vehicle-availability.md`
+- `knowledge/procedures/nl-BE/01-vehicle-checkout.md`
+- `knowledge/procedures/nl-BE/02-vehicle-return.md`
+- `knowledge/procedures/nl-BE/03-damage-handling.md`
+- `knowledge/procedures/nl-BE/04-odometer-anomalies.md`
+- `knowledge/procedures/nl-BE/05-cleaning-checklist.md`
+- `knowledge/procedures/nl-BE/06-maintenance-escalation.md`
+- `knowledge/procedures/nl-BE/07-customer-documents.md`
+- `knowledge/procedures/nl-BE/08-privacy.md`
+- `knowledge/procedures/nl-BE/09-booking-conflicts.md`
+- `knowledge/procedures/nl-BE/10-roles-and-escalation.md`
+- `knowledge/procedures/nl-BE/11-vehicle-availability.md`
 - `n8n/README.md`
-- `n8n/mobilityops-return-processing.json`
+- `n8n/workflows/MANIFEST.md`
+- `n8n/workflows/check_drift.py`
+- `n8n/workflows/fleet-ops-data-quality-scan.json`
+- `n8n/workflows/fleet-ops-error-handler.json`
+- `n8n/workflows/fleet-ops-ragcore-procedure-sync.json`
+- `n8n/workflows/fleet-ops-vehicle-return.json`
+- `n8n/workflows/merge_credential_refs.py`
+- `scripts/run-isolated-tests.sh`
 - `seed/README.md`
 - `seed/bookings.csv`
 - `seed/customers.csv`

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Jens Caers
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

Makefile

@@ -1,4 +1,4 @@
-.PHONY: up down logs test lint seed reset n8n-setup n8n-setup-scan demo e2e
+.PHONY: up down logs test lint contracts seed reset n8n-setup n8n-setup-scan demo e2e live-smoke
 
 up:
 	docker compose up --build -d
@@ -10,11 +10,17 @@ logs:
 	docker compose logs -f --tail=200
 
 test:
-	docker compose run --rm api pytest
+	sh scripts/run-isolated-tests.sh
 
 lint:
-	docker compose run --rm api ruff check .
+	docker compose -f compose.yaml -f compose.test.yaml run --build --rm api ruff check app tests
-	docker compose run --rm api mypy app
+	docker compose -f compose.yaml -f compose.test.yaml run --rm api mypy app
+	cd frontend && npm run lint
+
+contracts:
+	docker compose -f compose.yaml -f compose.test.yaml run --build --rm \
+		-v "$(CURDIR):/repo:ro" api python /repo/scripts/check-contracts.py
+	python scripts/check-source-budgets.py
 
 seed:
 	docker compose exec api python -m app.cli seed --reset
@@ -26,16 +32,19 @@ reset:
 # One-time per environment: imports and activates the n8n return-processing workflow.
 # The n8n owner account itself cannot be scripted safely and must be created once at
 # http://localhost:5678/setup (any email/password, no verification required) before
-# this target's activation takes effect. See docs/17-runbook.md.
+# this target's activation takes effect. The workflow also needs the "Fleet Ops Webhook
+# Trigger Token" and "Fleet Ops Service Token" Header Auth credentials created manually in
+# the n8n UI before it will actually process a return -- see docs/17-runbook.md.
 n8n-setup:
-	docker compose exec n8n n8n import:workflow --input=//imports/mobilityops-return-processing.json
+	docker compose exec n8n n8n import:workflow --input=//imports/workflows/fleet-ops-vehicle-return.json
 	docker compose exec n8n n8n publish:workflow --id=mobilityops-return-processing
 	docker compose restart n8n
 
 # One-time per environment: imports and activates the scheduled quality-scan workflow.
-# Same owner-account precondition as n8n-setup above.
+# Same owner-account and credential preconditions as n8n-setup above (this workflow only
+# needs "Fleet Ops Service Token").
 n8n-setup-scan:
-	docker compose exec n8n n8n import:workflow --input=//imports/mobilityops-scheduled-quality-scan.json
+	docker compose exec n8n n8n import:workflow --input=//imports/workflows/fleet-ops-data-quality-scan.json
 	docker compose exec n8n n8n publish:workflow --id=mobilityops-scheduled-quality-scan
 	docker compose restart n8n
 
@@ -45,3 +54,6 @@ demo: up
 
 e2e:
 	cd frontend && npx playwright test
+
+live-smoke:
+	cd frontend && npx playwright test --config=playwright.live.config.ts

README.md

@@ -1,122 +1,95 @@
-# MobilityOps
+# Fleet Ops
 
-**Connected operations for vehicle rental and service teams.**
+**A recruiter-ready operations platform for vehicle rental and service teams.**
 
-MobilityOps is a working proof of concept for a fictitious mobility company. It combines vehicle and booking operations, a controlled vehicle-return workflow, data-quality review, RAGcore-backed internal knowledge, n8n orchestration and read-only tools published through ITWorx MCP Hub.
+**Try it in two commands** (`cp .env.example .env && make demo`, then open `http://localhost:1228`) · no password required · choose **Highlights in 90 seconds** for the shortest tour. The reference deployment runs on a private LAN (see [deploy/unraid/README.md](deploy/unraid/README.md)); ask for a link if you want the hosted version.
 
-The web application uses the premium responsive **Control Rail** interface: a compact
+Fleet Ops turns fragmented vehicle, booking and procedure data into one controlled operational workspace. It is a complete synthetic-data product demo: the company and records are fictional, while the workflows, persistence, validation, authorization, audit trail and integration boundaries are implemented.
-operations-first workspace with persisted readiness metrics, evidence-led exceptions,
-review-before-commit return handling and mobile navigation designed down to 390 px. See
-`docs/design/design-directions.md` and `docs/design/implementation-validation.md` for the
-design decision and visual evidence.
 
-All people, companies, vehicles, bookings and documents are synthetic. The workflows, validation, integrations, audit logging and access boundaries are intended to be real.
+![Fleet Ops engineering overview](frontend/public/og-fleet-ops.svg)
 
-## Demo
+## The 90-second tour
 
-The demo presents itself as **Northstar Mobility**, a fictitious Belgian camper/van
+1. Open **Highlights** from the login screen.
-rental company — the login screen, a permanent "Synthetische demo" indicator, an in-app
+2. Follow a vehicle return from review to atomic commit, quality issue, outbox and correlated audit trace.
-guided tour (Demo Guide), a curated `/scenarios` overview, and an "Over deze demo" page
+3. Compare and merge a duplicate customer with explicit human confirmation.
-all make the fictional context, synthetic-data status, and real-vs-simulated boundaries
+4. Ask the Knowledge Hub a damage question and inspect its cited procedure evidence.
-explicit without any verbal explanation. See `docs/demo-release/` for the full demo
+5. Open **Engineering** for the architecture, reliability guarantees, test evidence and honest scope boundary.
-concept, the five named scenarios, the seed/date-anchoring strategy, the guided-tour
-design, and the operational runbook (5-minute and 10-minute demo flows, reset, redeploy,
-rollback).
 
-## Scope
+## What makes it more than a mock-up
 
-The PoC implements:
+- **Transactional operations:** a return writes the inspection, vehicle/booking state, audit events and outbox record atomically. n8n downtime never rolls back the local business transaction.
+- **Explainable data quality:** five persisted rule types, SLA deadlines, assignment, bulk queue controls and bounded resolution flows—not decorative warning cards.
+- **Grounded knowledge:** the live deployment uses RAGcore; insufficient or unavailable evidence produces no invented answer. Citations and provider provenance remain inspectable.
+- **Safe AI exposure:** four tenant-bound, service-authenticated, read-only Fleet Ops tools are published through ITWorx MCP Hub and audited with correlation IDs.
+- **Operational reliability:** bounded retries, delivery leases, health/readiness, Prometheus metrics, Grafana, scheduled verified backups and graceful external-dependency degradation.
+- **Real product ergonomics:** nl-BE, en-GB and fr-BE; responsive from 360 px; keyboard-accessible navigation; role-aware global search; route-level lazy loading; server-enforced permissions.
 
-- operations dashboard with a truthful aggregate n8n/MCP integration-status card;
+## Architecture
-- vehicle and booking views with working search, filters and pagination;
-- server-backed session lifecycle (refresh-safe, central 401 handling);
-- a role matrix enforced server-side and mirrored in the UI (see
-  `docs/12-security-and-audit.md`);
-- vehicle return capture → authoritative server-evaluated review → commit → result;
-- five deterministic data-quality checks, each with a bounded resolution flow, plus a
-  manual scan action;
-- human review and customer merge;
-- audit trail with human-readable before/after evidence and safe entity links;
-- role-aware global search across vehicles, bookings and (Operations Manager) issues;
-- safe, confirmed demo reset;
-- RAGcore-backed knowledge assistant with citations;
-- two n8n workflows: return processing, and a scheduled data-quality scan with
-  crash-recoverable outbox delivery leases;
-- four read-only MCP tools through ITWorx MCP Hub;
-- deterministic demo reset and five-minute showcase.
 
-It is not an ERP, CRM, accounting package, public booking site, payment system or autonomous agent.
+```mermaid
+flowchart LR
+    UI["React + TypeScript\nresponsive operations UI"] -->|session cookie| API["FastAPI\nbusiness rules + RBAC"]
+    API --> DB[(PostgreSQL)]
+    API -->|grounded retrieval| RAG[RAGcore]
+    DB --> OUT["Transactional outbox"]
+    OUT -->|bounded retry| N8N["Existing central n8n"]
+    N8N -->|authenticated callback| API
+    HUB["ITWorx MCP Hub"] -->|4 read-only tools| API
+```
 
-## Integration status
+Fleet Ops owns operational truth. RAGcore owns retrieval, n8n performs post-commit orchestration, and MCP Hub owns tool transport/publication. Neither RAGcore nor MCP Hub accesses the Fleet Ops database directly. See [the as-built architecture](artifacts/evidence/architecture.md).
 
-- **n8n**: fully implemented and verified against a real n8n instance, including
+## Demonstrable scope
-  degraded mode (n8n stopped mid-flow → return still commits, event stays `pending`
-  with backoff, self-heals once n8n returns), the failed-delivery manual-retry path,
-  stale-delivery-lease recovery after a simulated crash, and a second (scheduled
-  quality-scan) workflow live-verified end to end against a real n8n instance.
-  `GET /api/v1/integrations/status` reports a truthful aggregate state from outbox
-  delivery counts, not just the most recent event.
-- **RAGcore**: the demo `KnowledgeProvider` (deterministic TF-IDF extractive retrieval
-  over the local procedure documents) is what satisfies the knowledge-assistant
-  acceptance criteria and is fully verified. A `RAGcoreKnowledgeProvider` HTTP adapter is
-  implemented and unit-tested, including its unavailable-degradation path, but was never
-  exercised against a live RAGcore instance in this environment.
-- **ITWorx MCP Hub**: the four read-only provider endpoints are implemented, tested, and
-  directly `curl`-verified with correct auth enforcement and audit logging.
-  `MCP_HUB_REGISTRATION_ENABLED` is now actually wired into `Settings` (it was previously
-  declared in `.env.example` but silently dropped) and reported honestly by the
-  integration-status endpoint. No live Hub instance was reachable in this environment to
-  verify an actual Hub round trip.
 
-See `artifacts/functional-completion/final-summary.md` for the functional-completion
+- dashboard, vehicle fleet, booking lifecycle and controlled returns;
-audit evidence (supersedes the design-validation summary below for integration status),
+- data-quality queue, assignment, review, merge and resolution;
-and `artifacts/final-acceptance/summary.md` for the original M0–M7 acceptance evidence.
+- correlated human-readable audit history;
+- cited Knowledge Hub with honest provider state;
+- n8n delivery monitoring and manual retry;
+- user administration, privacy export/anonymisation and retention guards;
+- deterministic reset with 2 users, 180 customers, 50 vehicles, 254 bookings, 75 inspections, 40 maintenance records, 33 quality issues and 20 workflow runs.
 
-## Repository map
+This is deliberately not accounting, payments, a public reservation site, generic CRM, inventory, HR or an autonomous write agent.
 
-- `CLAUDE.md` — binding implementation rules.
+## Stack
-- `MASTER_BUILD_PROMPT.md` — prompt to start an autonomous Claude run.
-- `PROJECT_STATE.md` — short persistent project memory.
-- `docs/` — product, architecture, UX and acceptance specification.
-- `contracts/` — OpenAPI, event and MCP contracts.
-- `knowledge/` — fictitious source documents for the MobilityOps RAGcore workspace.
-- `seed/` — deterministic synthetic dataset and generator.
-- `n8n/` — importable workflow definitions.
-- `backend/` — FastAPI/SQLAlchemy/Alembic API.
-- `frontend/` — React/TypeScript/Vite web app, including the Playwright end-to-end suite (`frontend/e2e/`).
-- `artifacts/evidence/` — final acceptance evidence (screenshots, architecture, `final-summary.md`).
-- `artifacts/design-validation/` — baseline audit, Stitch direction references and implemented responsive captures.
-- `docs/functional-completion/` — the functional-completion audit and pre-work server baseline.
-- `artifacts/functional-completion/` — functional-completion acceptance evidence.
-- `docs/demo-release/` — demo concept, scenarios, seed/date-anchoring strategy, guided
-  tour, and runbook.
-- `artifacts/demo-release/` — demo-productization acceptance evidence.
 
-## Quickstart
+React, TypeScript, Vite, FastAPI, SQLAlchemy 2, PostgreSQL, Alembic, n8n, RAGcore, ITWorx MCP Hub, Docker Compose, Prometheus, Grafana and Playwright.
+
+## Run locally
 
 ```bash
 cp .env.example .env
 make demo
 ```
 
-This builds and starts the full stack (migrations run automatically) and loads the
-deterministic demo dataset. See `docs/17-runbook.md` for the one-time n8n workflow setup
-required for the automation demo, and the full operational runbook.
-
-Endpoints:
-
 - Web: `http://localhost:1228`
-- API health: `http://localhost:8128/health`
+- API readiness: `http://localhost:8128/health/ready`
-- n8n: `http://localhost:5678`
+- Existing n8n server: point `N8N_WEBHOOK_URL` at its return-processing webhook (see `.env.example`). The bundled `n8n` service in `compose.yaml` is a local fallback only; production reuses the server's central n8n (`compose.unraid.yaml` disables the bundled one).
 
-All defaults are configurable via `.env` (see `.env.example`).
+The deterministic local knowledge provider supports clean-checkout acceptance without pretending to be the live RAGcore integration. Configuration is documented in `.env.example`; operations and recovery are in [docs/17-runbook.md](docs/17-runbook.md).
 
 ## Quality gates
 
 ```bash
-make test    # backend: pytest (127 tests)
+make test    # isolated PostgreSQL backend suite
-make lint    # backend: ruff + mypy (strict, zero errors)
+make lint    # Ruff + strict mypy
-make e2e     # frontend: Playwright end-to-end (56 tests, live stack required)
+make e2e     # complete Playwright browser acceptance
+cd frontend && npm run build
+python scripts/run-readonly-load-smoke.py  # while the demo stack is running
 ```
 
-Frontend build/typecheck: `cd frontend && npm run build` (`tsc -b && vite build`).
+Release-scoped results and production evidence are recorded in [artifacts/final-acceptance/summary.md](artifacts/final-acceptance/summary.md); older milestone evidence remains explicitly historical. [PROJECT_STATE.md](PROJECT_STATE.md) records the commands and exact deployment revision.
+
+## Repository map
+
+- `backend/` — FastAPI domain, API, migrations and tests
+- `frontend/` — React app and Playwright acceptance suite
+- `contracts/` — OpenAPI, event and MCP contracts
+- `knowledge/` — versioned fictional procedures
+- `n8n/` — importable workflow definitions for the existing server
+- `seed/` — deterministic synthetic dataset
+- `docs/` — architecture, security, UX, testing and runbooks
+- `artifacts/` — dated, release-scoped acceptance evidence and screenshots
+
+“MobilityOps” remains the repository/deployment identifier; **Fleet Ops** is the product name shown to users.

SECURITY.md

@@ -0,0 +1,56 @@
+# Security Policy
+
+## Supported versions
+
+| Version | Security support |
+|---|---|
+| Latest tagged PoC release and current `master` | Supported |
+| Older commits, branches and untagged deployments | Not supported |
+
+MobilityOps is a synthetic-data proof of concept, not a production identity,
+payments or public reservation platform. Security fixes target the current
+release line only.
+
+## Reporting a vulnerability
+
+Do not disclose suspected vulnerabilities through a public issue.
+
+Report them privately to `jens@itworx.tech` with:
+
+- the affected revision, endpoint or component;
+- reproduction steps and prerequisites;
+- the observed and expected behaviour;
+- the security impact;
+- a minimal proof of concept, without unnecessary personal or secret data.
+
+Receipt should be acknowledged within three business days. An initial
+assessment or request for additional evidence should follow within ten
+business days. Remediation timing depends on severity and reproducibility.
+
+## Scope
+
+In scope:
+
+- MobilityOps backend, frontend, container and deployment code;
+- authentication, authorization, tenant boundaries and audit integrity;
+- database, outbox, backup and restore behaviour;
+- MobilityOps-owned n8n workflow definitions;
+- RAGcore and MCP Hub integration boundaries implemented in this repository.
+
+Out of scope:
+
+- denial-of-service or destructive testing against the hosted demo;
+- social engineering, credential stuffing or physical attacks;
+- synthetic demo-data exposure without a security-boundary failure;
+- vulnerabilities solely inside RAGcore, ITWorx MCP Hub, n8n or another
+  third-party service. Report those to their respective owners.
+
+Do not access data beyond what is required to demonstrate the issue, modify
+shared infrastructure, interrupt other services or retain obtained secrets.
+
+## Coordinated disclosure
+
+Good-faith research that respects this policy and applicable law will be
+handled constructively. Allow a reasonable remediation period before public
+disclosure. Submitted reports and evidence are used only for investigation,
+remediation and verification.

START_HERE.md

@@ -18,7 +18,7 @@ Claude must use `PROJECT_STATE.md` as its compact memory between sessions. Do no
 - architecture and domain decisions;
 - API and event contracts;
 - realistic deterministic synthetic seed data;
-- ten fictitious procedures for RAGcore;
+- eleven fictitious procedures for RAGcore (nl-BE, en-GB, fr-BE);
 - an initial n8n workflow export;
 - MCP tool definitions for ITWorx MCP Hub;
 - a minimal bootable frontend/API scaffold;

artifacts/evidence/architecture.md

@@ -1,63 +1,68 @@
-# MobilityOps — as-built architecture
+# Fleet Ops — as-built architecture
 
 ```mermaid
 flowchart TB
-    subgraph Browser
+    UI["Fleet Ops Web\nReact + TypeScript + Vite"]
-        UI["MobilityOps Web<br/>React + TypeScript"]
-    end
 
-    subgraph MobilityOps["MobilityOps (this repo)"]
+    subgraph CORE["Fleet Ops — this repository"]
-        API["FastAPI backend<br/>/api/v1/*"]
+        API["FastAPI /api/v1\nRBAC + domain rules"]
-        DISPATCH["Outbox dispatcher<br/>background thread"]
+        OUT["Outbox dispatcher\nleases + bounded retry"]
         DB[(PostgreSQL)]
+        OBS["Prometheus metrics\nGrafana dashboards"]
         API --> DB
-        DISPATCH --> DB
+        OUT --> DB
+        API --> OBS
     end
 
-    subgraph External["External central services"]
+    subgraph EXT["Existing external platforms"]
-        N8N["n8n<br/>return-processing workflow"]
+        N8N["Central n8n\nsecondary orchestration"]
-        RAGDEMO["Demo KnowledgeProvider<br/>TF-IDF extractive, local files"]
+        RAG["RAGcore\ngrounded procedure retrieval"]
-        RAGCORE["RAGcore<br/>(adapter built, no live instance)"]
+        HUB["ITWorx MCP Hub\ntool transport + publication"]
-        HUB["ITWorx MCP Hub<br/>(endpoints built, no live instance)"]
     end
 
-    UI -->|session cookie| API
+    UI -->|secure session cookie| API
-    API -->|GroundedAnswer| RAGDEMO
+    API -->|tenant/workspace adapter| RAG
-    API -.->|configurable, unavailable-safe| RAGCORE
+    OUT -->|vehicle.returned.v1| N8N
-    DISPATCH -->|POST vehicle.returned.v1| N8N
+    N8N -->|service-authenticated callback| API
-    N8N -->|callback, X-Service-Token| API
+    HUB -->|service-authenticated read-only tools| API
-    HUB -.->|X-Service-Token, read-only| API
-
-    classDef unverified stroke-dasharray: 5 5;
-    class RAGCORE,HUB unverified;
 ```
 
-Dashed boxes/arrows are implemented and unit/contract-tested but were never exercised
+## Ownership and trust boundaries
-against a live instance in this environment (no reachable RAGcore or ITWorx MCP Hub).
-Solid boxes were verified end-to-end, including a real n8n instance.
 
-## Component responsibility (unchanged from `docs/03-architecture.md`)
+| Component | Owns | Explicitly does not own |
+|---|---|---|
+| Fleet Ops | vehicles, customers, bookings, inspections, quality issues, audit, permissions, outbox state | external workflow execution or procedure retrieval |
+| RAGcore | indexing/retrieval and grounded procedure evidence | Fleet Ops database or business state |
+| ITWorx MCP Hub | MCP transport, connector publication and central tool-call audit | Fleet Ops database or write actions |
+| n8n | post-commit workflow orchestration | critical business rules or the source-of-truth transaction |
 
-| Component | Owns |
+## End-to-end return trace
-|---|---|
-| MobilityOps | vehicles, customers, bookings, inspections, data-quality issues, audit, outbox/delivery state |
-| RAGcore | procedure retrieval and grounded answers (demo provider substitutes locally) |
-| ITWorx MCP Hub | MCP transport, tool publication, central tool-call audit |
-| n8n | post-commit secondary orchestration only — never the source of truth for vehicle state |
 
-## Reliability boundaries verified in this build
+```mermaid
+sequenceDiagram
+    actor Operator
+    participant Web
+    participant API
+    participant DB
+    participant n8n
+    Operator->>Web: Review and confirm return
+    Web->>API: POST return with idempotency key
+    API->>DB: Lock booking and validate invariants
+    API->>DB: Commit inspection, state, audit and outbox atomically
+    API-->>Web: Result + correlation ID
+    Web-->>Operator: Human result and full processing trace
+    API->>n8n: Deliver persisted outbox event
+    n8n->>API: Authenticated status callback
+    API->>DB: Persist delivery/audit evidence
+```
 
-1. **Return commits atomically with its outbox event** — `app/services/returns.py`, one
+## Verified reliability properties
-   transaction; verified by `test_concurrent_returns_only_one_succeeds` (real Postgres row
+
-   locking, not mocked).
+1. Concurrent returns serialize through PostgreSQL row locking; only one can commit.
-2. **Outbox delivery is at-least-once, idempotent by event ID** — verified live: the n8n
+2. Local return success is independent of n8n availability. Pending delivery remains persisted and retryable.
-   callback checks for an existing `AuditEvent` by event ID before recording a second time.
+3. Outbox delivery is at-least-once and idempotent by event ID, with crash-recoverable leases and bounded backoff.
-3. **RAGcore failure disables knowledge answers only** — `RAGcoreKnowledgeProvider` degrades
+4. RAGcore failure affects knowledge answers only. The UI reports unavailable/insufficient evidence and does not invent an answer.
-   to `unavailable`; the rest of the app is unaffected because the knowledge router is the
+5. MCP endpoints are a separate tenant-bound, client-identity-validated, read-only surface; every call is audited with a correlation ID.
-   only consumer.
+6. Browser authorization is enforced again on the API. Hiding a navigation item is never the security boundary.
-4. **MCP Hub failure does not affect the web application** — the four MCP provider
+
-   endpoints are a separate authenticated surface (`X-Service-Token`), invisible to the
+The live deployment has exercised all three external boundaries. Local clean-checkout acceptance can use the deterministic extractive knowledge provider while reporting that mode honestly.
-   browser-facing API/UI.
-5. **n8n failure leaves events pending with bounded retries** — verified live: a seeded
-   `failed` event, retried through the UI, was picked up by the background dispatcher and
-   delivered through the real n8n instance within one poll cycle.

artifacts/evidence/final-summary.md

@@ -1,177 +1,74 @@
-# MobilityOps — final acceptance evidence
+# Fleet Ops — final acceptance evidence
 
-## Commit
+## Accepted implementation
 
-Built on top of commit `c5b7e21f81694f0339ad31e3bf044db952d0fbe0` (M6, "implement ITWorx
+- Deployment source marker: current `master`; the application image was built from runtime commit `cb7edb0` and the following commits contain acceptance/evidence only.
-MCP Hub publication"). This evidence file and the rest of M7's polish are committed as
+- Live demo: `http://192.168.10.150:1236` — public passwordless synthetic demo.
-`M7: portfolio polish and final acceptance` — run `git log --oneline` for the exact hash.
+- Deployment: Unraid `/mnt/user/appdata/mobilityops`, Compose project `mobilityops`.
+- Database migration: `b913a72e8c14 (head)`.
+- Product name: Fleet Ops; MobilityOps remains the technical repository/deployment ID.
 
-## Exact commands (clean checkout)
+## Clean acceptance — 2026-08-10
 
-```bash
+```text
-git clone <repo> && cd MobilityOps
+Backend, isolated PostgreSQL: 241 passed, zero warnings
-cp .env.example .env
+Playwright, fresh local build: 152 passed (5.4 min)
-make demo          # docker compose up --build -d ; migrations run automatically ; seed --reset
+Playwright, live production:   152 passed (4.7 min)
+Ruff:                          clean
+Mypy:                          clean across 58 source files
+Frontend lint/build:           passed
+npm audit, full and prod:      0 vulnerabilities
+Alembic:                       b913a72e8c14 (head)
 ```
 
-One-time n8n setup (see `docs/17-runbook.md` for full detail — this cannot be scripted
+The browser suite covers both roles, protected routes, booking/return/maintenance/user/privacy workflows, five data-quality resolution types, correlated audit, retryable orchestration, grounded knowledge in three languages, the complete guided demo, the recruiter highlights, keyboard behaviour and responsive layouts from 360 to 1440 px.
-end-to-end because it requires a one-time owner account created through n8n's web UI):
 
-```bash
+The provider matrix explicitly covers both valid title contracts: the deterministic provider uses Markdown frontmatter titles, while production RAGcore returns curated localized presentation titles. Both must include the language-specific source fragment. The focused live matrix passed 3/3, followed by the complete 152/152 green production run.
-# open http://localhost:5678/setup in a browser, create any owner account
+
-make n8n-setup
+## Production hand-off state
+
+The final reset completed immediately before the verified Hub call at `2026-08-10T19:16Z` and restored:
+
+```text
+users 2 · customers 180 · vehicles 50 · bookings 254 · inspections 75
+maintenance 40 · data-quality issues 33 · workflow runs 20
+scenario_integrity.all_ready = true (5/5 scenarios)
 ```
 
-Verification:
+API readiness is `ready`, PostgreSQL is `up`, and API, web, database, scheduled backup, Prometheus and Grafana are running healthy/current. No traceback or critical error appears in post-deployment API/backup logs.
 
-```bash
+A pre-deployment custom-format backup was created and independently verified by SHA-256 plus `pg_restore --list`:
-docker compose run --rm api pytest -q       # 66 passed
-docker compose run --rm api ruff check .    # All checks passed
-cd frontend && npm run build                # clean tsc + vite build
-cd frontend && npx playwright test          # 1 passed (full 5-minute demo script)
-```
 
-## Test counts
+`/mnt/user/appdata/mobilityops/backups/postgres/mobilityops-20260810T185833Z.dump`
 
-- **Backend**: 66 tests passing (`pytest`), 0 skipped, 0 failed. Ruff clean. Coverage by
+## External integration evidence
-  area: seed determinism (2), auth/roles (4), dashboard (3), vehicles (4), bookings (3),
-  return workflow incl. real concurrent-submission test (9), data quality incl. S2/S4
-  scenarios (10), audit (2), n8n dispatcher incl. malformed-payload regression (6),
-  n8n callback idempotency (3), workflows/retry (4), knowledge incl. S6 scenario (7),
-  MCP provider endpoints (8), health (1).
-- **Frontend**: `npm run build` — clean TypeScript + Vite build, zero errors.
-- **End-to-end**: 1 Playwright test (`frontend/e2e/demo.spec.ts`) automating the full
-  documented 5-minute demo script (login → dashboard → S1 return → S2 merge → S6 knowledge
-  question → audit → 360px responsive check) — **passing** against the live stack.
 
-## Screenshots of the seven main pages
+- **Existing central n8n:** configured, dispatch enabled and operational. All four expected workflows have recent healthy heartbeats. The reset state contains 19 successful runs, zero pending, zero unexpected failures and one explicitly synthetic failed-retry scenario.
+- **RAGcore:** reachable and ready for tenant `northstar-mobility-demo`, workspace `mobilityops`, collection `internal-procedures`. Exact identity lookups independently confirm 11/11 active published documents in each of NL/EN/FR; the latest n8n report separately records 33 synchronized documents and zero failures. The canonical parsed-artifact hash is not misrepresented as a raw-source hash.
+- **ITWorx MCP Hub:** registration is enabled, the Hub is reachable and Fleet Ops is **operational**. Three real `mobilityops.operations.summary` verification calls are audited under the Hub's tenant-bound client identity; the final short-lived bearer produced HTTP 200/JSON and HTTP 401 after immediate revocation. All exposed Fleet Ops tools remain read-only.
 
-Captured live against the deterministic seed (`artifacts/evidence/screenshots/`,
+## Recruiter and visual acceptance
-via `frontend/e2e/_capture-screenshots.spec.ts`):
 
-| # | Page | File |
+The public entry now offers a direct **90-second Highlights** route. It links three product actions to their engineering proof, while the **Engineering** workspace explains domain ownership, commit-before-orchestration, citation-bound AI, security and explicit scope.
-|---|---|---|
-| 1 | Login | `1-login.png` |
-| 2 | Dashboard | `2-dashboard.png` |
-| 3 | Vehicles | `3-vehicles.png` |
-| 4 | Bookings | `4-bookings.png` |
-| 5 | Data Quality | `5-data-quality.png` |
-| 6 | Knowledge (grounded S6 answer) | `6-knowledge.png` |
-| 7 | Automation | `7-automation.png` |
-| — | Audit (bonus, 8th nav item) | `8-audit.png` |
-| — | Dashboard at 360px (responsive proof) | `9-mobile-dashboard.png` |
 
-## RAGcore evidence
+Interactive live inspection confirmed:
 
-**Success (demo provider, the one actually satisfying acceptance in this environment)** —
+- no horizontal overflow on Highlights or Engineering at desktop; automated coverage confirms 360/390 px mobile layouts;
-S6 question against the real `/api/v1/knowledge/questions` endpoint:
+- clear hierarchy, consistent primary actions and readable proof cards;
+- compact remaining-attention action (count badge, two-line label, small directional icon) without the former stretched empty panel;
+- truthful live n8n/RAGcore/MCP labels after reset, including 11 verified RAG documents and the real Hub client/tool evidence;
+- no dead links, placeholder numbers or unexplained raw data-quality references;
+- complete synthetic-data disclosure on login and in the persistent shell.
 
-```json
+Current evidence images are in `artifacts/evidence/screenshots/`:
-{
-  "answer": "Per \"Vehicle return procedure\" (v2.0), section \"1. Register the return\": Open the active booking and record the ending odometer, fuel level, cleanliness, visible damage, technical warnings and relevant notes.",
-  "evidence_state": "grounded",
-  "sources": [
-    {"document_id": "vehicle-return-procedure", "title": "Vehicle return procedure", "version": "2.0", "section": "1. Register the return", "excerpt": "..."},
-    {"document_id": "vehicle-return-procedure", "title": "Vehicle return procedure", "version": "2.0", "section": "3. Determine next state", "excerpt": "..."},
-    {"document_id": "damage-procedure", "title": "Damage handling procedure", "version": "1.3", "section": "1. Immediate actions", "excerpt": "..."}
-  ],
-  "provider": "demo",
-  "correlation_id": "b50094b7-1c84-4e39-9055-1dc03e8fd1f8"
-}
-```
 
-**Unavailable (RAGcore adapter, live-demonstrated against an unreachable host)** —
+- `1-login.png`
-`KNOWLEDGE_PROVIDER=ragcore`, `RAGCORE_BASE_URL=http://ragcore-not-reachable:9999`:
+- `2-highlights.png`
+- `3-engineering-story.png`
+- `4-dashboard.png`
+- `5-knowledge-evidence.png`
+- `6-highlights-mobile.png`
 
-```
+## Deliberate boundary
-health: {'provider': 'ragcore', 'available': False, 'detail': 'RAGcore unavailable: ConnectError: ...', 'document_count': 0}
-ask:    {'answer': '', 'evidence_state': 'unavailable', 'sources': [], 'provider': 'ragcore', 'correlation_id': 'demo-correlation'}
-```
 
-No live RAGcore instance was reachable in this environment, so the adapter's actual
+This is a completed, production-shaped public demo—not claimed customer adoption and not a general ERP. Accounting, payments, public reservations, CRM, inventory, HR, a second RAG stack, a separate MCP server and autonomous write agents remain intentionally excluded. A real personal-data rollout would additionally require the adopting organisation's identity provider, retention approvals, secrets lifecycle, alert ownership and disaster-recovery governance.
-request/response contract against a real RAGcore is unverified beyond this
-degrade-safely behavior — see `contracts/ragcore-contract-assumptions.md` and
-`PROJECT_STATE.md`'s M5 notes.
-
-## n8n evidence
-
-**Success** — a real return registered on `BK-DEMO-RETURN`, delivered through the actual
-n8n instance (not mocked), confirmed via `GET /api/v1/workflows`:
-
-```json
-{"event_id": "aa5dfeee-90ca-452a-bdd1-0a0b6d3dd63f", "event_type": "vehicle.returned.v1", "aggregate_ref": "BK-DEMO-RETURN", "status": "succeeded", "attempts": 2, "last_error": null}
-```
-
-(`attempts: 2` because the first delivery attempt landed while n8n was mid-restart from
-the one-time workflow-activation step — the dispatcher's backoff-and-retry handled it
-without any manual intervention, which is itself evidence of the retry behavior working.)
-
-**Retry (S5 scenario)** — seeded `BK-H-0020` (event `00000000-...-0020`), initially
-`failed` after 3 attempts with `"Synthetic connection timeout to n8n"`:
-
-1. Before: `{"status": "failed", "attempts": 3, "last_error": "Synthetic connection timeout to n8n"}`
-2. Operations Manager clicks Retry on `/automation`.
-3. Within one ~3s dispatcher poll cycle, delivered through the live n8n instance.
-4. After: `{"status": "succeeded", "attempts": 4, "last_error": null}`
-
-## MCP tool sample calls
-
-All four provider endpoints, authenticated with `X-Service-Token`:
-
-```
-$ curl -H "X-Service-Token: <token>" http://localhost:8128/api/v1/integrations/mcp/operations-summary
-{"tenant":"northstar-mobility-demo","metrics":{"available":21,"rented":11,"cleaning":6,"maintenance":5,"blocked":7,"open_quality_issues":22,"pending_or_failed_workflows":1}}
-
-$ curl -H "X-Service-Token: <token>" "http://localhost:8128/api/v1/integrations/mcp/attention-vehicles?minimum_severity=high&limit=3"
-[{"vehicle_ref":"MO-016","severity":"high","rule_type":"booking_overlap",...},
- {"vehicle_ref":"MO-016","severity":"high","rule_type":"vehicle_status_conflict",...},
- {"vehicle_ref":"MO-031","severity":"high","rule_type":"missing_required_field",...}]
-
-$ curl -H "X-Service-Token: <token>" http://localhost:8128/api/v1/integrations/mcp/vehicles/MO-016
-{"public_ref":"MO-016","make":"Hymer","model":"Exsis","model_year":2021,"location":"Geel","operational_status":"available","odometer_km":30497,"next_service_km":40000,"open_quality_issue_count":2,"current_booking_ref":null}
-
-$ curl -H "X-Service-Token: <token>" -X POST -d '{"question":"What must I do when a vehicle returns with damage?","max_sources":2}' http://localhost:8128/api/v1/integrations/mcp/search-knowledge
-{"answer":"Per \"Vehicle return procedure\" ...","evidence_state":"grounded","sources":[...2 items...],"provider":"demo",...}
-```
-
-Auth verified: missing header → `422`; wrong token → `401`. All four calls confirmed
-recorded in `GET /api/v1/audit?action=mcp_tool_request` with `actor_type: "service"`.
-
-No live ITWorx MCP Hub instance was reachable in this environment — these are direct
-calls to MobilityOps's own provider endpoints, not a Hub round trip.
-
-## Known PoC limitations
-
-- **RAGcore and ITWorx MCP Hub were never reachable in this build environment.** Both
-  integrations are implemented against best-effort/documented contracts and are
-  unit/contract-tested (including their failure-degradation paths), but neither was
-  verified against a real counterpart service. The demo `KnowledgeProvider` is what
-  actually satisfies the knowledge-assistant acceptance criteria here.
-- **n8n requires a one-time manual owner-account setup** per fresh environment
-  (`docker compose down -v` wipes it) — this is a property of the n8n 2.x image itself
-  (`N8N_BASIC_AUTH_ACTIVE` no longer gates the UI), not something MobilityOps can bypass.
-  Documented precisely in `docs/17-runbook.md`; the workflow import/activation itself
-  *is* scripted (`make n8n-setup`).
-- **Inspection public refs are a simple `count+1` sequence**, not gap-safe under true
-  concurrent writers — acceptable for this single-tenant demo, would need a DB sequence
-  for a multi-writer production system.
-- **The five data-quality rules use simplified idempotency** — `(rule_type, entity_type,
-  entity_id)` while open, rather than the doc's literal evidence-fingerprint scheme — see
-  `PROJECT_STATE.md`'s M3 notes for the reasoning (the fingerprint scheme would have let
-  the scan double-report issues already present in the seeded CSV).
-- **No production authentication** — demo login is an HMAC-signed session cookie tied to
-  two fixed seeded users, appropriate for a PoC, not a real identity provider.
-
-## Portfolio wording (truthful)
-
-MobilityOps is a working proof of concept, not a production system and not deployed for
-any real company. All customers, vehicles, bookings, and documents are synthetic
-(deterministically generated). The application logic it demonstrates is real: a
-transactional vehicle-return workflow with idempotency and concurrency control tested
-against real concurrent database transactions; five explainable, deterministic
-data-quality rules with a working customer-merge UI; a background outbox dispatcher
-verified end-to-end against a real n8n instance including failure/retry; a
-TF-IDF-weighted extractive knowledge assistant that never fabricates answers; and four
-read-only, audited, service-authenticated integration endpoints. RAGcore and the ITWorx
-MCP Hub integrations are implemented and tested in isolation but were not verified
-against live instances of those systems in this environment.

artifacts/final-acceptance/summary.md

@@ -1,265 +1,77 @@
-# MobilityOps — final acceptance audit summary
+# Fleet Ops release acceptance
 
-This audit was run after M0–M7 had already been implemented and committed, specifically
+This file is release-scoped evidence, not a timeless claim. Older evidence under
-to independently re-verify the finished system end to end rather than trust the
+`artifacts/evidence/` is historical. Exact commands and production revisions are recorded
-milestone-by-milestone build log. It found and fixed one real category of defect
+in `PROJECT_STATE.md`.
-(`mypy` had never been run across the whole build) and confirmed everything else — every
-user journey, every button/filter/form, both external-dependency degraded modes, secret
-hygiene, and the clean-checkout path — works as documented.
 
-## Final commit
+## 2026-08-21 release candidate
 
-This audit's fixes are committed as the commit immediately following
+- Backend: **271/271** tests passed against an isolated clean PostgreSQL database.
-`108b5d04fc6f7c5ff9c47009032d6469df29cf3c` ("M7: portfolio polish and final acceptance").
+- Browser acceptance: **155/155** Chromium tests passed in 6.0 minutes.
-Run `git log -1 --format="%H %s"` for the exact hash.
+- Live-safe browser canary: **4/4** passed across Chromium and Firefox against the local
+  deployed stack; unlike the acceptance suite, it never resets or mutates demo records.
+- Accessibility: the principal login, dashboard, data-quality, knowledge, automation and
+  audit routes have no automated critical/serious WCAG 2 A/AA/2.1 AA violations.
+- Frontend: TypeScript, ESLint, production build, dependency audit and per-asset JS/CSS
+  budgets passed; committed visual baselines cover the public entry and engineering story.
+- Contracts: committed OpenAPI, event schema, MCP tools and all five n8n definitions match
+  their code/manifest sources.
+- Recovery: a custom-format PostgreSQL dump was restored into a disposable database; the
+  Alembic revision and non-zero canonical table counts matched the source database.
+- Operations: Prometheus/Alertmanager configuration validation passed, including the
+  watchdog and authenticated n8n receiver route.
 
-## Exact commands executed
+## 2026-08-21 production verification
 
-Clean-checkout drill (run twice during this audit, most recently against fully wiped
+- Immutable application revision `95c91797fa2c599443d69d9c96d83a85ee0711f7` was promoted
-Docker volumes):
+  from a checksum-verified source archive after a fresh production backup.
+- Source revision and both OCI revision labels matched. Public readiness was green,
+  Alembic was at head, all health-gated services were healthy and persisted demo data was
+  retained without a deployment reset.
+- Trivy found zero fixed HIGH/CRITICAL vulnerabilities in each exact production image.
+- Prometheus successfully scraped the bearer-protected API target, Alertmanager carried
+  the active delivery watchdog, and the authenticated n8n alert receiver remained active.
+- The final non-destructive HTTPS canary passed **4/4** across Chromium and Firefox,
+  including the core operator routes and a grounded answer from the real knowledge stack.
 
-```bash
+## 2026-08-21 resilience upgrade verification
-git status                              # working tree clean before starting
-docker compose down -v                  # wipe all volumes — genuinely clean state
-cp .env.example .env
-docker compose up --build -d            # migrations run automatically (backend/entrypoint.sh)
-docker compose exec api python -m app.cli seed --reset
-docker compose run --rm api pytest -q
-docker compose run --rm api ruff check .
-docker compose run --rm api mypy app
-cd frontend && npm run build
-cd frontend && npx playwright test
-```
 
-n8n one-time setup (owner account via browser at `http://localhost:5678/setup`, then):
+- Immutable revision `00191e9b54ee6b961648a6e02abbb3a57957dba0` was promoted from a
+  checksum-verified archive after a verified production dump. A stable gateway now routes
+  to two revision-specific API and two web replicas; stateful services are no longer
+  restarted by routine application releases.
+- Moving host port 1236 from the legacy web container to the gateway was a one-time
+  migration hand-off and produced 14 failures across 1,200 rapid probes. Future releases
+  do not move that port; their acceptance gate is the zero-error versioned gateway reload.
+- The subsequent M50 release exercised that steady-state path: the gateway remained online,
+  atomically switched revision-specific API/web aliases and sustained **700/700** external
+  readiness probes without interruption. Post-promotion Chromium/Firefox acceptance passed
+  **4/4** and 360 concurrent authenticated reads had zero errors at p95 **116.4 ms**.
+- The versioned gateway switch sustained **300/300** local rollout probes without an error.
+  Production's non-destructive Chromium/Firefox canary passed **4/4**, and 360 authenticated
+  concurrent reads returned zero errors at p95 **137.2 ms**.
+- PostgreSQL, backup, Prometheus, Alertmanager, Grafana and the gateway all reported healthy;
+  Alembic was at head, the protected Prometheus target was present, and backup plus real
+  restore-drill evidence remained current.
+- Trivy 0.74 found zero fixed HIGH/CRITICAL vulnerabilities in the exact production API,
+  web and gateway images. The separately built rclone/PostgreSQL backup-tools image is also
+  clean after rebuilding rclone 1.75.0 with Go 1.26.6.
+- The OneDrive worker is deployed as an opt-in profile but is not represented as active:
+  it requires the owner's one-time interactive Microsoft OAuth authorization. Until that
+  happens, verified local backups remain the active recovery source.
 
-```bash
+## Evidence boundary
-docker compose exec n8n n8n import:workflow --input=//imports/mobilityops-return-processing.json
-docker compose exec n8n n8n publish:workflow --id=mobilityops-return-processing
-docker compose restart n8n
-```
 
-Degraded-mode drills:
+The complete local suite uses the deterministic provider and an isolated database so it is
+repeatable and safely destructive. Production verification is deliberately smaller and
+non-destructive; it verifies the real RAGcore/MCP/n8n health surfaces without resetting the
+shared demo. A successful local result is never presented as proof that an external service
+was live. The production subsection is added only after the exact committed release is
+deployed and observed.
 
-```bash
+## Remaining product boundary
-docker compose stop n8n     # then register a return via the API — commits, event stays pending
-docker compose start n8n    # dispatcher self-heals, no manual intervention
-docker compose run --rm -e KNOWLEDGE_PROVIDER=ragcore -e RAGCORE_BASE_URL=http://ragcore-not-reachable:9999 \
-  api python -c "from app.services.knowledge import get_knowledge_provider; ..."
-```
 
-## Test and validation results
+Fleet Ops remains a synthetic single-tenant PoC. It is not a production identity provider,
-
+payment system, accounting package or public reservation platform. External RAGcore, MCP
-| Check | Command | Result |
+Hub and n8n services remain independently operated dependencies and are accessed only
-|---|---|---|
+through their documented adapters.
-| Backend unit/integration tests | `docker compose run --rm api pytest -q` | **66 passed**, 0 failed, 0 skipped |
-| Backend lint | `docker compose run --rm api ruff check .` | **All checks passed** |
-| Backend type check | `docker compose run --rm api mypy app` | **Success: no issues found in 44 source files** (found and fixed 43 pre-existing errors this audit — see below) |
-| Frontend build + typecheck | `cd frontend && npm run build` | Clean (`tsc -b && vite build`, zero errors) |
-| End-to-end (Playwright) | `cd frontend && npx playwright test` | **12 passed** (`demo.spec.ts` — full 9-step demo script; `interactive-elements.spec.ts` — 11 tests covering every nav item, filter, tab, and role boundary) |
-| Clean-checkout migrations | `docker compose down -v && docker compose up --build -d` | 11 tables created automatically, `alembic current` → `e7b08389f47f (head)`, zero manual step |
-| Deterministic seed | `docker compose exec api python -m app.cli seed --reset` | `users:2 customers:180 vehicles:50 bookings:246 inspections:75 maintenance:40 data_quality_issues:26 workflow_runs:20` — identical across every reseed this session |
-| Secret scan | `git ls-files \| grep -x .env`; `git log --all -p -- '*.env'`; history grep for AWS/private-key/`sk-` patterns | No `.env` ever committed; no secrets found in history |
-
-### mypy defects found and fixed (the one real gap this audit uncovered)
-
-`mypy` is a declared dev dependency (`backend/pyproject.toml`) but was never added to any
-milestone's validation loop — only `ruff` was run throughout M0–M7. Running it cold
-surfaced 43 errors across 10 files. All were triaged and fixed (not suppressed):
-
-- **Two genuine defensive-programming gaps**, not just type-annotation issues:
-  - `app/services/returns.py`: the vehicle lookup after acquiring the row lock had no
-    `None` guard; a dangling FK would have crashed with an unhandled 500 instead of a
-    clean `404 VEHICLE_NOT_FOUND`. Fixed.
-  - `app/api/routers/bookings.py`: same pattern in `get_booking` for the customer/vehicle
-    lookups — now returns a clean `500` with a message instead of an `AttributeError`.
-- `app/api/deps.py` / `app/api/routers/demo.py`: `CurrentUser.role` is validated by
-  Pydantic at runtime already, but `get_current_user` now explicitly checks role
-  membership before construction, turning a would-be unhandled `ValidationError` into a
-  clean `401` for a corrupted/tampered session cookie.
-- Two instances of reusing one variable name for both a `Vehicle` and a `Customer` across
-  branches (`dashboard.py`, `data_quality.py`) — renamed for clarity, not just to satisfy
-  mypy.
-- `Booking.__table__.update()` / `Customer.__table__.update()` switched to the idiomatic
-  `sqlalchemy.update(Model)` construct (also fixes the type error).
-- Remainder: deprecated `conint()` → `Annotated[int, Field(...)]`, a `Sequence` vs `list`
-  `.sort()` call, an `assert`-guarded None-narrow after a `WHERE ... IS NOT NULL` filter
-  mypy can't see through, and a couple of narrowly-scoped `# type: ignore[...]` comments
-  for known SQLAlchemy stub gaps (`Result.rowcount`).
-
-`make lint` now runs `ruff check .` **and** `mypy app`.
-
-## Application URLs and ports
-
-| Service | URL | Notes |
-|---|---|---|
-| Web (React SPA) | `http://localhost:1228` | nginx-served static build |
-| API | `http://localhost:8128` | FastAPI, `/health` for liveness |
-| API docs | `http://localhost:8128/docs` | auto-generated OpenAPI/Swagger UI |
-| n8n | `http://localhost:5678` | requires one-time owner setup, see below |
-| PostgreSQL | `localhost:5432` (container-internal only, no host port published) | |
-
-## Demo users and access method
-
-No passwords. Two demo-role buttons on `http://localhost:1228/login`:
-
-- **Open as Operations Manager** → `USR-OPS`, "Amelie De Ridder". Full access: dashboard,
-  data-quality resolution/merge, automation retry, demo reset, MCP/service-token routes
-  are separate (not user-facing).
-- **Open as Rental Employee** → `USR-EMP`, "Karim Boujaddaine". Can register returns and
-  browse vehicles/bookings/knowledge; Automation page is visible but shows a
-  role-restricted message instead of the delivery table (enforced both in the UI and by
-  the backend's `require_operations_manager` dependency — verified by
-  `test_retry_requires_operations_manager` and the e2e role-restriction test).
-
-Session is an HMAC-signed, `HttpOnly` cookie (`app/core/security.py`) — a demo mechanism,
-not a real identity provider (documented as a known limitation).
-
-## Implemented functionality
-
-- Operations dashboard with 100% database-backed metrics, attention items linking to the
-  underlying data-quality issue, "today" departures/returns, and recent automation runs.
-- Vehicle and booking list/detail pages with working filters (status, attention-only) and
-  a tabbed vehicle detail view (overview/bookings/inspections/maintenance/quality).
-- Full transactional vehicle-return workflow: row-locked, idempotent by
-  `Idempotency-Key`, canonical-odometer regression handling (never silently lowers the
-  canonical value), vehicle status derivation, two audit events, and a schema-compliant
-  outbox event — verified against real concurrent submissions (1×201 + 2×409).
-- Invalid-mileage rejection: negative values and non-numeric input both correctly
-  rejected with `422` and a precise Pydantic validation message.
-- Data Quality Workbench: five deterministic rules (duplicate customer via TF-IDF-style
-  weighted signal scoring, missing required field, odometer regression, booking overlap,
-  vehicle status conflict), issue list/detail/defer/reject, and a two-column
-  duplicate-customer compare-and-merge UI with an inline (non-native-dialog) confirmation
-  step, transactional booking rewiring, and audit logging.
-- Full audit trail: every significant action (login, return, vehicle status change,
-  data-quality issue lifecycle, customer merge, workflow retry, demo reset, n8n
-  callback, MCP tool request, knowledge question) is recorded with actor, correlation ID,
-  and before/after state; filterable by action.
-- Knowledge Assistant: deterministic TF-IDF-weighted extractive retrieval over the 10
-  procedure documents — never generative, always cites real excerpts, and honestly
-  reports `insufficient`/`unavailable` states rather than fabricating an answer.
-- n8n automation: background outbox dispatcher (`FOR UPDATE SKIP LOCKED` claim,
-  exponential backoff, no DB transaction held during the HTTP call), a live-verified
-  round trip through an actual n8n workflow, manual retry for failed deliveries, and an
-  Automation page (Operations Manager only) showing all runs with filtering.
-- Four read-only, service-token-authenticated MCP Hub provider endpoints, each recording
-  its own service-request audit event, with zero write/mutation endpoints anywhere in
-  that namespace.
-- Responsive UI verified down to 360px width (nav wraps, tables become cards, metric
-  tiles reflow to a 2-column grid, no horizontal overflow) — both by an automated
-  Playwright viewport/overflow assertion and by a captured screenshot.
-
-## RAGcore integration status: implemented, not live-verified
-
-The active `KnowledgeProvider` in this environment is `DemoKnowledgeProvider` — fully
-implemented, fully tested, fully live-verified, and what actually satisfies the
-knowledge-assistant acceptance criteria. A `RAGcoreKnowledgeProvider` HTTP adapter also
-exists (`app/services/knowledge/ragcore.py`), targeting a best-effort contract inferred
-from `contracts/ragcore-contract-assumptions.md` (no live RAGcore API spec was available).
-Its **unavailable-degradation path is live-verified this audit**: pointed at an
-unreachable host, it returns `{"evidence_state": "unavailable", "answer": "", "sources":
-[]}` with no fabrication, exactly as required — but an actual successful round trip
-against a real RAGcore instance has never been performed, because no such instance was
-reachable in this environment.
-
-## MCP Hub integration status: implemented, not live-verified
-
-All four contracted read-only tools (`mobilityops_get_operations_summary`,
-`mobilityops_list_attention_vehicles`, `mobilityops_get_vehicle_details`,
-`mobilityops_search_knowledge`) are implemented as service-token-protected endpoints under
-`/api/v1/integrations/mcp/`, directly `curl`-verified this audit (auth enforcement,
-correct data shape, no write methods, service-request audit logging). No live ITWorx MCP
-Hub instance was reachable in this environment, so an actual Hub-mediated tool call was
-never performed — only direct calls to MobilityOps's own provider API.
-
-## n8n integration status: implemented and fully live-verified
-
-The only external integration with a real, running counterpart service available in this
-environment. Fully verified this audit, including both success and degraded paths:
-
-- **Success**: a real return registered via the API was delivered by the background
-  dispatcher to an actual n8n instance (owner account + imported/activated workflow),
-  which called back into MobilityOps and was recorded `succeeded`.
-- **Degraded mode**: `docker compose stop n8n`, then a return was registered — it
-  **committed successfully** (`201`, booking `status: returned` persisted) exactly as
-  required by the architecture's reliability boundary ("a return command and its outbox
-  event commit in one transaction" and "n8n failure leaves events pending with bounded
-  retries"). The outbox event stayed `pending` with two real `ConnectError`s logged and
-  exponential backoff.
-- **Self-healing**: restarting n8n required no manual intervention — the background
-  dispatcher picked the pending event back up on its next poll cycle and delivered it to
-  `succeeded` (5 total attempts across the outage).
-- **Manual retry (S5 scenario)**: a seeded `failed` delivery, retried from the Automation
-  page, moved to `pending` and was delivered to `succeeded` by the live dispatcher within
-  one poll cycle.
-
-## Known limitations
-
-- RAGcore and the ITWorx MCP Hub were never reachable in this build/audit environment;
-  both integrations are implemented and tested against inferred/documented contracts but
-  not verified against real instances of those systems (see above).
-- n8n requires a one-time, per-fresh-environment manual owner-account setup through its
-  own web UI (`http://localhost:5678/setup`) — a property of the n8n 2.x image itself
-  (`N8N_BASIC_AUTH_ACTIVE` no longer gates the UI), not something MobilityOps can bypass.
-  The workflow import/activation itself *is* scripted (`make n8n-setup`).
-- Demo authentication is an HMAC-signed session cookie tied to two fixed seeded users —
-  appropriate for a PoC, not a production identity provider.
-- Inspection public references are assigned via a simple `count + 1` sequence, not
-  gap-safe under true concurrent writers (acceptable for this single-tenant demo).
-- The five data-quality rules use a simplified idempotency key
-  (`rule_type, entity_type, entity_id` while open) rather than the spec's literal
-  evidence-fingerprint scheme — documented rationale in `PROJECT_STATE.md`'s M3 notes.
-- `npm audit` reports one residual moderate `esbuild`/Vite-8 dev-server-only advisory
-  (fixable only by a Vite major version bump) and one high `react-router` RSC-mode
-  advisory that does not apply to this app (it never uses React Router's RSC/SSR mode).
-
-## Clean deployment instructions
-
-```bash
-git clone <repo> && cd MobilityOps
-cp .env.example .env
-make demo                      # build, start, migrate (automatic), seed
-```
-
-One-time n8n setup (only needed for the automation demo path; everything else works
-without it):
-
-```bash
-# open http://localhost:5678/setup in a browser, create any owner account
-# (8+ chars, 1 number, 1 capital letter — no email verification required)
-make n8n-setup
-```
-
-Verify:
-
-```bash
-curl http://localhost:8128/health              # {"status":"ok",...}
-curl -o /dev/null -w "%{http_code}\n" http://localhost:1228/   # 200
-make test                                      # 66 backend tests
-make lint                                      # ruff + mypy, zero errors
-make e2e                                        # 12 Playwright tests (stack must be running)
-```
-
-Full detail, recovery expectations, and required operational checks: `docs/17-runbook.md`.
-
-## Five-minute demonstration flow
-
-1. Open `http://localhost:1228` → **Open as Operations Manager**.
-2. **Dashboard**: point out the metrics are live counts (available/rented/cleaning/
-   maintenance/blocked vehicles, open quality issues, pending/failed workflows), and the
-   Attention Required list linking straight to the underlying issues.
-3. **Vehicles → MO-024** → open the active booking `BK-DEMO-RETURN`, register a return
-   with an odometer reading below MO-024's canonical value → the result panel shows the
-   inspection, the derived vehicle status, the automatically-created data-quality issue,
-   and the queued automation event — canonical odometer is confirmed unchanged.
-4. **Data Quality → DQ-DEMO-DUPLICATE**: the two-column CUS-0012/CUS-0178 comparison,
-   merge with the inline confirmation step, issue flips to `resolved`.
-5. **Knowledge**: ask "What must I do when a vehicle returns with damage?" → grounded
-   answer citing both the return and damage-handling procedures with real excerpts.
-6. **Automation**: filter to `failed`, retry the seeded delivery, watch it succeed within
-   a few seconds via the live n8n instance.
-7. **Audit**: filter by `return_registered` or `customer_merged` to show every action from
-   this walkthrough is recorded with actor, timestamp, and correlation ID.
-8. Resize the browser to 360px width to show the responsive layout (nav wraps, tables
-   become cards) — or run `make e2e` and point at the passing responsive assertion.

artifacts/final-integrations/final-summary.md

@@ -0,0 +1,161 @@
+# Fleet Ops final integrations — evidence summary
+
+Session date: 2026-08-05. Branch `feat/fleet-ops-final-integrations`.
+
+## Repository state
+
+| Repo | Start | End | Branch | Notes |
+|---|---|---|---|---|
+| Fleet Ops (MobilityOps) | `3ebca9e` (from `feat/live-n8n-ragcore-integration`) | `727c19a` (+ e2e test fixes, uncommitted at write time) | `feat/fleet-ops-final-integrations`, pushed to `origin` | 3 commits: `34df66d`, `2ae2044`, `727c19a` |
+| RAGcore | `64a908a` | `64a908a` (+1 isolated commit `ce0ad56`) | `main` | Only a backlog handoff entry committed; no code changes (36-file concurrent-session collision — see below) |
+| ITWorx MCP Hub | not modified this session | — | `feature/wp240-final-acceptance` | Connector already live in production before this session started; not touched |
+
+## Deployed revisions
+
+- Fleet Ops: `http://192.168.10.150:1236`, redeployed twice this session (after Batches
+  1-3 and after Batch 4), `docker compose -p mobilityops -f compose.yaml -f
+  compose.unraid.yaml up --build -d db api web`, `.deploy/source-revision` = `727c19a...`.
+- RAGcore: `http://192.168.10.150:1237`, `ragcore-app-1`. No image redeploy — the two live
+  fixes (filesystem permissions, reranker model pull) were applied directly to the
+  running container/Ollama instance, not via a code deploy.
+- ITWorx MCP Hub: `http://192.168.10.150:1100` (Tower), unchanged, already live before
+  this session at commit `c4a0f6d` per the Hub's own state.
+
+## GUI polish (Batch 1)
+
+- Dashboard Attention Queue: curated severity mix (grouped "Handle now / Follow up
+  today / Review later"), replacing pure severity-sort that let `high` crowd out
+  everything else.
+- Today's Movements: seed data curated (`seed/bookings.csv`) so a fresh reset shows ≥2
+  departures and ≥2 returns; new `test_seed_today_movements_are_a_credible_mix` test.
+  Live-verified after a real demo reset: 2 returns + 2 departures shown.
+- About Demo: restructured into a compact grid with `<details>` progressive disclosure
+  for architecture/security/testing sections.
+- Duplicate Customer Merge: match/conflict counts shown, matching fields hidden by
+  default (toggle to reveal), compact preview of the merged record before confirmation.
+- Repo hygiene: removed a stray empty `backend;C` dir and an untracked 31MB zip export;
+  `.gitignore` now excludes future archive exports.
+- All four live-verified via browser against the deployed instance (see screenshots
+  taken during the session — not separately saved to disk).
+
+## n8n (Batch 2)
+
+- 4 canonical workflows confirmed live: Vehicle Return Orchestration, Scheduled Data
+  Quality Scan, RAGcore Procedure Sync, Workflow Error Handler.
+- Fixed genuinely invalid JSON in the committed `fleet-ops-vehicle-return.json` (a
+  missing `},` between two node objects — the file could not be parsed).
+- Workflow 3 (RAGcore Procedure Sync): confirmed 6 real nodes built and saved. Found and
+  fixed two real defects via the safe `n8n import:workflow` CLI path (not the REST API,
+  which caused a documented wipe incident in an earlier session): three body-parameter
+  expressions had a stray trailing `}}`, and `settings.errorWorkflow` was unset. Exported
+  the corrected definition to `n8n/workflows/fleet-ops-ragcore-procedure-sync.json`,
+  added to `MANIFEST.md` and `check_drift.py`.
+- **Not published** — the Schedule Trigger runs daily at midnight; activating it starts
+  real unattended production runs, deliberately left as a separate go-live decision.
+- No no-op/sync/error-handler live-execution smoke test was run this session beyond the
+  structural CLI-export verification above (workflow remains unpublished).
+
+## RAGcore (Batch 3)
+
+- **Root cause found and fixed, live, user-approved**: the "zero retrieval candidates"
+  bug was a filesystem permission bug (`/workspace/.state/models/embedding_profiles.json`
+  was `root:root` mode `600` on the host bind mount, unreadable by the app's actual
+  runtime uid 10001) — not authorization, not Qdrant, not embeddings, all independently
+  verified healthy first. Fixed via `chown`/`chmod`; re-verified in-process (5 real hits,
+  up from 0).
+- **Second, deeper gap found, not fixed**: the reranker adapter calls
+  `{ollama}/api/rerank`, a route this Ollama version (`0.32.5`) does not serve (404).
+  Pulled a working model (`xitao/bge-reranker-v2-m3:latest`, 1.2GB, approved) — did not
+  fix it, since the problem is the HTTP route, not the model. `/v1/answers` still returns
+  `not_answerable`/0 citations for real questions against real matching content.
+- User decision: leave `KNOWLEDGE_PROVIDER=demo`; hand the reranker fix off to RAGcore's
+  own backlog (`docs/ai/BACKLOG.yaml`, task `M8-01`, committed in that repo as `ce0ad56`
+  — the only commit made in RAGcore this session) rather than editing RAGcore code amid
+  its own 36-file concurrent-session collision.
+- Side effect: minting the live-verification credential rotated the existing "Fleet Ops
+  Knowledge Assistant (production)" service account's credential (2-active-credential cap
+  reached). A fresh credential must be issued before actually flipping the provider live.
+
+## MCP Hub (Batch 4)
+
+- Confirmed the Fleet Ops connector is already live in production on the Hub side
+  (Tower, commit `c4a0f6d`), with a real contract fix already applied there
+  (`vehicle.get`'s wire parameter normalized to camelCase `vehicleRef`).
+- Fixed two concrete gaps in Fleet Ops's own `search-knowledge` endpoint: no `locale`
+  field existed at all (now `nl-BE`/`en-GB`/`fr-BE`, wired to the knowledge provider's
+  existing `language` param), and the correlation ID was always freshly minted, ignoring
+  any inbound `X-Correlation-Id` header. Added `get_correlation_id`, applied to all four
+  MCP endpoints.
+- `MCP_HUB_BASE_URL` was dead config (declared, never read); wired it for a real,
+  bounded Hub-reachability health check instead of an unneeded self-registration push
+  (the Hub's own registration is catalog-driven).
+- Renamed Fleet Ops's own internal audit tool labels `mobilityops_*` → `fleet_ops_*`
+  (mirrored in `contracts/mcp-tools.json`, `mobilityops_*` kept as deprecated aliases).
+  The live Hub connector's own dotted tool namespace (`mobilityops.operations.summary`
+  etc.) is a separate, Hub-owned naming layer, deliberately not touched.
+- Automation page's MCP card now shows real evidence (last tool/client/count/timestamp)
+  instead of only the registration-enabled boolean.
+
+## AI Operations Brief (Batch 5)
+
+Real MCP-client-shaped run via the live ITWorx MCP Hub connector's own
+`MobilityOpsClient` class against production Fleet Ops. Full runbook and live output in
+`docs/final-integrations/ai-operations-brief-runbook.md`. Summary:
+
+- Real operations summary (21 available / 11 rented / 6 cleaning / 5 maintenance /
+  7 blocked; 23 open quality issues).
+- Real most-pressing vehicle identified (`MO-031`, missing operational inspection).
+- Real vehicle detail lookup.
+- Real grounded knowledge answer (English damage-handling question): 2 real citations,
+  `evidence_state: grounded`.
+- Dutch/French variants of the same question honestly returned `insufficient` (no
+  fabrication) — root cause: the live Hub connector doesn't yet send the new `locale`
+  field, a Hub-side follow-up, not silently worked around.
+- Correlation IDs verified end-to-end in Fleet Ops's own audit log
+  (`GET /api/v1/audit?action=mcp_tool_request`), matching the response payloads exactly.
+- No write actions performed at any point.
+
+## Testing per batch
+
+- Backend: **176 passed**, `ruff check .` clean, `mypy app` clean (50 source files) —
+  verified against a freshly rebuilt image after discovering mid-session that
+  `docker compose run --rm api` (no bind mount on the `api` service) silently tests a
+  stale image otherwise. One genuinely stale test assertion found and fixed as a result.
+- Frontend: `tsc -b && vite build` clean.
+- E2e (Playwright, against the live deployed instance,
+  `MOBILITYOPS_PUBLIC_URL=http://192.168.10.150:1236`): every spec file run this
+  session passed — `demo.spec.ts`, `interactive-elements.spec.ts` (26),
+  `responsive-i18n.spec.ts` + `demo-accessibility.spec.ts` + `guided-demo-full.spec.ts`
+  (28), `i18n-coverage.spec.ts` + `error-messages.spec.ts` + `clickable-rows.spec.ts` +
+  `demo-guide.spec.ts` + `demo-entry.spec.ts` + `demo-legibility.spec.ts` +
+  `fleet-ops-correction.spec.ts` + `ui-redesign.spec.ts` + `greeting.spec.ts` +
+  `greeting-live.spec.ts` (28, after fixing 2 pre-existing fragile locators unrelated to
+  this session's feature work — a `.data-table` ambiguity now that Automation has two
+  tables, and a `Technische details` toggle ambiguity for the same reason; plus one
+  pre-existing untranslated-loanword false positive in `i18n-coverage.spec.ts`).
+
+## Known limitations, stated plainly
+
+- `KNOWLEDGE_PROVIDER` is still `demo`, not `ragcore` — blocked on RAGcore's own
+  reranker gap (handed off, not fixed this session).
+- n8n workflow 3 is built and correct but not published (deliberate, separate decision).
+- The live MCP Hub connector doesn't yet send the new `locale` field, so
+  locale-aware knowledge search only works when called directly against Fleet Ops (as
+  proven by the backend tests), not yet through the live Hub connector as deployed.
+- No public-demo-readiness checklist, About Demo Guide "completed" end-state polish
+  (section 4E), or dashboard MCP "activity showcase after Demo Complete" gating were
+  built this session — the MCP evidence display exists on the Automation page
+  unconditionally rather than gated behind guided-demo completion.
+- No security-review pass was run separately this session (existing gates: ruff, mypy,
+  the repo's own auth/audit test coverage).
+
+## Rollback
+
+- Fleet Ops: prior working revision `0571a40` remains in `.deploy/` as
+  `source-0571a40.tar.gz` on the Unraid host; redeploy by re-extracting and re-running
+  the same `docker compose up --build -d` sequence with that archive.
+- RAGcore: `chown`/`chmod` change is trivially reversible (`chown 0:0` +
+  `chmod 600` on the same path) if needed, though there is no reason to revert a
+  permission fix. Ollama model pull (`xitao/bge-reranker-v2-m3:latest`) can be removed
+  with `ollama rm` if unwanted; it is inert until RAGcore's own code is changed to use it.
+- MCP Hub: not modified this session.

artifacts/fleet-ops-correction/final-summary.md

@@ -0,0 +1,284 @@
+# Fleet Ops correction and release — final evidence
+
+**Result: PASS**
+
+## Commits
+
+- Source branch / commit (verified pre-correction baseline): `master` @ `18344bc8b7a75a2f868bf15bf498fc030ac6c34c`
+- Fix branch: `fix/fleet-ops-i18n-status-flow`
+- Final fix-branch commit: `284b3c7` (merged content identical to `2e4fb43`, which carries the evidence-summary localization fix)
+- Main-before-merge: `18344bc8b7a75a2f868bf15bf498fc030ac6c34c` (confirmed unchanged via `git fetch` + `git rev-parse origin/master` immediately before merging — no unexpected commits landed on master while this branch was in progress)
+- Merge commit: `de0bdea84fea01b4501deb7099107bc753c2e6d7` (`git merge --no-ff fix/fleet-ops-i18n-status-flow -m "merge: complete Fleet Ops localization and status resolution"`, zero conflicts)
+- Final main commit: `de0bdea84fea01b4501deb7099107bc753c2e6d7`
+- Deployed commit: `de0bdea84fea01b4501deb7099107bc753c2e6d7` (`.deploy/source-revision` on Unraid)
+- Gitea main branch: `master` (confirmed via `git fetch origin && git rev-parse origin/master` matching local `master` after push)
+- Live URL: `http://192.168.10.150:1236`
+
+Fix-branch commit history: `6deb955`, `e6539d1`, `ac4b163`, `1fdd2b3`, `1e40775`, `a7ac5ed`, `7851e80`, `cda2c32`, `2e4fb43`, `284b3c7`.
+
+## What this correction fixed
+
+1. **Status-recommendation flow redesigned** (sections 8A–8F). The old single opaque
+   "calculate and apply recommended status" action is replaced by a single shared, pure
+   evaluator (`backend/app/services/vehicle_status.py::evaluate_vehicle_status`,
+   documented in `docs/fleet-ops-correction/vehicle-status-decision-table.md`) used
+   identically by the scanner, a non-mutating preview endpoint
+   (`POST /api/v1/data-quality/issues/{ref}/status-recommendation`), and a
+   transactional apply endpoint (`POST .../apply-recommended-status`) that locks the
+   row, recomputes facts, rejects a stale `recommendation_token`, refuses unsafe/manual-
+   review recommendations, and re-validates post-write before resolving the issue.
+   - Forbidden shortcuts eliminated: "maintenance + active booking" no longer
+     auto-recommends "rented" (being in maintenance is itself now a blocking fact);
+     "maintenance with nothing else wrong" no longer auto-clears to "available" (no
+     fact proves maintenance is actually finished — release stays a manual decision).
+   - Frontend: "Review recommendation" → a localized decision panel (current/
+     recommended status, why, evidence, consequences) → an exact "Change status to
+     <status>" confirm action → result, or a distinct "Manual review required"
+     state offering no generic apply button.
+2. **MO-016 order independence** (section 9). Order independence does not mean "same
+   final status regardless of order" — resolving the booking overlap first genuinely
+   removes the conflict, correctly leaving nothing to apply. What holds either way: the
+   recommendation always reflects real current facts (never a stale proxy), and nothing
+   unsafe is ever applied (never "rented"). Proven by a backend test explicitly scoped
+   to MO-016/DQ-DEMO-STATUS (the original version wasn't — `_first_open()` returned
+   whichever of ~14 open `vehicle_status_conflict` issues was most recent, not
+   necessarily MO-016's) and a browser-level Playwright test covering both orders.
+3. **"Fleet Ops" is a non-localizable brand constant** (`frontend/src/product.ts`,
+   backend `PRODUCT_NAME`), wired via `{{productName}}` interpolation everywhere the
+   brand appeared in locale prose. A permanent test fails the build if any locale file
+   ever defines the brand name or an `appName` key again.
+4. **Dynamic backend prose converted to message codes + params** (sections 5/6/10):
+   return status reasons, audit field/actor-type labels, automation `last_error` (new
+   `last_error_code` column, migration `799d8800e241`), search results (sections/
+   vehicles/bookings/issues), and — found live on Unraid — the data-quality evidence
+   summary. Raw technical text is demoted to a "Technical details" disclosure
+   everywhere.
+5. **Knowledge-base fixes**: the demo provider's tokenizer silently dropped accented
+   characters (`[a-z0-9]+` split "véhicule" into "v"+"hicule"), breaking French
+   retrieval broadly — fixed to include the Latin-1 accented range. Reweighted section
+   scoring so a body match (real substance) outranks a heading/title match (a shallow
+   structural hint) — the old weighting misranked the damage procedure behind an
+   unrelated document for the brief's exact validation question in all 3 languages.
+   Removed leftover "MobilityOps"/"PoC" mentions from 9 procedure documents.
+6. **Search, audit, automation, maintenance/inspections localized** (section 10):
+   backend returns stable codes + params only; the frontend localizes section labels,
+   vehicle summaries, booking/issue statuses, audit action/field/actor labels,
+   automation error explanations, and maintenance/inspection type labels.
+7. **i18n test suite strengthened** (section 11): key parity, brand invariant,
+   translation-quality (cross-locale identical-value detection), a hardcoded-JSX-text
+   static scan (had to anchor on backreferenced closing-tag names — a naive `>text<`
+   regex misread TypeScript generics as JSX), and a 3-language route matrix (every main
+   route, no console errors, correct `html[lang]`, real page headings).
+
+## Live-caught bug (the deployment validation earning its keep)
+
+Live validation on the freshly-deployed fix branch directly caught a real defect: every
+data-quality issue's top-of-page evidence summary was unconditionally showing raw,
+always-English text (e.g. *"vehicle marked available while reserved bookings
+conflict"*) in **all three languages**, because the frontend never finished the
+`evidence.signals` localization the backend had already been emitting (the backend code
+even had a comment describing the intended design that the frontend didn't implement).
+Fixed in commit `2e4fb43`:
+- `DataQualityIssueDetail.tsx` now renders `evidence.signals` through the operator's
+  locale as the primary evidence text.
+- The four `DQ-DEMO-*` seed rows that anchor the guided demo's scripted scenarios now
+  carry real, accurate signals computed at seed time (the duplicate-customer similarity
+  score is the actual `SequenceMatcher` ratio on the seeded names, not invented).
+- Rows with no structured signals fall back to raw text rather than showing a blank
+  summary; the one known filler placeholder gets its own localized rendering.
+- A regression test locks this in: the vehicle-status-conflict evidence summary must
+  show localized text and must never contain the specific raw English sentence that was
+  live-visible before the fix, in all 3 languages.
+
+Also found and fixed along the way: a frontend logic bug conflating "no conflict" with
+"manual review required" (both carry `safe_to_apply: false`), which showed a false
+"manual review required" panel for MO-016 after its booking overlap was resolved
+instead of the correct "no change needed" state (fixed in `1fdd2b3`).
+
+## Translation coverage
+
+- All three locale files (`nl-BE`, `en-GB`, `fr-BE`) define exactly the same key set
+  for every namespace (`i18n-coverage.spec.ts`, structural guarantee).
+- No locale file contains an empty string value.
+- No locale file defines the brand name or an `appName` key (brand-invariant test).
+- Cross-locale translation-quality check: for every string ≥8 characters of real prose,
+  nl-BE ≠ en-GB, fr-BE ≠ en-GB, fr-BE ≠ nl-BE, with a precise, audited allowlist for
+  genuine proper nouns/cognates (23 entries, each with a documented reason).
+- Hardcoded-JSX-text static scan: zero findings against the current codebase (verified
+  against both false positives — TypeScript generics — and a deliberately-injected-
+  then-reverted false negative).
+- 3-language route matrix: every main route (dashboard, vehicles, vehicle detail,
+  bookings, booking detail, data quality, issue detail, automation, knowledge, audit,
+  scenarios, about) opens cleanly in all 3 languages with no console errors, correct
+  `html[lang]`, and a real page heading.
+- **Remaining visible wrong-language text**: none found. The one gap that existed (the
+  data-quality evidence summary) was found live and fixed before merge.
+
+## Branding
+
+- Visible product name: **Fleet Ops**, exactly, in all 3 languages, everywhere (login,
+  topbar, footer "Fleet Ops Demo", document title, About page, Demo Guide, knowledge
+  base). Verified structurally (brand-invariant test) and live (branding test across
+  dashboard/vehicles/data-quality/audit/automation/knowledge pages in all 3 languages;
+  visual screenshots of the login screen in nl-BE and fr-BE).
+- Technical identifier retained (by design, per the brief): repository name, local
+  directory, package/module names, Compose project, deployment directory, database
+  name, and the `/health` endpoint's `service: "mobilityops-api"` field remain
+  "mobilityops" — none of these are visible UI text.
+- No visible "MobilityOps" or "PoC" anywhere in the UI or the demo knowledge base
+  (9 procedure documents cleaned up; regression test in `test_knowledge.py` scans every
+  procedure file for both strings).
+
+## Status-preview / apply / manual-review / MO-016 ordering
+
+- **Preview**: verified non-mutating — the issue's `status` stays `"open"` after
+  calling the preview endpoint and re-fetching it via a fresh request.
+- **Apply**: the confirm button names the exact target status ("Change status to
+  Blocked" / "Status wijzigen naar Geblokkeerd" / "Changer le statut vers Bloqué");
+  applying resolves the issue and updates the vehicle atomically.
+- **Manual review**: MO-024 (active rental + service-threshold reached, a genuine fact
+  contradiction) shows "Manual review required" with no generic apply button rendered
+  at all.
+- **Stale token**: simulated by resolving the underlying booking overlap after the
+  preview was fetched but before applying — the apply call is correctly rejected
+  (`RECOMMENDATION_STALE`), the UI shows the "situation has changed" message, and the
+  user must review again before a new apply is possible.
+- **MO-016 ordering**: both orders tested. Resolving the overlap first correctly leaves
+  nothing to apply (vehicle stays "available", genuinely correct). Resolving the status
+  conflict first safely blocks the vehicle; resolving the now-redundant overlap
+  afterwards does not disturb it. Neither order ever produces "rented".
+
+## Knowledge (per language)
+
+The brief's exact validation question, in each language, grounds on the damage
+procedure as the **primary** (not just top-3) source:
+- nl-BE: *"Wat moet ik doen wanneer een voertuig beschadigd terugkomt?"* → damage
+  procedure, Dutch source, Dutch excerpt.
+- en-GB: *"What should I do when a vehicle returns with damage?"* → damage procedure,
+  English source, English excerpt.
+- fr-BE: *"Que dois-je faire lorsqu'un véhicule revient endommagé ?"* → damage
+  procedure, French source, French excerpt.
+
+This required two real fixes: a tokenizer bug that silently dropped accented
+characters (breaking French retrieval broadly) and a scoring-weight rebalance (body
+matches now outrank heading/title matches).
+
+## Audit / automation
+
+- Audit: action labels localized (`workflow_retry` → "automatisering opnieuw
+  geprobeerd" / "automation retried" / "automatisation relancée", etc.), field names
+  localized (`operational_status` → "Operationele status" / "Operational status" /
+  "Statut opérationnel"), actor types localized, raw technical codes only inside
+  "Technical details". Verified live and via a dedicated Playwright test.
+- Automation: the seeded synthetic failure shows a localized primary explanation
+  ("De workflowdienst was tijdelijk niet bereikbaar…") with the raw technical message
+  ("Synthetic connection timeout to n8n") only under "Technical details". Verified live
+  and via a dedicated Playwright test.
+
+## Backend tests / lint / types
+
+- `pytest`: **151 passed**, 0 failed (clean checkout, local dev, and post-merge master
+  — run four times across this correction, always 151/151).
+- `ruff check .`: all checks passed, every run.
+- `mypy app` (strict): no issues found in 49 source files, every run.
+- Alembic: `alembic upgrade head` from empty database lands on `799d8800e241`
+  (the new `outbox_events.last_error_code` column); `downgrade -1` / `upgrade head`
+  round-trip verified.
+
+## Frontend build / Playwright
+
+- `npm ci`, `tsc -b`, `vite build`: clean, every run.
+- Full Playwright suite: **116 tests**, run repeatedly against the local dev stack, an
+  isolated clean-checkout stack, the live fix-branch deployment, and the live
+  post-merge master deployment — **116/116 passed** on the final master-deployment run
+  and on the final local run. A handful of transient, sequential-run-only flakes
+  occurred at various points across ~10 full-suite runs today (different test each
+  time, e.g. a pre-existing logout-timing race in `AuthContext.logout()` unrelated to
+  this branch); every single one was confirmed to pass cleanly in isolation.
+- Guided demo covered indirectly via `guided-demo-full.spec.ts`,
+  `demo-guide.spec.ts`, and the route matrix across all 3 languages — no dedicated
+  "run the guided tour end-to-end in French" script exists beyond what those specs plus
+  the branding/route-matrix tests already exercise, since the guided tour's steps route
+  through the same pages already covered per-language.
+
+## Clean-checkout drill
+
+Fresh `git clone --branch fix/fleet-ops-i18n-status-flow` of only committed files into
+an isolated Compose project (`cleancheckfleetops`, ports 8129/1229/5679 to avoid
+colliding with the working dev stack). From empty volumes: build → up → `alembic
+upgrade head` → `reset_and_seed` (50 vehicles / 180 customers / 246 bookings / 27
+data-quality issues / 20 workflow runs) → 151 backend tests + Ruff + mypy green →
+frontend build green → full Playwright suite green → final reset →
+`scenario_integrity.all_ready: true`. Isolated stack, containers, volumes, and images
+torn down afterward; working dev environment confirmed untouched.
+
+## Unraid deployment
+
+Deployed via `git archive` → `scp` → extract into `/mnt/user/appdata/mobilityops`
+(preserving `.env` and persistent volumes) → `.deploy/source-revision` → rebuild
+`api`+`web` → `alembic upgrade head` → reset/reseed. Done twice: once for the fix
+branch (caught the evidence-summary bug), once for the final merged master. Both times:
+containers healthy, no errors in `api`/`web` container logs, full Playwright suite
+green against the live server, `scenario_integrity.all_ready: true` after final reset.
+RAGcore and MCP Hub were not activated (the demo `KnowledgeProvider` — deterministic
+local retrieval — remains what's live, per the brief's constraint against activating
+unvalidated live integrations).
+
+## Responsive / accessibility
+
+- Breakpoint matrix (1440×1000, 1280×800, 1024×768, 768×1024, 430×932, 390×844,
+  360×800) × 3 languages: no horizontal overflow, localized headings visible
+  (`responsive-i18n.spec.ts`).
+- Status-recommendation panel: keyboard-only activation of "Review recommendation" and
+  "Change status to X" verified via focus assertions (not just click); reduced-motion
+  emulated during the flow; status never conveyed by colour alone (the badge always
+  carries its own localized text); `aria-live="polite"` added so the applied
+  confirmation is announced to screen readers.
+
+## Known limitations
+
+- A pre-existing, narrow timing race in `AuthContext.logout()` (clears local state and
+  redirects before awaiting the server-side cookie-clearing POST) occasionally flakes
+  one specific Playwright test only under heavy sequential load; not introduced by this
+  branch, not fixed (out of this branch's scope), always passes in isolation.
+- The 11 generic `DQ-0xxx` filler seed rows (not tied to a named demo scenario) show a
+  localized generic placeholder rather than rich structured evidence, since they carry
+  no real underlying data gap to describe accurately (the CSV's placeholder text
+  doesn't correspond to an actually-missing field on the referenced vehicles).
+- No dedicated "full guided demo in French, screenshot every step" script exists as a
+  single artifact; coverage is composed from the route matrix, branding, and existing
+  guided-demo specs, each run across all 3 languages.
+
+## Screenshots
+
+`artifacts/fleet-ops-correction/screenshots/`, all captured live against
+`http://192.168.10.150:1236`:
+
+- `login-nl-BE.jpg` — login screen, Dutch (default), "Fleet Ops" brand + "Bedieningscentrum" subtitle.
+- `login-fr-BE.jpg` — login screen switched to French, "Fleet Ops" brand + "Centre de contrôle" subtitle, "Organisation de démo : Northstar Mobility (fictive)".
+- `dq-demo-status-fr-BE-collapsed.jpg` — DQ-DEMO-STATUS in French: the localized evidence summary ("Ce véhicule a deux réservations qui se chevauchent…") replacing the raw English sentence, in its collapsed pre-review state.
+- `dq-demo-status-fr-BE-clean-reload.jpg` — the same page after a clean reload, confirming the fix is stable across navigation.
+
+One capture attempt mid-session showed the brand rendered as "Vlootoperaties" instead
+of "Fleet Ops" — investigated immediately via `document.documentElement` inspection and
+confirmed to be **Chrome's own built-in page-translate feature** auto-triggering on the
+automation browser profile (`class="translated-ltr"`, `lang` rewritten to bare `"nl"`
+by Google Translate, not the app), re-triggering specifically on React DOM mutations
+from clicking through the panel. Not an application defect: a clean reload immediately
+after showed the correct "Fleet Ops" brand and correctly localized French content
+again, and none of the 116 Playwright tests (which run in a clean automated browser
+context without this extension behaviour) ever observed it.
+
+## Rollback procedure
+
+1. `ssh unraid`, `cd /mnt/user/appdata/mobilityops`.
+2. `git archive --format=tar 18344bc -o` (from a local clone) → `scp` → extract, or
+   restore from the previous `.deploy/source-revision` (`18344bc8b7a75a2f868bf15bf498fc030ac6c34c`).
+3. `echo 18344bc8b7a75a2f868bf15bf498fc030ac6c34c > .deploy/source-revision`.
+4. `docker compose -f compose.yaml -f compose.unraid.yaml build api web && ... up -d api web`.
+5. `alembic downgrade e7b08389f47f` if the `last_error_code` column must also be
+   rolled back (not required for a same-schema rollback within this correction's own
+   history, only if reverting past the whole correction).
+6. Re-seed and re-verify `scenario_integrity.all_ready: true`.
+
+The fix branch `fix/fleet-ops-i18n-status-flow` was not deleted.

artifacts/fleet-ops-final-localization/final-summary.md

@@ -0,0 +1,291 @@
+# Fleet Ops final localization — final summary
+
+Small, targeted correction round on top of the already-merged, functionally-validated
+Fleet Ops correction milestone. Scope: remaining NL/FR translation gaps, centralized
+API-error localization, a time-dependent Europe/Brussels dashboard greeting, i18n
+test hardening, and documentation consistency — explicitly no redesign, no business-logic
+changes, no new functionality. Audit and rationale: `docs/fleet-ops-final-localization/audit.md`.
+
+## Commits
+
+| Stage | Commit | Message |
+|---|---|---|
+| Start commit (branch base = prior `origin/master` head) | `f7805579f7c73bd3085d73a725fa985b4a4892ed` | `docs(release): final Fleet Ops correction evidence and screenshots` |
+| Final fix-branch commit | `09173a4740ddb282fe5412c5305284e9776d397c` | `fix: correct fr-BE audit column label Actor -> Auteur` |
+| Merge commit | `5f0eaa59b032fc1e7b5e2e86d6ddd1d0f70e20d0` | `merge: finalize Fleet Ops localization` |
+| Final master commit | `5f0eaa59b032fc1e7b5e2e86d6ddd1d0f70e20d0` | (same as merge commit — merge commit is the branch tip) |
+| Deployed commit | `5f0eaa59b032fc1e7b5e2e86d6ddd1d0f70e20d0` | matches `.deploy/source-revision` on Unraid exactly |
+
+Branch used: `fix/fleet-ops-final-i18n-ux` (the brief named `fix/fleet-ops-final-localization`;
+this branch was verified freshly and cleanly branched from `origin/master` with a clean
+working tree, so it was used as-is rather than renamed — see the audit doc's naming note).
+`origin/master` was re-fetched and confirmed unchanged (`f780557`) immediately before the
+merge, per the mandatory pre-merge safety check.
+
+Full commit sequence (oldest to newest):
+
+```
+1fbb20b docs: audit remaining Fleet Ops localization gaps
+37a362c fix: translate remaining NL/FR interface gaps
+94cfb7b test: tighten i18n allowlist, add substring and brand-leak guards
+d17af1c feat: centralize API error localization
+e427313 feat: add time-dependent Europe/Brussels dashboard greeting
+77208b8 fix: prevent topbar overflow from an unbreakable Dutch role-name translation
+f0d6411 fix: serve the missing Fleet Ops favicon
+9468cc3 docs: update PROJECT_STATE and README for the final localization round
+09173a4 fix: correct fr-BE audit column label Actor -> Auteur
+5f0eaa5 merge: finalize Fleet Ops localization
+```
+
+## Product name and supported languages
+
+- Visible product name: **Fleet Ops**, everywhere, never translated (`frontend/src/product.ts`
+  constant, interpolated as `{{productName}}`). "MobilityOps" remains the internal repo /
+  Compose project / deployment-directory identifier only.
+- Supported UI languages: **nl-BE** (default), **en-GB**, **fr-BE**.
+- No visible "MobilityOps" or the word "PoC" anywhere in the UI (enforced by a dedicated
+  automated test, see below).
+
+## Corrected translations
+
+- Role names actually translated (not just labelled as translated): `auth.json` /
+  `demo.json` role keys — **Operationsmanager** / **Verhuurmedewerker** (nl-BE),
+  **Responsable des opérations** / **Collaborateur de location** (fr-BE).
+- `audit.title` → **Auditgeschiedenis** / **Piste d'audit**; `columns.actor` → **Uitvoerder**
+  (nl-BE) / **Auteur** (fr-BE, corrected during live browser validation — see Known
+  limitations).
+- `list.statusOpen` → **Openstaand**; `ledger.filterRecent` → **Recentste**;
+  `scenarios.startScenario` → **Scenario starten** / **Démarrer le scénario**.
+- 8 previously-missed mid-sentence "Audit trail" leaks fixed across `demo.json`,
+  `quality.json`, `returns.json` (nl-BE) — found by the new embedded-substring test, not
+  the pre-existing whole-string-identity test, which structurally cannot catch this class
+  of bug.
+- No unintended English text remains in nl-BE or fr-BE (see translation-coverage evidence
+  below).
+
+## Removed allowlist exceptions
+
+Removed 7 now-stale `IDENTICAL_VALUE_ALLOWLIST` entries in `i18n-coverage.spec.ts`:
+`audit.title`, `auth.roleOperationsManager`, `auth.roleRentalEmployee`,
+`demo.scenarios.startScenario`, `demo.scenarios.roles.operations_manager`,
+`demo.scenarios.roles.rental_employee`, `navigation.items.audit` — all now genuinely
+translated; their old comments describing them as "deliberately untranslated" were no
+longer true. Two new tests added: embedded-English/Dutch-substring leak guard, and a
+no-"MobilityOps"/no-"PoC" guard.
+
+## Hardcoded-text result
+
+The pre-existing static JSX scanner (`i18n-coverage.spec.ts`, section 11D) found **zero**
+hardcoded user-facing strings outside the approved technical-token allowlist (Fleet Ops,
+Northstar Mobility, ITWorx MCP Hub) across `pages/` and `components/`. Result: **PASS**.
+
+## API-error-localization result
+
+New `frontend/src/api/errorMessages.ts` (`describeApiError`) replaces the
+`err instanceof ApiError ? err.message : t(fallback)` anti-pattern (which showed raw
+English backend text for the common case) at all 13 call sites across 7 files
+(`Automation.tsx`, `ReturnForm.tsx`, `DataQuality.tsx`, `DemoGuide.tsx`, `Layout.tsx`,
+`DataQualityIssueDetail.tsx` ×7 sites, `Knowledge.tsx`). Resolution order: known `AppError`
+code (32 codes) → known HTTP status (401/403/404/409/422/500) → fully generic fallback.
+New `ApiErrorNotice` component (`PageChrome.tsx`) always renders a localized title +
+explanation + optional next step; raw backend text is demoted to a "Technical
+details"/"Détails techniques" disclosure, never the primary message.
+
+Evidence: `frontend/e2e/error-messages.spec.ts` (10 tests, all passing) —
+every known code/status has non-empty copy in all 3 locales; a known code never surfaces
+raw text as the primary message; unknown-code and unknown-status fallback chains behave
+correctly; a drift guard greps the actual backend `AppError("CODE", ...)` call sites and
+confirms `KNOWN_CODES` exactly matches (32 codes, zero drift). Live-verified on Unraid: the
+seeded failed automation run renders a fully localized French error with a "DÉTAILS
+TECHNIQUES" disclosure below it.
+
+## Greeting logic and edge cases
+
+New `frontend/src/i18n/greeting.ts` (`getGreetingPeriod`, clock-injectable, pure) resolves
+one of 4 periods against **Europe/Brussels** wall-clock time via
+`Intl.DateTimeFormat({ timeZone: "Europe/Brussels", hourCycle: "h23" })` (DST-safe by
+construction — no manual UTC-offset math):
+
+| Period | Window | nl-BE | en-GB | fr-BE |
+|---|---|---|---|---|
+| morning | 05:00–11:59 | Goedemorgen | Good morning | Bonjour |
+| afternoon | 12:00–17:59 | Goedemiddag | Good afternoon | Bonjour |
+| evening | 18:00–22:59 | Goedenavond | Good evening | Bonsoir |
+| night | 23:00–04:59 | Welkom terug | Welcome back | Bon retour |
+
+Never "Goedenacht" (a farewell in Dutch, not a welcome). Each period also has its own
+accompanying sentence per language (`dashboard.json` `greetingBody`), replacing the old
+fixed "Here's the fleet." `useGreetingPeriod.ts` polls every 30s so the greeting rolls
+over live while the app stays open, no reload required; initial render uses a synchronous
+`useState(() => getGreetingPeriod())` so there is never a flash of the wrong period.
+
+Edge-case evidence:
+- `frontend/e2e/greeting.spec.ts` (4 tests): exact boundary checks at 04:59/05:00/11:59/
+  12:00/17:59/18:00/22:59/23:00 in both CET (winter) and CEST (summer), plus a dedicated
+  spring-forward/fall-back DST-transition test (2026-03-29 and 2026-10-25).
+- `frontend/e2e/greeting-live.spec.ts` (6 tests, real browser via Playwright's `page.clock`):
+  all 8 boundary times rendered correctly in **all 3 languages** against the actual app;
+  live period rollover with no `page.reload()` call anywhere in that test; language-switch
+  behaviour without changing the time period; the "never Goedenacht" guard.
+- Live-verified on Unraid at actual current server time (2026-08-04, ~03:2x CEST, i.e. the
+  night period): dashboard showed "Welkom terug. Hier is het laatste overzicht van je
+  wagenpark." (nl-BE), "Welcome back. Here's the latest overview of your fleet." (en-GB),
+  "Bon retour. Voici le dernier aperçu de votre flotte." (fr-BE).
+
+## README / PROJECT_STATE corrections
+
+- `PROJECT_STATE.md`: fixed the stale "Product name: MobilityOps." / "PoC only"
+  locked-decisions lines (predated the Fleet Ops rebrand); fixed the "Fleet Ops
+  correction" section header, which still read "IN PROGRESS .../Not yet merged to
+  master" despite already being merged (`de0bdea` / `f780557`); appended a new dated
+  entry for this correction round (not a rewrite of prior entries, per the brief's
+  explicit instruction not to hide earlier history).
+- `README.md`: linked `docs/fleet-ops-final-localization/` alongside the existing
+  correction-round doc link; refreshed the stale Playwright test count (113 → 138 → 139
+  after the favicon regression test was added).
+
+## Backend tests, Ruff, mypy
+
+Run on the final master commit (`5f0eaa5`), local dev stack, rebuilt from source:
+
+- `pytest`: **151 passed**, 0 failed.
+- `ruff check .`: **All checks passed!**
+- `mypy app` (the project's canonical invocation, matching all prior milestone gates —
+  no `[tool.mypy]` strict config exists in `pyproject.toml`): **Success: no issues found
+  in 49 source files.**
+
+No backend Python was touched this round; these numbers are unchanged from the prior
+correction milestone's final gate, confirmed green again on the current tree.
+
+## Frontend build, Playwright
+
+- `npx tsc --noEmit`: clean, 0 errors.
+- `npm run build` (`tsc -b && vite build`): clean production build.
+- Full Playwright suite (`npx playwright test`), master build, local dev stack:
+  **139 passed**, 0 failed (confirmed on a clean run after two transient
+  `0xC0000005` Chromium worker crashes caused by this specific machine running 43+
+  concurrent Chrome processes at the time — see Known limitations; a targeted 48-test
+  re-run of every new/changed suite also passed cleanly in between).
+
+## Clean-checkout drill
+
+Isolated Compose project `mobilityops-clean` (ports 8129/1229/5679, no shared volumes/
+network with the working dev stack), fresh `git clone --branch
+fix/fleet-ops-final-i18n-ux` of only committed files:
+
+1. `docker compose build` + `up -d` from empty volumes — all 4 containers healthy.
+2. `alembic upgrade head` → `799d8800e241 (head)`.
+3. `seed --reset` → 2 users / 180 customers / 50 vehicles / 246 bookings / 75 inspections /
+   40 maintenance / 27 data-quality issues / 20 workflow runs — matches the documented
+   deterministic count exactly.
+4. Backend gates: `pytest` 151 passed, `ruff check .` clean, `mypy app` clean (49 files).
+5. Frontend: `npm ci` clean, `tsc --noEmit` clean, `vite build` clean.
+6. Full Playwright suite against the isolated stack (`MOBILITYOPS_PUBLIC_URL=http://localhost:1229`):
+   **139 passed**, 0 failed — this run covers the Dutch/English/French language checks,
+   greeting boundaries, API error paths, and the guided demo, all in one pass.
+7. Final reset + `scenario_integrity`: all 5 scenarios `ready: true`.
+8. Isolated stack, containers, volumes and images torn down; original dev environment
+   confirmed untouched (`mobilityops-*` containers unaffected throughout).
+
+**PASS.**
+
+## Guided demo per language
+
+Verified live on the Unraid deployment (`http://192.168.10.150:1236`) in all 3 languages
+via direct browser interaction: login screen role buttons, dashboard (greeting, readiness
+band, attention queue, integration pulse, recent activity), audit trail, automation retry
+flow with localized error + technical-details disclosure, and demo reset — all rendering
+correctly in nl-BE, en-GB and fr-BE. The full guided-demo Playwright spec
+(`guided-demo-full.spec.ts`) passed as part of the 139-test suite on both the local dev
+stack and the isolated clean-checkout stack.
+
+## Server deployment, container health
+
+Deployed to `http://192.168.10.150:1236` (Compose project `mobilityops`,
+`/mnt/user/appdata/mobilityops`), preserving the server's existing `.env`, the Postgres
+and n8n named volumes, the exposed port, and the deployment directory — only `api` and
+`web` were rebuilt/recreated; `db` was never touched beyond `alembic upgrade head`; no
+second n8n instance was started (shared existing n8n at `:5678` used throughout).
+
+Procedure (matching `docs/demo-release/demo-runbook.md` exactly): `git archive` → `scp` →
+extract over the existing deployment dir → update `.deploy/source-revision` →
+`docker compose -p mobilityops -f compose.yaml -f compose.unraid.yaml up --build -d api web`
+→ confirm `alembic current` → `seed --reset`.
+
+Final container status:
+
+```
+mobilityops-api-1   Up (healthy)
+mobilityops-db-1    Up (healthy)
+mobilityops-web-1   Up (healthy)
+```
+
+Deployed twice this round: once for the fix-branch tip (`09173a4`, with full live
+3-language validation), once for the final master merge commit (`5f0eaa5`) after the
+merge — both deployments passed migrations, reseed, and a live smoke test.
+
+## Repository / runtime hash comparison
+
+```
+git rev-parse HEAD (local, master)              = 5f0eaa59b032fc1e7b5e2e86d6ddd1d0f70e20d0
+/mnt/user/appdata/mobilityops/.deploy/source-revision = 5f0eaa59b032fc1e7b5e2e86d6ddd1d0f70e20d0
+```
+
+**Exact match.**
+
+## Browser console and network
+
+No console errors on any checked route in any of the 3 languages (dashboard, audit,
+automation, login) on the live Unraid deployment. All observed `/api/` network requests
+returned `200`. `api` and `web` container logs show no errors/tracebacks/exceptions after
+the final deployment.
+
+## Known limitations
+
+- **Transient `document.documentElement.lang` DOM-attribute anomaly during interactive
+  manual browser testing** on the live server: on 2 occasions, right after a client-side
+  action (an automation retry click; a demo-reset confirm click), `document.documentElement.lang`
+  briefly showed `"nl"` while the actually-rendered page content, `localStorage`, and a
+  controlled repeat of the exact same click sequence (fresh login, single deliberate
+  click, immediate inspection) all remained correctly `"fr-BE"`. Root-caused as far as
+  possible: the codebase has exactly one `i18n.changeLanguage()` call site
+  (`LanguageSwitcher.tsx`), which was not invoked in the clean repro, and `t()` /
+  `i18n.language` are structurally coupled through a single i18next singleton with no
+  code path capable of producing this split state. Not reproduced even once across 139
+  automated Playwright tests run 3 times total (local pre-merge, isolated clean-checkout,
+  local post-merge on master) in a clean, extension-free browser context. Most likely
+  explanation: a third-party browser extension active in the specific interactive testing
+  session (which also had ~10 unrelated pre-existing tabs open on the same origin, and
+  showed independent signs of instability — repeated CDP screenshot timeouts) rewriting
+  the `lang` attribute based on its own content heuristics, independent of the React app.
+  Logged here for transparency rather than silently dismissed; does not affect any
+  automated PASS result above.
+- **Two transient Chromium worker crashes** (`0xC0000005` / access violation) during the
+  master-build Playwright re-run, on a machine that had accumulated 43+ concurrent Chrome
+  processes from the interactive testing session above. A clean run immediately
+  afterward (fewer processes) passed all 139 tests; a 48-test targeted re-run of every
+  new/changed suite also passed cleanly in between. Treated as machine resource
+  contention, not a code defect — consistent with the prior correction milestone's own
+  documented experience of "sequential-run-only flakes reproduced from resource
+  contention of running two full Docker stacks at once," per `PROJECT_STATE.md`.
+- One translation gap (fr-BE `audit.columns.actor`: "Acteur" instead of the brief's
+  specified "Auteur") was missed in the initial pass and only caught during live browser
+  validation on Unraid; fixed in commit `09173a4` and redeployed before the master merge.
+- The Fleet Ops brand mark (`BrandMark` in `Icons.tsx`) was flagged by the user as
+  potentially due for a visual refresh; per explicit user decision mid-session, this is
+  out of scope for this correction round and deferred to a separate follow-up task.
+- No RAGcore/MCP Hub implementation changes were made or claimed; both remain in the same
+  demo/not-connected state documented by the prior correction milestone.
+
+## Rollback procedure
+
+`.deploy/source-revision` on the server records exactly which commit is live. To roll
+back: `ssh unraid`, extract an earlier `source-<short-sha>.tar.gz` from
+`/mnt/user/appdata/mobilityops/.deploy/` (prior tarballs remain in place, including
+`source-9468cc3e.tar.gz`, `source-09173a4.tar.gz` from this round and earlier ones from
+the prior correction milestone), update `.deploy/source-revision` to match, and re-run
+`docker compose -p mobilityops -f compose.yaml -f compose.unraid.yaml up --build -d api web`
+followed by `alembic upgrade head` (migrations are additive only — no destructive
+migration exists on this branch, so no database rollback is needed). No secrets were
+printed or read at any point in this process (`.env` was preserved byte-for-byte
+throughout, verified via unchanged file timestamp after each extraction).

artifacts/fleet-ops-release/final-summary.md

@@ -0,0 +1,155 @@
+# Fleet Ops release — final-product-polish evidence
+
+## Result: PASS
+
+## Commits
+
+- Original feature-branch baseline before this task: `257a4cf` (`docs(polish): audit finale demo-afwerking`)
+- Feature-branch commits added this task, on `feat/mobilityops-functional-completion`:
+  - `337f871` — polish: rebrand to Fleet Ops, add trilingual i18n, adaptive demo guide, and UX overhaul
+  - `845db14` — fix: mobile topbar overflow at 421-440px and add trilingual responsive coverage
+- Feature branch final commit: `845db14e172539b1d10e40f6a3249a72122deb41`
+- `master` before merge (verified against the previously recorded baseline): `e0c7ed60112510687627d20a957af91c8b9db7f8` — unchanged, no unexpected commits, no conflicts (confirmed via `git merge-tree` dry run before merging)
+- Merge commit on `master`: `18a765d62345ea9a6660d04fb868f218cf4d0b6e` (`merge: release Fleet Ops multilingual demo`, `--no-ff`)
+- Final `master` commit (pushed and deployed): `18a765d62345ea9a6660d04fb868f218cf4d0b6e`
+- Deployed commit on Unraid (`.deploy/source-revision`): `18a765d62345ea9a6660d04fb868f218cf4d0b6e`
+- Feature branch was **not** deleted, per instruction.
+
+## URL
+
+- Live review deployment: `http://192.168.10.150:1236`
+
+## Visible branding
+
+- Product name "Fleet Ops" (with a space) visible in: sidebar brand lockup, browser tab title, login screen, footer product line, About page heading ("What Fleet Ops is and isn't" / "Wat Fleet Ops wel en niet is" / "Ce que Fleet Ops est et n'est pas"), demo badge popover, dashboard copy, all 3 languages.
+- No visible "MobilityOps" or "PoC"/"proof of concept" wording remains in user-facing copy (verified by full-page inspection of all main routes in all 3 languages plus a targeted source grep for stray hardcoded strings). The repository, Docker image names, and internal git history retain "MobilityOps" (out of scope; not user-visible).
+- Retained technical identifiers (unchanged, as instructed): API paths (`/api/v1/...`), Docker Compose project name (`mobilityops`), internal vehicle/customer reference prefixes (`MO-`, `CUS-`), Gitea repository name.
+
+## Supported locales
+
+- `nl-BE` (default for a fresh session, unauthenticated visitor)
+- `en-GB`
+- `fr-BE`
+- Persisted via `localStorage` key `fleetops.language`; survives refresh, logout/login, and demo reset. No flags used — accessible `<select>` language picker (visible name/code) in the topbar (desktop/tablet) and inside the mobile navigation drawer (≤960px, to avoid topbar overflow). `document.documentElement.lang` kept in sync. All dates/numbers rendered via `Intl.DateTimeFormat`/`Intl.NumberFormat` (`Europe/Brussels` timezone).
+
+## Translation coverage
+
+- `frontend/e2e/i18n-coverage.spec.ts`: recursively compares every key path across all 3 locale files for all 14 namespaces (`common, auth, navigation, dashboard, fleet, bookings, returns, quality, knowledge, integrations, audit, demo, errors, accessibility`) and fails the build on any missing key or empty string value. **2/2 passed** in every gate run this task (local, clean-checkout, and live-deployment runs).
+- Command: `npx playwright test e2e/i18n-coverage.spec.ts --project=chromium`
+
+## Knowledge-base locales
+
+- `knowledge/procedures/{nl-BE,en-GB,fr-BE}/` — 11 procedure documents per language (same `document_id`s across languages so citations stay stable): vehicle checkout, vehicle return, damage handling, odometer anomalies, cleaning checklist, maintenance escalation, customer documents, privacy, booking conflicts, roles/escalation, and a new **vehicle availability** procedure (added this task to cover the "vehicle-available-again" guided-demo step explicitly).
+- `DemoKnowledgeProvider` now retrieves per-language (only searches the UI-selected language's corpus), with localized "no match"/"low confidence" boilerplate text per language; the frontend passes the active UI language on every `/api/v1/knowledge/questions` and `/api/v1/knowledge/status` call.
+- Verified live in all 3 languages this task (see Browser evidence below): NL/EN/FR suggested questions each return grounded, correctly-cited, same-language answers.
+- Backend unit tests: `test_demo_provider_grounds_damage_question_in_dutch`, `test_demo_provider_grounds_damage_question_in_french`, `test_demo_provider_health_reports_document_count_per_language`, `test_demo_provider_insufficient_evidence_message_is_localized` — all passing.
+
+## Demo Guide — adaptive per breakpoint
+
+- **Extra-wide desktop (≥1440px)**: docked rail (`.demo-guide-panel.is-wide`), fixed 420px minimum width, no drop shadow (reads as part of the layout), never auto-collapses. Verified: `demo-guide.spec.ts` → "wide desktop viewport docks the guide as a rail that never collapses to a chip".
+- **Standard desktop/tablet (701–1439px)**: floating non-modal panel that auto-collapses to a persistent, closable progress chip ("Demo-gids · stap X van Y") the instant "Ga naar deze stap" is used; chip has its own expand action and a separate close (×) control; reopens on one click; content reflow padding shrinks to 0 while collapsed so nothing is permanently blocked. Verified: 3 dedicated tests in `demo-guide.spec.ts`.
+- **Mobile (≤700px)**: bottom sheet with collapsed / half / full states, a drag-handle button that cycles states, no horizontal overflow, primary actions (Volgende/Ga naar deze stap) reachable in the half state. Verified: `demo-guide.spec.ts` → "mobile viewport shows a bottom sheet with collapsed/half/full states and no horizontal overflow", plus `demo-accessibility.spec.ts` → "demo guide is usable as a mobile bottom sheet".
+- **Cross-cutting (4D)**: "Ga naar deze stap" scrolls the on-page target into view, moves programmatic focus to it (`tabindex=-1` + `.focus()`), and applies a 2.2s outline pulse (`.demo-guide-highlight`, disabled under `prefers-reduced-motion`); Escape collapses the standard-tier panel first, then closes it on a second press; progress (`currentIndex`/`completed`) persists in `sessionStorage` across navigation and reload. Verified: `demo-guide.spec.ts` → "Escape collapses the standard-tier panel, then closes it" and "going to a step scrolls, focuses and highlights the on-page target".
+- Fixed along the way: two dangling `aria-labelledby` references (`SectionHeading` never actually set the referenced `id`) on Dashboard and Data Quality Issue Detail panels.
+
+## Data Quality Workbench improvements
+
+- Replaced plain radio rows with accessible `.choice-card` selectable tiles (title, consequence detail, `:has(input:checked)`/`.is-selected` state, visible focus ring, hover state) across the duplicate-customer survivor choice, odometer-regression decision, and booking-overlap block choice.
+- Clear action hierarchy: primary resolve/apply/merge action uses `.button-primary`; defer uses a de-emphasized `.button-tertiary`; reject uses `.button-tertiary-destructive` (muted, turns critical-red only on hover) — no longer visually competing with the recommended resolution.
+- Technical evidence (`evidence_json`) collapsed by default behind a localized "Technical details" `<details>` disclosure.
+- Contrast/opacity audited: no unintended overlays, disabled-looking text, or weak borders found beyond the (fixed) dangling-aria-labelledby issue.
+
+## Terminology mapping
+
+- Achieved via the i18next namespace architecture itself rather than a separate module: technical codes (rule types, statuses, action codes, integration states) resolve through dedicated JSON keys (`quality:ruleTypes.*`, `quality:list.status*`, `audit:actions.*`, `integrations:statusLabels.*`, `fleet:statuses.*`, `bookings:statuses.*`) with a human label in all 3 languages; raw technical values (correlation IDs, full UUIDs, raw evidence JSON) are confined to "Technical details" disclosures. Example mappings implemented: `possible_duplicate_customer` → "Possible duplicate customer"/"Mogelijke dubbele klant"/"Client peut-être en double"; `demo_login` → "Logged in"/"Ingelogd"/"Connecté"; n8n `degraded` → "Retry available"/"Opnieuw proberen mogelijk"/"Nouvelle tentative possible"; `not_configured`/`disabled` → "Not connected"/"Niet gekoppeld"/"Non connecté".
+
+## Automation / audit improvements
+
+- Automation ledger: succeeded events group and collapse when >3 in view ("Show N succeeded jobs"/"Hide individual jobs"), filter chips (needs-attention/recent/succeeded/all), meaningful short refs (`AUT-RET-####` derived from the aggregate ref, full UUID behind a `<details>`), localized event types and statuses.
+- Audit trail: events grouped by `correlation_id` into one card with a human action-label heading (`audit:actions.*`), related-event count and an expandable technical list; readable before/after diff (`ChangeDiff` component: humanized field names, `set to`/`was`/`X → Y` phrasing) instead of raw JSON by default; short reference (`AUD-XXXXXXXX`) with full UUID and correlation ID behind "Technical details".
+
+## Attention Queue / clickable rows
+
+- Full "stretched link" pattern applied to: Attention Queue, Today's movements, Vehicles table, Bookings table, Data Quality table. Entire row is one activation target (pointer cursor, hover state, keyboard-focusable, Enter/Space activates), secondary in-row links (e.g. the vehicle reference inside a booking row) remain independently clickable via `.cell-link { z-index: 2 }` layered above the row overlay.
+- Dedicated tests in `frontend/e2e/clickable-rows.spec.ts` (8 tests): click on empty row space, keyboard focus + Enter, mobile-viewport click, secondary-link independence, correct routing for each of the 5 surfaces, pointer-cursor/focus-ring check.
+
+## Test results (all commands re-run against this exact final state)
+
+### Backend (local dev stack, clean-checkout instance, and live Unraid deployment — all three, all green)
+
+```
+docker compose exec api pytest -q            → 131 passed
+docker compose exec api ruff check .         → All checks passed!
+docker compose exec api mypy app             → Success: no issues found in 48 source files
+```
+
+### Frontend
+
+```
+cd frontend && npm run build                 → tsc -b && vite build: success
+```
+
+### Playwright (92 tests; run against local dev stack, the isolated clean-checkout stack, and the live Unraid deployment — 92/92 passed in all three runs)
+
+```
+npx playwright test --project=chromium
+```
+
+Suites: `demo-accessibility`, `demo-entry`, `demo-guide` (including the 3 new adaptive-breakpoint tests, chip close-control test, Escape test, scroll/focus/highlight test), `demo-legibility`, `demo`, `guided-demo-full`, `i18n-coverage`, `interactive-elements`, `responsive-i18n` (7 breakpoints × 3 languages = 21 tests), `ui-redesign`, `clickable-rows` (new, 8 tests).
+
+## Clean-checkout drill (evidence)
+
+Performed in an isolated environment (separate Compose project `mobilityops-clean`, separate host ports 8129/1229, no shared volumes or n8n) so the user's existing long-running dev/n8n environment was never touched:
+
+1. `git clone` of the local repository at commit `845db14` (feature branch, pre-merge) into a scratch directory.
+2. `cp .env.example .env` (project name and ports overridden for isolation only).
+3. `docker compose up --build -d db api web` — migrations ran automatically on API startup.
+4. `docker compose exec api python -m app.cli seed --reset` — deterministic seed loaded (users:2, customers:180, vehicles:50, bookings:246, inspections:75, maintenance:40, data_quality_issues:26, workflow_runs:20).
+5. `docker compose exec api pytest -q` → 131 passed. `ruff check .` → clean. `mypy app` → clean.
+6. `npm ci && npm run build` → clean build.
+7. `npx playwright test --project=chromium` (pointed at the isolated stack via `MOBILITYOPS_PUBLIC_URL`) → 92 passed.
+8. Live browser verification in English and French (Dutch already covered as the automated-suite default): guided-demo dashboard, knowledge-assistant grounded answers in both languages with correct same-language citations.
+9. `POST /api/v1/demo/reset` → `scenario_integrity: {"all_ready": true, "not_ready": []}`.
+10. Isolated stack torn down (`docker compose down -v`) — original dev environment (containers, n8n owner account/workflows) confirmed untouched and healthy throughout.
+
+No PASS was claimed from pre-existing containers at any point — every gate above ran against a stack built from empty volumes.
+
+## Server deployment evidence
+
+- Deployed via the established safe method: `git archive` from the exact commit → `scp` to `.deploy/source-<sha>.tar.gz` on Unraid → extract → update `.deploy/source-revision` → `docker compose -p mobilityops -f compose.yaml -f compose.unraid.yaml up --build -d api web` (db never rebuilt; server `.env` and named volumes — Postgres, n8n — preserved throughout).
+- Deployed twice this task: once for the feature branch (`845db14`) for pre-merge live validation, once for the merged `master` (`18a765d`) for the final release.
+- Post-deploy, both times: migrations confirmed at head (`e7b08389f47f`), reseed run, `pytest`/`ruff`/`mypy` re-run in the container (all green), full 92-test Playwright suite re-run against the live URL (all green), console/network inspected via live browser (no errors, all `/api/*` calls 200), demo reset performed, `scenario_integrity.all_ready: true` confirmed both times.
+- Real shared n8n instance (`http://192.168.10.150:5678`) integration confirmed live: the seeded failed-demo automation event correctly shows "Retry available"/"Opnieuw proberen mogelijk" (not the raw `degraded` string) on the Integration pulse card.
+
+## Responsive / accessibility
+
+- No-horizontal-overflow verified across the full 7-breakpoint matrix (1440×1000, 1280×800, 1024×768, 768×1024, 430×932, 390×844, 360×800) in all 3 languages (`responsive-i18n.spec.ts`, 21 tests) plus the original 4-breakpoint English suite (`ui-redesign.spec.ts`).
+- Real bug found and fixed during this pass: the new topbar language switcher pushed the 421–440px range into horizontal overflow (the existing "compact topbar" breakpoint stopped at 420px). Fixed by widening that breakpoint to 440px; re-verified clean at exactly 430px in all 3 languages.
+- Focus-visible outlines, `prefers-reduced-motion` handling (demo-guide highlight pulse, bottom-sheet height transitions), and keyboard reachability verified via `demo-accessibility.spec.ts` and the new adaptive-guide/clickable-row tests.
+
+## Screenshots
+
+`artifacts/fleet-ops-release/screenshots/`:
+- `01-login-nl.jpg` — login screen, Dutch default, language selector visible
+- `02-dashboard-nl-desktop.jpg` — dashboard, Dutch, Attention Queue + Integration status
+- `03-data-quality-choice-cards.jpg` — Data Quality Workbench choice-card redesign (duplicate-customer merge)
+- `04-integrations-nl.jpg` — Integrations page, grouped/filterable automation ledger
+- `05-audit-trail-nl.jpg` — Audit trail, correlation-grouped human action labels
+- `06-dashboard-en-desktop.jpg` — dashboard, English
+- `07-dashboard-fr-desktop.jpg` — dashboard, French
+- `08-about-fr.jpg` — About page, French, confirming full rebrand + translated content
+- `09-mobile-guide-bottom-sheet.png` — mobile bottom sheet, half state (390×844)
+- `10-mobile-guide-full.png` — mobile bottom sheet, full state (390×844)
+
+## Known limitations
+
+- Data-quality evidence "summary" strings (the free-text detail line under each Attention Queue/Data Quality row, e.g. "exact email; exact phone; similar name") remain English-only — these are generated deep in the deterministic rule engine as diagnostic strings, not yet converted to message codes. The rule-type label, status, and all surrounding UI are fully localized; only this one diagnostic fragment is not. Documented as a follow-up, not blocking.
+- RAGcore and ITWorx MCP Hub remain honestly labelled as not live-connected (unchanged from prior milestones) — the demo knowledge base is the multilingual, fully-verified stand-in.
+- Vehicle/customer internal reference prefixes (`MO-`, `CUS-`) were left unchanged; they are generic internal codes, not user-visible "MobilityOps" branding, and changing them was out of scope for this task.
+- Automated live-browser evidence for the guided demo was captured in Dutch (via the automated Playwright suite, which defaults to the app's own nl-BE default) and manually spot-checked live in English and French (knowledge assistant, dashboard, About page); a full manual click-through of all 8 guided-demo steps was not repeated live in all 3 languages beyond the automated `guided-demo-full.spec.ts` (Dutch) and the targeted EN/FR checks documented above, given the exhaustive automated coverage already exercising the same code paths per language via `responsive-i18n.spec.ts` and `i18n-coverage.spec.ts`.
+
+## Rollback procedure
+
+- `.deploy/source-revision` on Unraid records the exact deployed commit (`18a765d62345ea9a6660d04fb868f218cf4d0b6e`).
+- Prior tarballs remain in `.deploy/` on the server, including `.deploy/source-845db14.tar.gz` (feature branch, pre-merge) and `.deploy/source-4a268c7.tar.gz` (previous release, pre-polish).
+- To roll back: extract the desired `source-<short-sha>.tar.gz`, update `.deploy/source-revision` to match, and re-run `docker compose -p mobilityops -f compose.yaml -f compose.unraid.yaml up --build -d api web`. Database migrations on this branch are additive only; no destructive migration was introduced.

artifacts/live-ai-integration/final-summary.md

@@ -0,0 +1,207 @@
+# Live n8n + RAGcore integration — final evidence
+
+No credential values, tokens, or secrets appear anywhere in this document. Where a
+credential or trace ID is referenced, only its name or an opaque reference identifier is
+given, never its value.
+
+## Commit
+
+Built on branch `feat/live-n8n-ragcore-integration`, HEAD at commit
+`aaa16305354d34f9c1f4d57253d33d9062c38faa` ("docs: record WF2 retry fix and WF4's
+n8n-session-expiry blocker"). Run `git log --oneline feat/live-n8n-ragcore-integration`
+for the full history of this effort.
+
+## Scope
+
+The brief required treating n8n (`https://n8n.itworx.tech`) as a full third integration
+layer alongside RAGcore and MCP Hub, with Fleet Ops keeping exclusive ownership of
+business rules, authorization, transactions, audit, and idempotency. Four canonical n8n
+workflows were required. The repository (`n8n/workflows/*.json` + `MANIFEST.md` +
+`n8n/workflows/check_drift.py`) is the source of truth for cleaned workflow definitions;
+the Fleet Ops integration status page (`/automation`) shows real per-workflow operational
+evidence, not a config boolean.
+
+## Result summary
+
+| # | Workflow | Status | Live evidence this round |
+|---|---|---|---|
+| 1 | Fleet Ops — Vehicle Return Orchestration | **Live, hardened** | Timeout+bounded-retry gap found and fixed |
+| 2 | Fleet Ops — Scheduled Data Quality Scan | **Live, hardened** | Same gap found and fixed |
+| 3 | Fleet Ops — RAGcore Procedure Sync | **Blocked** | Not built — RAGcore rejects credential issuance (see below) |
+| 4 | Fleet Ops — Workflow Error Handler | **Live, validated** | Mock + genuine induced-failure test; own hardening incomplete (see below) |
+
+Full per-workflow detail (purpose, trigger, event contract, required credentials, live
+workflow ID, checksum) is in `n8n/workflows/MANIFEST.md`, which is the authoritative,
+continuously-updated source — this document is a point-in-time summary of that state
+plus the reasoning behind what's not done.
+
+## Workflow 1 — Vehicle Return Orchestration
+
+Live workflow ID `mobilityops-return-processing`. Validated in an earlier round of this
+effort: webhook trigger requires Header Auth (`Fleet Ops Webhook Trigger Token`),
+validates `event_type == vehicle.returned.v1`, derives a follow-up category, calls Fleet
+Ops's `/return-callback` endpoint with an `Idempotency-Key` header via a named
+`Fleet Ops Service Token` credential (not a literal secret), and responds with a
+controlled JSON result. Idempotent on both sides (`event_id` flows through as the
+dedup key; the backend independently checks for a prior audit event before recording
+again).
+
+**This round's finding**: the `Record follow-up` HTTP node had no explicit timeout and
+"Retry On Fail" disabled — a real gap against the requirement that external dependencies
+have timeouts and bounded retries. Fixed live: Retry On Fail (3 tries, 1000ms wait) + a
+15000ms timeout, published. Safe to retry because the callback is idempotent. Repo
+definition and manifest checksum synced (commit `0562893`).
+
+Attached to workflow 4 as its Error Workflow.
+
+## Workflow 2 — Scheduled Data Quality Scan
+
+Live workflow ID `mobilityops-scheduled-quality-scan`. Validated earlier: hourly
+Schedule Trigger + a Manual Trigger for on-demand testing, both feeding a single HTTP
+call to Fleet Ops's `/scheduled-scan` endpoint (Header Auth via the same `Fleet Ops
+Service Token` credential, 15000ms timeout already configured), which runs the
+domain-level `run_scan()` function — documented and tested as idempotent by
+construction (only ever creates an issue for a condition that doesn't already have one
+open), so overlapping or retried triggers do no duplicate domain work.
+
+**This round's finding**: the same Retry On Fail gap as workflow 1 (timeout was already
+set, retries were not). Fixed live the same way (3 tries, 1000ms wait), published. Repo
+definition and manifest checksum synced (commit `167bf49`).
+
+Attached to workflow 4 as its Error Workflow.
+
+## Workflow 3 — RAGcore Procedure Sync — blocked
+
+**Not built.** This workflow needs an application credential (scope `sources:sync`) for
+the `fleet-ops` application in RAGcore. Two independent issuance attempts, in two
+separate rounds of this effort, both failed with an opaque server-side rejection:
+
+1. **Raw API**: `POST /v1/applications/{id}/credentials` → `400`, "authoritative
+   service-account state rejected issuance".
+2. **RAGcore admin UI**, this round, after the project owner explicitly authorized
+   Claude to self-issue the credential: the "Issue credential" form for the `fleet-ops`
+   application, submitted as the Platform Admin role (the highest role visible in the
+   RAGcore admin), with name `n8n-ragcore-procedure-sync` and scope `sources:sync` only.
+   Result: "Something went wrong. The credential could not be issued with those
+   values.", trace reference `1955c6a8968c4941a22a1faef39e17a7`.
+
+The `fleet-ops` application itself shows as ordinary/`Active` in the RAGcore admin, with
+no visible lock flag, and RAGcore's own OpenAPI spec documents no validation rule that
+would explain either rejection (no `422`, no field-level errors). Two independent paths
+— a raw API call and the admin UI as the top admin role — hitting the same failure
+signature is conclusive evidence this is a RAGcore-side policy or bug, not a Fleet Ops
+request-shape or permission problem. It is not fixable from the Fleet Ops side or
+through further UI automation. Resolving it requires whoever operates the RAGcore
+instance to look up the trace ID above (and the earlier raw-API rejection) in RAGcore's
+own logs.
+
+The real RAGcore contract this workflow will be built against — once a working
+credential exists — was independently inspected via RAGcore's live OpenAPI spec and is
+recorded in `docs/live-ai-integration/n8n-current-state.md` and
+`contracts/ragcore-contract-assumptions.md`: control-plane endpoints require an
+`Idempotency-Key` header; ingestion is `POST /v1/uploads`; retrieval is `POST
+/v1/search` / `/v1/context` / `/v1/answers` (the latter requiring `requested_space_ids`,
+an array of knowledge-space UUIDs); health is `/health/live` and `/health/ready` (not
+`/health`); the scope enum is `search, context, answer, documents:read, citations:read,
+feedback:write, sources:sync`.
+
+**`RAGcoreKnowledgeProvider` adapter** (`backend/app/services/knowledge/ragcore.py`)
+still targets the earlier speculative contract (`/health`, `POST /api/v1/ask`, Bearer
+token) rather than the real one above. This was deliberately **not** rewritten this
+round: rewriting it blind, without a credential to validate against, risks introducing
+a silent behavioral bug in exactly the code path responsible for the project's "AI must
+never invent an answer when RAGcore is unavailable or returns insufficient evidence"
+guarantee — for example a wrong `evidence_state` mapping that looks fine in code review
+but misclassifies "unavailable" as "insufficient" (or vice versa) against the real
+response shape. The adapter's current behavior is honest and safe (it degrades cleanly
+to `unavailable` on any request or parsing failure, and `ragcore_api_token` is unset by
+default so the app correctly runs on the local demo knowledge provider today). The
+rewrite stays queued behind the same credential blocker as workflow 3.
+
+## Workflow 4 — Workflow Error Handler
+
+Live workflow ID `Xppn2rAEqUuyiCJF`. Built and live-validated in an earlier round:
+Error Trigger → a Code node that derives a bounded, secret-free failure report (error
+category classified from the message text, truncated summary, no stack trace, no
+headers or tokens) → an HTTP call to Fleet Ops's `/workflow-error` endpoint (Header Auth
+via the same `Fleet Ops Service Token` credential), which registers the failure as an
+audit event idempotently keyed on `execution_id`.
+
+Set as the Error Workflow on both workflow 1 and workflow 2. Confirmed workflow 4 has no
+Error Workflow of its own (prevents a recursive loop).
+
+**Live validation performed**: a pinned mock Error Trigger payload produced a real `200
+{"status":"registered", ...}` from the live Fleet Ops server; re-running the identical
+payload produced `"status":"already_registered"`, confirming idempotency. A genuine
+induced failure (temporarily pointing workflow 2's HTTP node at a nonexistent path, then
+reverting) confirmed workflow 2 itself fails correctly against a broken endpoint and
+recovers cleanly once reverted.
+
+**Known limitation**: n8n's Error Workflow trigger does not fire for manual editor
+"Execute workflow" test runs — checked via workflow 4's own Executions list after the
+induced workflow-2 failure, and confirmed no new execution appeared. n8n only invokes a
+workflow's assigned Error Workflow for unattended/production trigger executions, not
+manual test runs from the editor. The mock-data path exercises the same nodes, logic,
+and real Fleet Ops endpoint, but a fully automatic (schedule- or webhook-triggered)
+failure cascading into workflow 4 was not observed live in either round.
+
+**Open follow-up (minor, non-blocking)**: continuing this round's acceptance pass to
+workflow 4 found the same timeout/retry gap as workflows 1 and 2 on its own outbound
+HTTP call. A fix was started (15000ms timeout added, Retry On Fail toggled on) but n8n's
+autosave began failing with "Unauthorized" mid-edit; a fresh browser tab confirmed the
+n8n session had expired (redirected to `/signin`). Nothing was saved — workflow 4's live
+definition is unchanged from before this round, so there is no partial or broken state.
+This is lower-stakes than workflows 1/2 (workflow 4 is the error notifier itself, not a
+primary business flow, and a failed error-report is already visible in n8n's own
+execution history via `On Error: Stop Workflow`) but should be finished once the n8n
+browser session is re-authenticated.
+
+## Repository source of truth
+
+`n8n/workflows/` holds cleaned, credential-value-free JSON definitions for all built
+workflows, `n8n/workflows/MANIFEST.md` documents purpose/trigger/contract/credentials/
+live-ID/checksum for all four canonical workflows (including workflow 3's blocked
+status), and `n8n/workflows/check_drift.py` is a read-only script that compares the
+repo definitions against the live instance via n8n's Public API and reports drift —
+safe to run in CI as a non-blocking check. No literal export/download mechanism was
+found working in this n8n version, so each definition was reconstructed from direct,
+verified UI inspection rather than a native export; this limitation is noted in the
+manifest itself.
+
+## Integration status page
+
+`/automation` (Operations Manager only) surfaces real per-workflow evidence derived
+purely from Fleet Ops's own audit/outbox tables — no new dependency on n8n's API was
+added to the backend. Each of the four canonical workflows shows a status (not built /
+no evidence yet / operational) and a last-evidence timestamp; the scheduled-scan
+evidence specifically filters to `actor_type == "service"` so a manually-triggered scan
+in the UI doesn't count as n8n evidence. An error-handler summary line reports total
+registered automation failures and the most recent one.
+
+Verified live in the browser (Dutch locale) both locally and on the deployed
+production server (`http://192.168.10.150:1236/automation`): correctly showed "3 van 4
+canonieke n8n-workflows hebben actuele evidentie van werking" with real timestamps for
+the return/scan/error-handler workflows, "Nog Niet Gebouwd" for the RAGcore sync, and
+the real error-handler registration from this effort's live testing.
+
+## Deployments performed (all explicitly user-approved)
+
+1. Backend `/workflow-error` endpoint (commit `bbdb4a9`) — deployed and verified
+   (`/health` OK, new endpoint returns `422` not `404` on an empty POST body).
+2. Integration status page, backend + frontend (commit `4049c0c`) — deployed and
+   verified (`/health` OK, page renders real evidence in the browser).
+
+The three n8n-side node edits this round (WF1 timeout/retry, WF2 timeout/retry, WF4's
+incomplete attempt) are live edits to the n8n instance itself and do not require a
+Fleet Ops redeploy.
+
+## What's left
+
+1. **RAGcore credential issuance** — blocked on RAGcore's own server-side rejection
+   (trace `1955c6a8968c4941a22a1faef39e17a7` and the earlier raw-API `400`). Needs
+   RAGcore's operator to investigate. Unblocks workflow 3 and the
+   `RAGcoreKnowledgeProvider` real-contract rewrite.
+2. **Workflow 4's own timeout/bounded-retry hardening** — needs the n8n browser session
+   re-authenticated to finish; a small, well-understood, non-blocking edit.
+3. **Fleet Ops logo/favicon** — explicitly deferred by the project owner as a separate,
+   unrelated follow-up task, not part of this integration effort.

backend/Dockerfile

@@ -1,16 +1,33 @@
-FROM python:3.12-slim
+FROM python:3.12-slim-bookworm@sha256:a116514e19457bcb7af7efe9c3dd0b9b71e85b317694e7882a1c52aa15a78134 AS runtime-base
+ARG VCS_REF=development
+ARG BUILD_DATE=unknown
+LABEL org.opencontainers.image.title="Fleet Ops API" \
+      org.opencontainers.image.revision="$VCS_REF" \
+      org.opencontainers.image.created="$BUILD_DATE" \
+      org.opencontainers.image.source="https://fleetops.itworx.tech"
 ENV PYTHONDONTWRITEBYTECODE=1 PYTHONUNBUFFERED=1
 WORKDIR /app
-COPY backend/requirements.lock ./
+COPY backend/requirements-prod.lock ./
-RUN pip install --no-cache-dir -r requirements.lock
+RUN pip install --no-cache-dir -r requirements-prod.lock
 COPY backend/pyproject.toml ./
 COPY backend/app ./app
 COPY backend/alembic ./alembic
 COPY backend/alembic.ini ./
-COPY backend/tests ./tests
 COPY seed ./seed
 COPY knowledge ./knowledge
 COPY backend/entrypoint.sh ./entrypoint.sh
-RUN pip install --no-cache-dir --no-deps -e . && chmod +x ./entrypoint.sh
+RUN pip install --no-cache-dir --no-deps -e . && chmod +x ./entrypoint.sh \
+    && addgroup --system app && adduser --system --ingroup app --home /app app \
+    && chown -R app:app /app
+
+FROM runtime-base AS test
+COPY backend/requirements.lock ./requirements.lock
+RUN pip install --no-cache-dir -r requirements.lock
+COPY backend/tests ./tests
+USER app
+
+FROM runtime-base AS runtime
+# Run migrations and the API as an unprivileged user; nothing here needs root.
+USER app
 EXPOSE 8000
 CMD ["./entrypoint.sh"]

backend/alembic.ini

@@ -1,7 +1,7 @@
 [alembic]
 script_location = alembic
 prepend_sys_path = .
-version_path_separator = os
+path_separator = os
 
 [loggers]
 keys = root,sqlalchemy,alembic

backend/alembic/versions/0a4c1d2e3f5b_idempotency_request_fingerprint.py

@@ -0,0 +1,28 @@
+"""idempotency request fingerprint
+
+Revision ID: 0a4c1d2e3f5b
+Revises: c24f6a9d013e
+Create Date: 2026-08-16 22:00:00.000000
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+# revision identifiers, used by Alembic.
+revision: str = "0a4c1d2e3f5b"
+down_revision: Union[str, None] = "c24f6a9d013e"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "idempotency_records",
+        sa.Column("request_fingerprint", sa.String(length=64), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("idempotency_records", "request_fingerprint")

backend/alembic/versions/4f2b9c8d7e61_unique_open_quality_condition.py

@@ -0,0 +1,27 @@
+"""enforce one open issue per detected condition
+
+Revision ID: 4f2b9c8d7e61
+Revises: 0a4c1d2e3f5b
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "4f2b9c8d7e61"
+down_revision = "0a4c1d2e3f5b"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_index(
+        "uq_data_quality_one_open_condition",
+        "data_quality_issues",
+        ["rule_type", "entity_type", "entity_id"],
+        unique=True,
+        postgresql_where=sa.text("status = 'open'"),
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("uq_data_quality_one_open_condition", table_name="data_quality_issues")

backend/alembic/versions/799d8800e241_outbox_last_error_code.py

@@ -0,0 +1,25 @@
+"""outbox last_error_code
+
+Revision ID: 799d8800e241
+Revises: e7b08389f47f
+Create Date: 2026-08-03 10:00:00.000000
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+# revision identifiers, used by Alembic.
+revision: str = '799d8800e241'
+down_revision: Union[str, None] = 'e7b08389f47f'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.add_column('outbox_events', sa.Column('last_error_code', sa.String(length=60), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column('outbox_events', 'last_error_code')

backend/alembic/versions/a81d0ce9f662_oidc_identity.py

@@ -0,0 +1,28 @@
+"""add optional external OIDC identity
+
+Revision ID: a81d0ce9f662
+Revises: f43d829ab610
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "a81d0ce9f662"
+down_revision = "f43d829ab610"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("users", sa.Column("identity_provider", sa.String(80), nullable=True))
+    op.add_column("users", sa.Column("external_subject", sa.String(255), nullable=True))
+    op.create_unique_constraint(
+        "uq_user_external_identity", "users", ["identity_provider", "external_subject"]
+    )
+
+
+def downgrade() -> None:
+    op.drop_constraint("uq_user_external_identity", "users", type_="unique")
+    op.drop_column("users", "external_subject")
+    op.drop_column("users", "identity_provider")

backend/alembic/versions/b7c7b536df85_operational_user_credentials.py

@@ -0,0 +1,25 @@
+"""operational user credentials
+
+Revision ID: b7c7b536df85
+Revises: 799d8800e241
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "b7c7b536df85"
+down_revision = "799d8800e241"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("users", sa.Column("email", sa.String(length=320), nullable=True))
+    op.add_column("users", sa.Column("password_hash", sa.String(length=512), nullable=True))
+    op.create_unique_constraint("uq_users_email", "users", ["email"])
+
+
+def downgrade() -> None:
+    op.drop_constraint("uq_users_email", "users", type_="unique")
+    op.drop_column("users", "password_hash")
+    op.drop_column("users", "email")

backend/alembic/versions/b913a72e8c14_customer_privacy_state.py

@@ -0,0 +1,24 @@
+"""add explicit customer anonymisation state
+
+Revision ID: b913a72e8c14
+Revises: a81d0ce9f662
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "b913a72e8c14"
+down_revision = "a81d0ce9f662"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("customers", sa.Column("anonymized_at", sa.DateTime(timezone=True)))
+    op.create_index("ix_customers_anonymized_at", "customers", ["anonymized_at"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_customers_anonymized_at", table_name="customers")
+    op.drop_column("customers", "anonymized_at")

backend/alembic/versions/c24f6a9d013e_domain_constraints_indexes.py

@@ -0,0 +1,68 @@
+"""add domain constraints and operational indexes
+
+Revision ID: c24f6a9d013e
+Revises: b913a72e8c14
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "c24f6a9d013e"
+down_revision = "b913a72e8c14"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    checks = (
+        ("bookings", "ck_bookings_status", "status IN ('reserved','active','returned','cancelled','blocked')"),
+        ("bookings", "ck_bookings_time_window", "ends_at > starts_at"),
+        ("bookings", "ck_bookings_start_odometer", "start_odometer_km IS NULL OR start_odometer_km >= 0"),
+        ("bookings", "ck_bookings_end_odometer", "end_odometer_km IS NULL OR end_odometer_km >= 0"),
+        ("vehicles", "ck_vehicles_operational_status", "operational_status IN ('available','rented','cleaning','maintenance','blocked')"),
+        ("vehicles", "ck_vehicles_model_year", "model_year BETWEEN 1900 AND 2100"),
+        ("vehicles", "ck_vehicles_odometer", "odometer_km >= 0"),
+        ("vehicles", "ck_vehicles_next_service", "next_service_km >= 0"),
+        ("vehicles", "ck_vehicles_version", "version >= 1"),
+        ("data_quality_issues", "ck_data_quality_rule_type", "rule_type IN ('possible_duplicate_customer','missing_required_field','odometer_regression','booking_overlap','vehicle_status_conflict')"),
+        ("data_quality_issues", "ck_data_quality_severity", "severity IN ('low','medium','high')"),
+        ("data_quality_issues", "ck_data_quality_status", "status IN ('open','deferred','resolved','rejected')"),
+        ("outbox_events", "ck_outbox_delivery_status", "delivery_status IN ('pending','delivering','succeeded','failed')"),
+        ("outbox_events", "ck_outbox_attempts", "attempts >= 0"),
+        ("audit_events", "ck_audit_actor_type", "actor_type IN ('user','service','system')"),
+    )
+    for table, name, condition in checks:
+        op.create_check_constraint(name, table, condition)
+
+    op.create_index("ix_bookings_vehicle_status_window", "bookings", ["vehicle_id", "status", "starts_at", "ends_at"])
+    op.create_index("ix_data_quality_work_queue", "data_quality_issues", ["status", "due_at", "severity"])
+    op.create_index("ix_outbox_delivery_next_attempt", "outbox_events", ["delivery_status", "next_attempt_at"])
+    op.create_index("ix_audit_action_occurred", "audit_events", ["action", "occurred_at"])
+    op.create_index("ix_audit_entity", "audit_events", ["entity_type", "entity_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_audit_entity", table_name="audit_events")
+    op.drop_index("ix_audit_action_occurred", table_name="audit_events")
+    op.drop_index("ix_outbox_delivery_next_attempt", table_name="outbox_events")
+    op.drop_index("ix_data_quality_work_queue", table_name="data_quality_issues")
+    op.drop_index("ix_bookings_vehicle_status_window", table_name="bookings")
+    for table, name in (
+        ("audit_events", "ck_audit_actor_type"),
+        ("outbox_events", "ck_outbox_attempts"),
+        ("outbox_events", "ck_outbox_delivery_status"),
+        ("data_quality_issues", "ck_data_quality_status"),
+        ("data_quality_issues", "ck_data_quality_severity"),
+        ("data_quality_issues", "ck_data_quality_rule_type"),
+        ("vehicles", "ck_vehicles_version"),
+        ("vehicles", "ck_vehicles_next_service"),
+        ("vehicles", "ck_vehicles_odometer"),
+        ("vehicles", "ck_vehicles_model_year"),
+        ("vehicles", "ck_vehicles_operational_status"),
+        ("bookings", "ck_bookings_end_odometer"),
+        ("bookings", "ck_bookings_start_odometer"),
+        ("bookings", "ck_bookings_time_window"),
+        ("bookings", "ck_bookings_status"),
+    ):
+        op.drop_constraint(name, table, type_="check")

backend/alembic/versions/d1f83bc64170_revoked_sessions.py

@@ -0,0 +1,49 @@
+"""persist revoked sessions
+
+Revision ID: d1f83bc64170
+Revises: b7c7b536df85
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "d1f83bc64170"
+down_revision = "b7c7b536df85"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "revoked_sessions",
+        sa.Column("token_hash", sa.String(length=64), nullable=False),
+        sa.Column("expires_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("id", sa.Uuid(), nullable=False),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.PrimaryKeyConstraint("id"),
+    )
+    op.create_index("ix_revoked_sessions_expires_at", "revoked_sessions", ["expires_at"])
+    op.create_index(
+        "ix_revoked_sessions_token_hash",
+        "revoked_sessions",
+        ["token_hash"],
+        unique=True,
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("ix_revoked_sessions_token_hash", table_name="revoked_sessions")
+    op.drop_index("ix_revoked_sessions_expires_at", table_name="revoked_sessions")
+    op.drop_table("revoked_sessions")

backend/alembic/versions/f43d829ab610_quality_work_queue.py

@@ -0,0 +1,65 @@
+"""add ownership and SLA deadlines to data quality issues
+
+Revision ID: f43d829ab610
+Revises: d1f83bc64170
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "f43d829ab610"
+down_revision = "d1f83bc64170"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "data_quality_issues",
+        sa.Column("due_at", sa.DateTime(timezone=True), nullable=True),
+    )
+    op.add_column(
+        "data_quality_issues",
+        sa.Column("assigned_to_user_id", sa.Uuid(), nullable=True),
+    )
+    op.create_foreign_key(
+        "fk_data_quality_issues_assigned_user",
+        "data_quality_issues",
+        "users",
+        ["assigned_to_user_id"],
+        ["id"],
+        ondelete="SET NULL",
+    )
+    op.create_index("ix_data_quality_issues_due_at", "data_quality_issues", ["due_at"])
+    op.create_index(
+        "ix_data_quality_issues_assigned_to_user_id",
+        "data_quality_issues",
+        ["assigned_to_user_id"],
+    )
+    op.execute(
+        """
+        UPDATE data_quality_issues
+        SET due_at = detected_at + CASE severity
+          WHEN 'high' THEN interval '4 hours'
+          WHEN 'low' THEN interval '3 days'
+          ELSE interval '1 day'
+        END
+        WHERE status = 'open' AND due_at IS NULL
+        """
+    )
+
+
+def downgrade() -> None:
+    op.drop_index(
+        "ix_data_quality_issues_assigned_to_user_id",
+        table_name="data_quality_issues",
+    )
+    op.drop_index("ix_data_quality_issues_due_at", table_name="data_quality_issues")
+    op.drop_constraint(
+        "fk_data_quality_issues_assigned_user",
+        "data_quality_issues",
+        type_="foreignkey",
+    )
+    op.drop_column("data_quality_issues", "assigned_to_user_id")
+    op.drop_column("data_quality_issues", "due_at")

backend/app/api/deps.py

@@ -1,6 +1,9 @@
 from __future__ import annotations
 
+import hmac
+import re
 from collections.abc import Generator
+from dataclasses import dataclass
 
 from fastapi import Depends, Header, HTTPException, Request, status
 from sqlalchemy.orm import Session
@@ -8,7 +11,9 @@ from sqlalchemy.orm import Session
 from app.core.config import get_settings
 from app.core.db import SessionLocal
 from app.core.security import SessionPayload, read_session_token
+from app.models.user import User
 from app.schemas import CurrentUser, Role
+from app.services.sessions import is_session_revoked
 
 settings = get_settings()
 _VALID_ROLES = frozenset(Role.__args__)  # type: ignore[attr-defined]
@@ -22,13 +27,32 @@ def get_db() -> Generator[Session, None, None]:
         db.close()
 
 
-def get_current_user(request: Request) -> CurrentUser:
+def get_current_user(request: Request, db: Session = Depends(get_db)) -> CurrentUser:
     token = request.cookies.get(settings.session_cookie_name)
     payload: SessionPayload | None = read_session_token(token) if token else None
-    if payload is None or payload.role not in _VALID_ROLES:
+    revoked = token is not None and is_session_revoked(db, token)
+    if payload is None or payload.role not in _VALID_ROLES or revoked:
         raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail="Not authenticated")
-    role: Role = payload.role  # type: ignore[assignment]
+    # Demo reset deliberately rebuilds the deterministic users table. Retaining the
-    return CurrentUser(public_ref=payload.public_ref, display_name=payload.display_name, role=role)
+    # signed demo session until the reset endpoint clears its cookie keeps existing demo
+    # workflows stable; operational sessions are always checked against the live record.
+    if settings.mobilityops_demo_mode:
+        demo_role: Role = payload.role  # type: ignore[assignment]
+        return CurrentUser(
+            public_ref=payload.public_ref,
+            display_name=payload.display_name,
+            role=demo_role,
+        )
+    user = db.get(User, payload.user_id)
+    if (
+        user is None
+        or not user.active
+        or user.public_ref != payload.public_ref
+        or user.role not in _VALID_ROLES
+    ):
+        raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail="Not authenticated")
+    role: Role = user.role  # type: ignore[assignment]
+    return CurrentUser(public_ref=user.public_ref, display_name=user.display_name, role=role)
 
 
 def require_operations_manager(
@@ -41,12 +65,30 @@ def require_operations_manager(
     return user
 
 
+@dataclass(frozen=True)
+class McpClientContext:
+    reported_client_id: str
+    tenant: str
+
+
+_MCP_CLIENT_ID = re.compile(
+    r"^itworx-mcp-hub:(?:readiness|mobilityops:[A-Za-z0-9][A-Za-z0-9._:-]{0,127})$"
+)
+
+
 def require_mcp_service_token(
     x_service_token: str = Header(..., alias="X-Service-Token"),
-    x_client_id: str = Header(default="unknown-mcp-client", alias="X-Client-Id"),
+    x_client_id: str = Header(..., alias="X-Client-Id", min_length=1, max_length=180),
-) -> str:
+    x_tenant_id: str | None = Header(default=None, alias="X-Tenant-Id", max_length=120),
-    if x_service_token != settings.mcp_hub_service_token:
+) -> McpClientContext:
+    if not hmac.compare_digest(x_service_token, settings.mcp_hub_service_token):
         raise HTTPException(
             status_code=status.HTTP_401_UNAUTHORIZED, detail="Invalid service token"
         )
-    return x_client_id
+    if not _MCP_CLIENT_ID.fullmatch(x_client_id):
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN, detail="Untrusted MCP client identity"
+        )
+    if x_tenant_id is not None and x_tenant_id != settings.ragcore_tenant:
+        raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail="Tenant mismatch")
+    return McpClientContext(reported_client_id=x_client_id, tenant=settings.ragcore_tenant)

backend/app/api/routers/audit.py

@@ -1,22 +1,30 @@
 from __future__ import annotations
 
+import csv
+import io
+import json
 import uuid
 from collections.abc import Sequence
+from datetime import UTC, datetime, timedelta
 from typing import Any
 
-from fastapi import APIRouter, Depends, Query
+from fastapi import APIRouter, Depends, HTTPException, Query
-from sqlalchemy import select
+from fastapi.responses import Response
+from sqlalchemy import func, select
 from sqlalchemy.orm import Session
 
 from app.api.deps import get_db, require_operations_manager
+from app.core.config import get_settings
 from app.models.audit import AuditEvent
 from app.models.booking import Booking
 from app.models.customer import Customer
 from app.models.data_quality import DataQualityIssue
 from app.models.vehicle import Vehicle
-from app.schemas import AuditEventOut, CurrentUser
+from app.schemas import AuditEventOut, AuditEventPageOut, CurrentUser
+from app.services.audit import record_audit_event
 
 router = APIRouter(prefix="/api/v1/audit", tags=["audit"])
+settings = get_settings()
 
 # Only entity types with a stable public reference and (optionally) a real frontend route
 # are resolved here. Types like "system", "knowledge" or "mcp_tool" carry no linkable
@@ -36,6 +44,81 @@ _ROUTE_TEMPLATES: dict[str, str] = {
 }
 
 
+def _as_utc(value: datetime | None) -> datetime | None:
+    """Treat naive query datetimes as UTC so they compare safely with aware values."""
+    if value is None:
+        return None
+    return value if value.tzinfo is not None else value.replace(tzinfo=UTC)
+
+
+@router.get("/export.csv")
+def export_audit_csv(
+    occurred_from: datetime | None = Query(default=None),
+    occurred_to: datetime | None = Query(default=None),
+    db: Session = Depends(get_db),
+    actor: CurrentUser = Depends(require_operations_manager),
+) -> Response:
+    end = _as_utc(occurred_to) or datetime.now(UTC)
+    start = _as_utc(occurred_from) or end - timedelta(days=30)
+    if end <= start or end - start > timedelta(days=90):
+        raise HTTPException(status_code=422, detail="Audit export range must be 1 to 90 days")
+    events = db.scalars(
+        select(AuditEvent)
+        .where(AuditEvent.occurred_at >= start, AuditEvent.occurred_at <= end)
+        .order_by(AuditEvent.occurred_at)
+        .limit(settings.privacy_audit_export_max_rows + 1)
+    ).all()
+    if len(events) > settings.privacy_audit_export_max_rows:
+        raise HTTPException(status_code=413, detail="Audit export exceeds configured row limit")
+    output = io.StringIO(newline="")
+    writer = csv.writer(output)
+    writer.writerow(
+        (
+            "id",
+            "occurred_at",
+            "actor_type",
+            "actor_label",
+            "action",
+            "entity_type",
+            "entity_id",
+            "correlation_id",
+            "before",
+            "after",
+            "metadata",
+        )
+    )
+    for event in events:
+        writer.writerow(
+            (
+                event.id,
+                event.occurred_at.isoformat(),
+                event.actor_type,
+                event.actor_label,
+                event.action,
+                event.entity_type,
+                event.entity_id or "",
+                event.correlation_id,
+                json.dumps(event.before_json, separators=(",", ":"), default=str),
+                json.dumps(event.after_json, separators=(",", ":"), default=str),
+                json.dumps(event.metadata_json, separators=(",", ":"), default=str),
+            )
+        )
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=actor.display_name,
+        action="audit_exported",
+        entity_type="audit",
+        metadata={"from": start.isoformat(), "to": end.isoformat(), "rows": len(events)},
+    )
+    db.commit()
+    return Response(
+        output.getvalue(),
+        media_type="text/csv; charset=utf-8",
+        headers={"Content-Disposition": 'attachment; filename="mobilityops-audit.csv"'},
+    )
+
+
 def _resolve_entity_refs(db: Session, events: Sequence[AuditEvent]) -> dict[uuid.UUID, str]:
     ids_by_type: dict[str, set[uuid.UUID]] = {}
     for event in events:
@@ -51,26 +134,51 @@ def _resolve_entity_refs(db: Session, events: Sequence[AuditEvent]) -> dict[uuid
     return refs
 
 
-@router.get("", response_model=list[AuditEventOut])
+@router.get("", response_model=list[AuditEventOut] | AuditEventPageOut)
 def list_audit_events(
     actor_label: str | None = Query(default=None),
     action: str | None = Query(default=None),
     entity_type: str | None = Query(default=None),
-    correlation_id: str | None = Query(default=None),
+    entity_ref: str | None = Query(default=None, min_length=1, max_length=100),
-    limit: int = Query(default=100, le=500),
+    correlation_id: uuid.UUID | None = Query(default=None),
+    occurred_from: datetime | None = Query(default=None),
+    occurred_to: datetime | None = Query(default=None),
+    page: int | None = Query(default=None, ge=1),
+    page_size: int = Query(default=25, ge=1, le=25),
     db: Session = Depends(get_db),
     _user: CurrentUser = Depends(require_operations_manager),
-) -> list[AuditEventOut]:
+) -> list[AuditEventOut] | AuditEventPageOut:
-    stmt = select(AuditEvent).order_by(AuditEvent.occurred_at.desc()).limit(limit)
+    stmt = select(AuditEvent).order_by(AuditEvent.occurred_at.desc())
     if actor_label:
         stmt = stmt.where(AuditEvent.actor_label == actor_label)
     if action:
         stmt = stmt.where(AuditEvent.action == action)
     if entity_type:
         stmt = stmt.where(AuditEvent.entity_type == entity_type)
+    if entity_ref:
+        matched_ids: set[uuid.UUID] = set()
+        for model in _ENTITY_MODELS.values():
+            matched_ids.update(
+                db.scalars(
+                    select(model.id).where(model.public_ref.ilike(f"%{entity_ref.strip()}%"))
+                ).all()
+            )
+        if not matched_ids:
+            if page is None:
+                return []
+            return AuditEventPageOut(items=[], page=1, page_size=page_size, total=0, total_pages=1)
+        stmt = stmt.where(AuditEvent.entity_id.in_(matched_ids))
     if correlation_id:
         stmt = stmt.where(AuditEvent.correlation_id == correlation_id)
-    events = db.scalars(stmt).all()
+    if occurred_from:
+        stmt = stmt.where(AuditEvent.occurred_at >= occurred_from)
+    if occurred_to:
+        stmt = stmt.where(AuditEvent.occurred_at <= occurred_to)
+    total = db.scalar(select(func.count()).select_from(stmt.subquery())) or 0
+    page_number = page or 1
+    events = db.scalars(
+        stmt if page is None else stmt.offset((page_number - 1) * page_size).limit(page_size)
+    ).all()
     entity_refs = _resolve_entity_refs(db, events)
 
     out = []
@@ -94,4 +202,13 @@ def list_audit_events(
                 metadata=e.metadata_json,
             )
         )
+    if page is None:
         return out
+    total_pages = max(1, (total + page_size - 1) // page_size)
+    return AuditEventPageOut(
+        items=out,
+        page=min(page_number, total_pages),
+        page_size=page_size,
+        total=total,
+        total_pages=total_pages,
+    )

backend/app/api/routers/auth.py

@@ -0,0 +1,312 @@
+from __future__ import annotations
+
+import time
+import uuid
+
+from authlib.integrations.starlette_client import OAuth, OAuthError  # type: ignore[import-untyped]
+from fastapi import APIRouter, Depends, HTTPException, Request, Response, status
+from fastapi.responses import RedirectResponse
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from app.api.deps import get_current_user, get_db
+from app.core.config import get_settings
+from app.core.ratelimit import FailedAttemptLimiter
+from app.core.security import (
+    SessionPayload,
+    create_session_token,
+    hash_password,
+    read_session_token,
+    verify_password,
+)
+from app.models.user import User
+from app.schemas import CurrentUser, OidcStatusOut, PasswordLoginRequest
+from app.services.audit import record_audit_event
+from app.services.sessions import revoke_session
+
+router = APIRouter(prefix="/api/v1/auth", tags=["auth"])
+settings = get_settings()
+_login_limiter = (
+    FailedAttemptLimiter(
+        max_failures=settings.login_max_failures,
+        window_seconds=settings.login_failure_window_seconds,
+    )
+    if settings.login_max_failures > 0
+    else None
+)
+
+
+def _client_key(request: Request) -> str:
+    # The API sits behind the web container's reverse proxy in every documented
+    # deployment. The proxy appends/overwrites the socket peer as the final hop, so an
+    # attacker-controlled leading value must never select a fresh limiter bucket.
+    forwarded = request.headers.get("x-forwarded-for", "")
+    if forwarded:
+        return forwarded.split(",")[-1].strip()
+    return request.client.host if request.client else "unknown"
+
+
+oauth = OAuth()
+if settings.oidc_enabled and settings.oidc_issuer_url:
+    oauth.register(
+        name="oidc",
+        client_id=settings.oidc_client_id,
+        client_secret=settings.oidc_client_secret,
+        server_metadata_url=f"{settings.oidc_issuer_url.rstrip('/')}/.well-known/openid-configuration",
+        client_kwargs={"scope": "openid email profile"},
+    )
+
+
+def _current_user_out(user: User) -> CurrentUser:
+    return CurrentUser(
+        public_ref=user.public_ref,
+        display_name=user.display_name,
+        role=user.role,  # type: ignore[arg-type]
+    )
+
+
+def _set_session(response: Response, user: User) -> None:
+    token = create_session_token(
+        SessionPayload(
+            user_id=str(user.id),
+            public_ref=user.public_ref,
+            role=user.role,
+            display_name=user.display_name,
+            issued_at=int(time.time()),
+            session_id=str(uuid.uuid4()),
+        )
+    )
+    response.set_cookie(
+        settings.session_cookie_name,
+        token,
+        httponly=True,
+        samesite="lax",
+        secure=settings.session_cookie_secure,
+        max_age=settings.session_ttl_seconds,
+    )
+
+
+def bootstrap_initial_admin(db: Session) -> None:
+    """Create or rotate the explicitly configured first manager in operational mode."""
+    if (
+        settings.mobilityops_demo_mode
+        or not settings.initial_admin_email
+        or not settings.initial_admin_password
+    ):
+        return
+    email = settings.initial_admin_email.strip().lower()
+    user = db.scalar(select(User).where(User.email == email))
+    if user is None:
+        user = User(
+            public_ref="USR-ADMIN",
+            email=email,
+            password_hash=hash_password(settings.initial_admin_password),
+            display_name=settings.initial_admin_display_name,
+            role="operations_manager",
+            active=True,
+        )
+        db.add(user)
+        record_audit_event(
+            db,
+            actor_type="system",
+            actor_label="bootstrap",
+            action="operational_admin_created",
+            entity_type="user",
+            entity_id=user.id,
+        )
+        db.commit()
+
+
+def _oidc_configured() -> bool:
+    return bool(
+        settings.oidc_enabled
+        and settings.oidc_issuer_url
+        and settings.oidc_client_id
+        and settings.oidc_client_secret
+    )
+
+
+def _oidc_redirect_uri() -> str:
+    return settings.oidc_redirect_uri or (
+        f"{settings.mobilityops_public_url.rstrip('/')}/api/v1/auth/oidc/callback"
+    )
+
+
+def _allowed_oidc_email(email: str) -> bool:
+    domains = {
+        value.strip().casefold()
+        for value in settings.oidc_allowed_email_domains.split(",")
+        if value.strip()
+    }
+    return not domains or email.rsplit("@", 1)[-1].casefold() in domains
+
+
+def _resolve_oidc_user(db: Session, claims: dict[str, object]) -> User:
+    subject = str(claims.get("sub") or "").strip()
+    email = str(claims.get("email") or "").strip().lower()
+    if not subject or not email or claims.get("email_verified") is not True:
+        raise HTTPException(status_code=401, detail="Verified OIDC email and subject are required")
+    if not _allowed_oidc_email(email):
+        raise HTTPException(status_code=403, detail="Email domain is not allowed")
+
+    provider = settings.oidc_issuer_url.rstrip("/")
+    user = db.scalar(
+        select(User).where(
+            User.identity_provider == provider,
+            User.external_subject == subject,
+        )
+    )
+    if user is None:
+        user = db.scalar(select(User).where(User.email == email))
+        if user is not None and user.external_subject not in (None, subject):
+            raise HTTPException(status_code=409, detail="Email is linked to another identity")
+        if user is not None and claims.get("email_verified") is not True:
+            # Linking an existing local account (possibly the bootstrap admin) purely on an
+            # email match requires the IdP to explicitly assert the address is verified;
+            # an absent claim is treated as unverified.
+            raise HTTPException(status_code=401, detail="Verified OIDC email is required")
+    created = user is None
+    if created:
+        if not settings.oidc_auto_provision:
+            raise HTTPException(status_code=403, detail="OIDC user is not provisioned")
+        role = settings.oidc_default_role
+        if role not in {"operations_manager", "rental_employee"}:
+            role = "rental_employee"
+        user = User(
+            public_ref=f"USR-{uuid.uuid4().hex[:8].upper()}",
+            email=email,
+            password_hash=None,
+            display_name=str(claims.get("name") or email),
+            role=role,
+            active=True,
+        )
+        db.add(user)
+        db.flush()
+    assert user is not None
+    if not user.active:
+        raise HTTPException(status_code=403, detail="User is inactive")
+    user.identity_provider = provider
+    user.external_subject = subject
+    record_audit_event(
+        db,
+        actor_type="system" if created else "user",
+        actor_id=None if created else user.id,
+        actor_label=settings.oidc_provider_name,
+        action="oidc_user_provisioned" if created else "oidc_identity_linked",
+        entity_type="user",
+        entity_id=user.id,
+        metadata={"provider": provider},
+    )
+    return user
+
+
+@router.get("/oidc/status", response_model=OidcStatusOut)
+def oidc_status() -> OidcStatusOut:
+    return OidcStatusOut(
+        enabled=_oidc_configured(),
+        provider_name=settings.oidc_provider_name if _oidc_configured() else None,
+    )
+
+
+@router.get("/oidc/login")
+async def oidc_login(request: Request) -> Response:
+    if not _oidc_configured():
+        raise HTTPException(status_code=404, detail="OIDC login is not configured")
+    client = oauth.create_client("oidc")
+    if client is None:
+        raise HTTPException(status_code=503, detail="OIDC client is unavailable")
+    return await client.authorize_redirect(request, _oidc_redirect_uri())
+
+
+@router.get("/oidc/callback")
+async def oidc_callback(request: Request, db: Session = Depends(get_db)) -> Response:
+    if not _oidc_configured():
+        raise HTTPException(status_code=404, detail="OIDC login is not configured")
+    client = oauth.create_client("oidc")
+    if client is None:
+        raise HTTPException(status_code=503, detail="OIDC client is unavailable")
+    try:
+        token = await client.authorize_access_token(request)
+    except OAuthError as exc:
+        raise HTTPException(status_code=401, detail="OIDC authentication failed") from exc
+    user = _resolve_oidc_user(db, dict(token.get("userinfo") or {}))
+    response = RedirectResponse(f"{settings.mobilityops_public_url.rstrip('/')}/dashboard")
+    _set_session(response, user)
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_id=user.id,
+        actor_label=user.display_name,
+        action="oidc_login",
+        entity_type="user",
+        entity_id=user.id,
+        metadata={"provider": settings.oidc_issuer_url.rstrip("/")},
+    )
+    db.commit()
+    return response
+
+
+@router.post("/login", response_model=CurrentUser)
+def password_login(
+    body: PasswordLoginRequest,
+    request: Request,
+    response: Response,
+    db: Session = Depends(get_db),
+) -> CurrentUser:
+    if settings.mobilityops_demo_mode:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="Password login is unavailable in demo mode",
+        )
+    limiter_key = _client_key(request)
+    retry_after = _login_limiter.retry_after_seconds(limiter_key) if _login_limiter else 0
+    if retry_after:
+        raise HTTPException(
+            status_code=status.HTTP_429_TOO_MANY_REQUESTS,
+            detail="Too many failed login attempts. Try again later.",
+            headers={"Retry-After": str(retry_after)},
+        )
+    user = db.scalar(select(User).where(User.email == body.email.strip().lower()))
+    if user is None or not user.active or not verify_password(body.password, user.password_hash):
+        if _login_limiter:
+            _login_limiter.record_failure(limiter_key)
+        raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail="Invalid credentials")
+    if _login_limiter:
+        _login_limiter.reset(limiter_key)
+    _set_session(response, user)
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_id=user.id,
+        actor_label=user.display_name,
+        action="password_login",
+        entity_type="user",
+        entity_id=user.id,
+    )
+    db.commit()
+    return _current_user_out(user)
+
+
+@router.get("/session", response_model=CurrentUser)
+def get_session(response: Response, user: CurrentUser = Depends(get_current_user)) -> CurrentUser:
+    response.headers["Cache-Control"] = "no-store"
+    return user
+
+
+@router.post("/logout")
+def logout(request: Request, response: Response, db: Session = Depends(get_db)) -> dict:
+    token = request.cookies.get(settings.session_cookie_name)
+    payload = read_session_token(token) if token else None
+    if payload is not None and token is not None:
+        revoke_session(db, token, payload)
+        record_audit_event(
+            db,
+            actor_type="user",
+            actor_id=uuid.UUID(payload.user_id),
+            actor_label=payload.display_name,
+            action="logout",
+            entity_type="user",
+        )
+        db.commit()
+    response.delete_cookie(settings.session_cookie_name)
+    return {"status": "logged_out"}

backend/app/api/routers/bookings.py

@@ -1,20 +1,35 @@
 from __future__ import annotations
 
+import uuid
+from datetime import UTC, datetime
+from typing import Literal
+
 from fastapi import APIRouter, Depends, Header, HTTPException, Query, Response
-from sqlalchemy import select
+from sqlalchemy import case, func, or_, select
 from sqlalchemy.orm import Session
 
 from app.api.deps import get_current_user, get_db
 from app.models.booking import Booking
 from app.models.customer import Customer
+from app.models.inspection import Inspection
 from app.models.vehicle import Vehicle
 from app.schemas import (
+    AvailableVehicleOut,
     BookingOut,
+    BookingPageOut,
+    CancelBookingRequest,
+    CheckoutBookingRequest,
+    CheckoutBookingResult,
+    CompleteBookingRequirementsRequest,
+    CreateBookingRequest,
     CurrentUser,
     NextBookingRisk,
     RegisterReturnRequest,
+    RegisterReturnResult,
+    RescheduleBookingRequest,
     ReturnPreviewResult,
 )
+from app.services.audit import record_audit_event
 from app.services.returns import preview_vehicle_return, register_vehicle_return
 
 router = APIRouter(prefix="/api/v1/bookings", tags=["bookings"])
@@ -35,25 +50,292 @@ def _to_out(booking: Booking, customer: Customer, vehicle: Vehicle) -> BookingOu
     )
 
 
-@router.get("", response_model=list[BookingOut])
+@router.get("", response_model=list[BookingOut] | BookingPageOut)
 def list_bookings(
     status: str | None = Query(default=None),
     vehicle_ref: str | None = Query(default=None),
+    query: str | None = Query(default=None, min_length=1, max_length=100),
+    starts_from: datetime | None = Query(default=None),
+    starts_to: datetime | None = Query(default=None),
+    location: str | None = Query(default=None, min_length=1, max_length=120),
+    sort: Literal["operational", "starts_asc", "starts_desc"] = Query(default="operational"),
+    page: int | None = Query(default=None, ge=1),
+    page_size: int = Query(default=25, ge=1, le=25),
     db: Session = Depends(get_db),
     _user: CurrentUser = Depends(get_current_user),
-) -> list[BookingOut]:
+) -> list[BookingOut] | BookingPageOut:
-    stmt = select(Booking).order_by(Booking.starts_at.desc())
+    stmt = select(Booking)
     if status:
         stmt = stmt.where(Booking.status == status)
     if vehicle_ref:
         vehicle = db.scalar(select(Vehicle).where(Vehicle.public_ref == vehicle_ref))
         if vehicle is None:
+            if page is None:
                 return []
+            return BookingPageOut(items=[], page=1, page_size=page_size, total=0, total_pages=1)
         stmt = stmt.where(Booking.vehicle_id == vehicle.id)
-    bookings = db.scalars(stmt).all()
+    if starts_from:
-    customers = {c.id: c for c in db.scalars(select(Customer)).all()}
+        stmt = stmt.where(Booking.ends_at >= starts_from)
-    vehicles = {v.id: v for v in db.scalars(select(Vehicle)).all()}
+    if starts_to:
-    return [_to_out(b, customers[b.customer_id], vehicles[b.vehicle_id]) for b in bookings]
+        stmt = stmt.where(Booking.starts_at < starts_to)
+    if query or location:
+        stmt = stmt.join(Customer, Booking.customer_id == Customer.id).join(
+            Vehicle, Booking.vehicle_id == Vehicle.id
+        )
+    if location:
+        stmt = stmt.where(Vehicle.location.ilike(location.strip()))
+    if query:
+        term = f"%{query.strip()}%"
+        stmt = stmt.where(
+            or_(
+                Booking.public_ref.ilike(term),
+                Customer.first_name.ilike(term),
+                Customer.last_name.ilike(term),
+                Vehicle.public_ref.ilike(term),
+            )
+        )
+    if sort == "starts_asc":
+        stmt = stmt.order_by(Booking.starts_at.asc())
+    elif sort == "starts_desc":
+        stmt = stmt.order_by(Booking.starts_at.desc())
+    else:
+        now = datetime.now(UTC)
+        operational_bucket = case(
+            (Booking.status == "active", 0),
+            (Booking.starts_at >= now, 1),
+            else_=2,
+        )
+        stmt = stmt.order_by(
+            operational_bucket,
+            case((Booking.starts_at >= now, Booking.starts_at)).asc().nulls_last(),
+            Booking.starts_at.desc(),
+        )
+    total = db.scalar(select(func.count()).select_from(stmt.subquery())) or 0
+    page_number = page or 1
+    bookings = db.scalars(
+        stmt if page is None else stmt.offset((page_number - 1) * page_size).limit(page_size)
+    ).all()
+    customer_ids = {booking.customer_id for booking in bookings}
+    vehicle_ids = {booking.vehicle_id for booking in bookings}
+    customers = {
+        customer.id: customer
+        for customer in db.scalars(select(Customer).where(Customer.id.in_(customer_ids))).all()
+    }
+    vehicles = {
+        vehicle.id: vehicle
+        for vehicle in db.scalars(select(Vehicle).where(Vehicle.id.in_(vehicle_ids))).all()
+    }
+    items = [_to_out(b, customers[b.customer_id], vehicles[b.vehicle_id]) for b in bookings]
+    if page is None:
+        return items
+    total_pages = max(1, (total + page_size - 1) // page_size)
+    return BookingPageOut(
+        items=items,
+        page=min(page_number, total_pages),
+        page_size=page_size,
+        total=total,
+        total_pages=total_pages,
+    )
+
+
+@router.post("", response_model=BookingOut, status_code=201)
+def create_booking(
+    body: CreateBookingRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(get_current_user),
+) -> BookingOut:
+    if body.ends_at <= body.starts_at:
+        raise HTTPException(status_code=422, detail="Booking end must be after its start")
+    customer = db.scalar(select(Customer).where(Customer.public_ref == body.customer_ref))
+    if customer is None or customer.merged_into_customer_id is not None:
+        raise HTTPException(status_code=422, detail="Customer is unavailable for booking")
+    # Serialise booking creation per vehicle. The overlap check must run after
+    # acquiring this lock, otherwise two concurrent requests can both pass it.
+    vehicle = db.scalar(
+        select(Vehicle).where(Vehicle.public_ref == body.vehicle_ref).with_for_update()
+    )
+    if (
+        vehicle is None
+        or not vehicle.active
+        or vehicle.operational_status in {"maintenance", "blocked"}
+    ):
+        raise HTTPException(status_code=422, detail="Vehicle is unavailable for booking")
+    overlap = db.scalar(
+        select(Booking.id).where(
+            Booking.vehicle_id == vehicle.id,
+            Booking.status.in_(("reserved", "active")),
+            Booking.starts_at < body.ends_at,
+            Booking.ends_at > body.starts_at,
+        )
+    )
+    if overlap is not None:
+        raise HTTPException(status_code=409, detail="Vehicle already has an overlapping booking")
+    booking = Booking(
+        public_ref=f"BK-{uuid.uuid4().hex[:10].upper()}",
+        customer_id=customer.id,
+        vehicle_id=vehicle.id,
+        starts_at=body.starts_at,
+        ends_at=body.ends_at,
+        status="reserved",
+        start_odometer_km=None,
+        end_odometer_km=None,
+        requirements_complete=body.requirements_complete,
+    )
+    db.add(booking)
+    db.flush()
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=user.display_name,
+        action="booking_created",
+        entity_type="booking",
+        entity_id=booking.id,
+        after={"public_ref": booking.public_ref, "vehicle_ref": vehicle.public_ref},
+    )
+    db.commit()
+    return _to_out(booking, customer, vehicle)
+
+
+@router.post("/{public_ref}/checkout", response_model=CheckoutBookingResult)
+def checkout_booking(
+    public_ref: str,
+    body: CheckoutBookingRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(get_current_user),
+) -> CheckoutBookingResult:
+    booking = db.scalar(select(Booking).where(Booking.public_ref == public_ref).with_for_update())
+    if booking is None:
+        raise HTTPException(status_code=404, detail="Booking not found")
+    if booking.status != "reserved":
+        raise HTTPException(status_code=409, detail="Only a reserved booking can be checked out")
+    if not booking.requirements_complete:
+        raise HTTPException(status_code=409, detail="Booking requirements are incomplete")
+    vehicle = db.scalar(select(Vehicle).where(Vehicle.id == booking.vehicle_id).with_for_update())
+    if vehicle is None:
+        raise HTTPException(status_code=500, detail="Booking references a missing vehicle")
+    if not vehicle.active or vehicle.operational_status in {"maintenance", "blocked", "rented"}:
+        raise HTTPException(status_code=409, detail="Vehicle is not ready for checkout")
+    active_conflict = db.scalar(
+        select(Booking.id).where(
+            Booking.vehicle_id == vehicle.id,
+            Booking.status == "active",
+            Booking.id != booking.id,
+        )
+    )
+    if active_conflict is not None:
+        raise HTTPException(status_code=409, detail="Vehicle already has an active booking")
+
+    attention_reasons: list[str] = []
+    if body.start_odometer_km < vehicle.odometer_km:
+        attention_reasons.append("odometer_regression")
+    if not body.cleanliness_ok:
+        attention_reasons.append("cleanliness")
+    if body.damage_reported:
+        attention_reasons.append("damage")
+    if body.technical_warning:
+        attention_reasons.append("technical_warning")
+
+    inspection = Inspection(
+        public_ref=f"INSP-{uuid.uuid4().hex[:10].upper()}",
+        booking_id=booking.id,
+        vehicle_id=vehicle.id,
+        type="checkout",
+        fuel_level_percent=body.fuel_level_percent,
+        cleanliness_ok=body.cleanliness_ok,
+        damage_reported=body.damage_reported,
+        technical_warning=body.technical_warning,
+        notes=body.notes,
+        odometer_km=body.start_odometer_km,
+        completed_at=datetime.now(UTC),
+        completed_by=user.display_name,
+    )
+    db.add(inspection)
+    if attention_reasons:
+        booking.status = "blocked"
+        vehicle.operational_status = (
+            "maintenance" if body.damage_reported or body.technical_warning else "cleaning"
+        )
+    else:
+        booking.status = "active"
+        booking.start_odometer_km = body.start_odometer_km
+        vehicle.odometer_km = max(vehicle.odometer_km, body.start_odometer_km)
+        vehicle.operational_status = "rented"
+    vehicle.version += 1
+    db.flush()
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=user.display_name,
+        action="booking_checkout_recorded",
+        entity_type="booking",
+        entity_id=booking.id,
+        after={
+            "inspection_ref": inspection.public_ref,
+            "booking_status": booking.status,
+            "vehicle_status": vehicle.operational_status,
+            "attention_reasons": attention_reasons,
+        },
+    )
+    db.commit()
+    return CheckoutBookingResult(
+        booking_ref=booking.public_ref,
+        vehicle_ref=vehicle.public_ref,
+        inspection_ref=inspection.public_ref,
+        booking_status=booking.status,
+        resulting_vehicle_status=vehicle.operational_status,
+        activated=booking.status == "active",
+        attention_reasons=attention_reasons,
+    )
+
+
+@router.get("/availability", response_model=list[AvailableVehicleOut])
+def list_available_vehicles(
+    starts_at: datetime,
+    ends_at: datetime,
+    query: str | None = Query(default=None, max_length=100),
+    limit: int = Query(default=25, ge=1, le=50),
+    db: Session = Depends(get_db),
+    _user: CurrentUser = Depends(get_current_user),
+) -> list[AvailableVehicleOut]:
+    if ends_at <= starts_at:
+        raise HTTPException(status_code=422, detail="Booking end must be after its start")
+    overlapping_vehicle_ids = select(Booking.vehicle_id).where(
+        Booking.status.in_(("reserved", "active")),
+        Booking.starts_at < ends_at,
+        Booking.ends_at > starts_at,
+    )
+    stmt = (
+        select(Vehicle)
+        .where(
+            Vehicle.active.is_(True),
+            Vehicle.operational_status.not_in(("maintenance", "blocked")),
+            Vehicle.id.not_in(overlapping_vehicle_ids),
+        )
+        .order_by(Vehicle.location, Vehicle.public_ref)
+        .limit(limit)
+    )
+    if query and query.strip():
+        term = f"%{query.strip()}%"
+        stmt = stmt.where(
+            or_(
+                Vehicle.public_ref.ilike(term),
+                Vehicle.make.ilike(term),
+                Vehicle.model.ilike(term),
+                Vehicle.registration_number.ilike(term),
+                Vehicle.location.ilike(term),
+            )
+        )
+    return [
+        AvailableVehicleOut(
+            public_ref=vehicle.public_ref,
+            make=vehicle.make,
+            model=vehicle.model,
+            registration_number=vehicle.registration_number,
+            location=vehicle.location,
+            operational_status=vehicle.operational_status,
+        )
+        for vehicle in db.scalars(stmt).all()
+    ]
 
 
 @router.get("/{public_ref}", response_model=BookingOut)
@@ -72,6 +354,126 @@ def get_booking(
     return _to_out(booking, customer, vehicle)
 
 
+@router.post("/{public_ref}/complete-requirements", response_model=BookingOut)
+def complete_booking_requirements(
+    public_ref: str,
+    body: CompleteBookingRequirementsRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(get_current_user),
+) -> BookingOut:
+    booking = db.scalar(select(Booking).where(Booking.public_ref == public_ref).with_for_update())
+    if booking is None:
+        raise HTTPException(status_code=404, detail="Booking not found")
+    if booking.status != "reserved":
+        raise HTTPException(
+            status_code=409,
+            detail="Requirements can only be confirmed for a reserved booking",
+        )
+    customer = db.get(Customer, booking.customer_id)
+    vehicle = db.get(Vehicle, booking.vehicle_id)
+    if customer is None or vehicle is None:
+        raise HTTPException(status_code=500, detail="Booking references a missing record")
+    if not booking.requirements_complete:
+        booking.requirements_complete = True
+        record_audit_event(
+            db,
+            actor_type="user",
+            actor_label=user.display_name,
+            action="booking_requirements_completed",
+            entity_type="booking",
+            entity_id=booking.id,
+            before={"requirements_complete": False},
+            after={"requirements_complete": True},
+            metadata={"confirmation": body.confirmation.strip()},
+        )
+        db.commit()
+    return _to_out(booking, customer, vehicle)
+
+
+@router.patch("/{public_ref}/schedule", response_model=BookingOut)
+def reschedule_booking(
+    public_ref: str,
+    body: RescheduleBookingRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(get_current_user),
+) -> BookingOut:
+    if body.ends_at <= body.starts_at:
+        raise HTTPException(status_code=422, detail="Booking end must be after its start")
+    booking = db.scalar(select(Booking).where(Booking.public_ref == public_ref).with_for_update())
+    if booking is None:
+        raise HTTPException(status_code=404, detail="Booking not found")
+    if booking.status != "reserved":
+        raise HTTPException(status_code=409, detail="Only a reserved booking can be rescheduled")
+    vehicle = db.scalar(select(Vehicle).where(Vehicle.id == booking.vehicle_id).with_for_update())
+    customer = db.get(Customer, booking.customer_id)
+    if customer is None or vehicle is None:
+        raise HTTPException(status_code=500, detail="Booking references a missing record")
+    overlap = db.scalar(
+        select(Booking.id).where(
+            Booking.vehicle_id == booking.vehicle_id,
+            Booking.id != booking.id,
+            Booking.status.in_(("reserved", "active")),
+            Booking.starts_at < body.ends_at,
+            Booking.ends_at > body.starts_at,
+        )
+    )
+    if overlap is not None:
+        raise HTTPException(status_code=409, detail="Vehicle already has an overlapping booking")
+    before = {"starts_at": booking.starts_at.isoformat(), "ends_at": booking.ends_at.isoformat()}
+    booking.starts_at = body.starts_at
+    booking.ends_at = body.ends_at
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=user.display_name,
+        action="booking_rescheduled",
+        entity_type="booking",
+        entity_id=booking.id,
+        before=before,
+        after={"starts_at": booking.starts_at.isoformat(), "ends_at": booking.ends_at.isoformat()},
+        metadata={"reason": body.reason.strip()},
+    )
+    db.commit()
+    return _to_out(booking, customer, vehicle)
+
+
+@router.post("/{public_ref}/cancel", response_model=BookingOut)
+def cancel_booking(
+    public_ref: str,
+    body: CancelBookingRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(get_current_user),
+) -> BookingOut:
+    booking = db.scalar(select(Booking).where(Booking.public_ref == public_ref).with_for_update())
+    if booking is None:
+        raise HTTPException(status_code=404, detail="Booking not found")
+    if booking.status not in ("reserved", "blocked"):
+        # A booking blocked at checkout (damage, technical warning, ...) has no other exit:
+        # it never became active, so it can neither be returned nor completed. Cancelling
+        # it (audited, with a reason) is the only way to close the file.
+        raise HTTPException(
+            status_code=409, detail="Only a reserved or blocked booking can be cancelled"
+        )
+    customer = db.get(Customer, booking.customer_id)
+    vehicle = db.get(Vehicle, booking.vehicle_id)
+    if customer is None or vehicle is None:
+        raise HTTPException(status_code=500, detail="Booking references a missing record")
+    before = {"status": booking.status}
+    booking.status = "cancelled"
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=user.display_name,
+        action="booking_cancelled",
+        entity_type="booking",
+        entity_id=booking.id,
+        before=before,
+        after={"status": booking.status, "reason": body.reason.strip()},
+    )
+    db.commit()
+    return _to_out(booking, customer, vehicle)
+
+
 @router.post("/{public_ref}/return-preview", response_model=ReturnPreviewResult)
 def preview_return(
     public_ref: str,
@@ -89,6 +491,8 @@ def preview_return(
         resulting_odometer_km=evaluation.resulting_odometer_km,
         resulting_vehicle_status=evaluation.resulting_vehicle_status,
         status_reason=evaluation.status_reason,
+        status_reason_code=evaluation.status_reason_code,
+        status_reason_params=evaluation.status_reason_params,
         would_create_quality_issue=evaluation.would_create_quality_issue,
         attention_reasons=evaluation.attention_reasons,
         next_booking_risk=(
@@ -99,7 +503,7 @@ def preview_return(
     )
 
 
-@router.post("/{public_ref}/return")
+@router.post("/{public_ref}/return", response_model=RegisterReturnResult)
 def register_return(
     public_ref: str,
     body: RegisterReturnRequest,
@@ -107,7 +511,7 @@ def register_return(
     idempotency_key: str = Header(..., alias="Idempotency-Key", min_length=8, max_length=128),
     db: Session = Depends(get_db),
     user: CurrentUser = Depends(get_current_user),
-) -> dict:
+) -> RegisterReturnResult:
     status_code, result = register_vehicle_return(db, public_ref, body, idempotency_key, user)
     response.status_code = status_code
-    return result
+    return RegisterReturnResult(**result)

backend/app/api/routers/customers.py

@@ -0,0 +1,41 @@
+from fastapi import APIRouter, Depends, Query
+from sqlalchemy import or_, select
+from sqlalchemy.orm import Session
+
+from app.api.deps import get_current_user, get_db
+from app.models.customer import Customer
+from app.schemas import CurrentUser, CustomerOptionOut
+
+router = APIRouter(prefix="/api/v1/customers", tags=["customers"])
+
+
+@router.get("", response_model=list[CustomerOptionOut])
+def search_customers(
+    query: str = Query(min_length=2, max_length=100),
+    limit: int = Query(default=20, ge=1, le=50),
+    db: Session = Depends(get_db),
+    _user: CurrentUser = Depends(get_current_user),
+) -> list[CustomerOptionOut]:
+    term = f"%{query.strip()}%"
+    customers = db.scalars(
+        select(Customer)
+        .where(
+            Customer.merged_into_customer_id.is_(None),
+            or_(
+                Customer.public_ref.ilike(term),
+                Customer.first_name.ilike(term),
+                Customer.last_name.ilike(term),
+                Customer.email.ilike(term),
+            ),
+        )
+        .order_by(Customer.last_name, Customer.first_name)
+        .limit(limit)
+    ).all()
+    return [
+        CustomerOptionOut(
+            public_ref=customer.public_ref,
+            display_name=f"{customer.first_name} {customer.last_name}",
+            email=customer.email,
+        )
+        for customer in customers
+    ]

backend/app/api/routers/dashboard.py

@@ -1,7 +1,8 @@
 from __future__ import annotations
 
-from datetime import UTC, date, datetime
+from datetime import date, datetime
 from typing import Literal
+from zoneinfo import ZoneInfo
 
 from fastapi import APIRouter, Depends
 from sqlalchemy import select
@@ -19,6 +20,7 @@ from app.schemas import (
     AutomationRunOut,
     CurrentUser,
     DashboardOut,
+    EvidenceSignalOut,
     TodayItem,
 )
 from app.services.operations import compute_metrics
@@ -29,10 +31,20 @@ settings = get_settings()
 _SEVERITY_ORDER = {"high": 0, "medium": 1, "low": 2}
 
 
+def _local_tz() -> ZoneInfo:
+    return ZoneInfo(settings.demo_timezone)
+
+
 def _today() -> date:
     # Seeded dates are shifted to the real reset moment by `seed_loader.py`'s anchor
     # shift, so "today" must be real wall-clock time, not the frozen `demo_today` setting.
-    return datetime.now(UTC).date()
+    # Timestamps are stored in UTC but the operational day is the local (Europe/Brussels)
+    # calendar day, so a 23:30Z departure belongs to tomorrow's schedule in summer.
+    return datetime.now(_local_tz()).date()
+
+
+def _local_date(value: datetime) -> date:
+    return value.astimezone(_local_tz()).date()
 
 
 @router.get("", response_model=DashboardOut)
@@ -61,19 +73,34 @@ def get_dashboard(
             entity = customers_by_id.get(issue.entity_id)
             link_type = "customer"
         link_ref = entity.public_ref if entity else ""
+        # The backend never emits prose for the attention queue -- only stable signal
+        # codes + raw data params, exactly like the issue detail page's evidence list
+        # (see app/services/data_quality.py::_open_issue). The frontend is the one place
+        # that turns these into the operator's selected language; `evidence_json["summary"]`
+        # is a technical fallback only, never rendered here.
+        signals = [
+            EvidenceSignalOut(code=s["code"], params=s.get("params", {}))
+            for s in issue.evidence_json.get("signals", [])
+        ]
         attention_items.append(
             AttentionItem(
                 kind="quality_issue",
                 severity=issue.severity,
                 rule_type=issue.rule_type,
-                detail=issue.evidence_json.get("summary", ""),
+                evidence_signals=signals,
                 link_type=link_type,
                 link_ref=link_ref,
                 issue_ref=issue.public_ref,
             )
         )
-    attention_items.sort(key=lambda item: _SEVERITY_ORDER.get(item.severity, 3))
+    # Curate a credible severity mix instead of letting `high` dominate every slot:
-    attention_items = attention_items[:8]
+    # each item's real severity is unchanged, only the display selection is capped per
+    # tier (a handful of "now", then "today", then "later") so a heavy day of high-severity
+    # issues doesn't crowd out medium/low ones the operator should still see.
+    high_items = [i for i in attention_items if i.severity == "high"]
+    medium_items = [i for i in attention_items if i.severity == "medium"]
+    low_items = [i for i in attention_items if i.severity == "low"]
+    attention_items = (high_items[:3] + medium_items[:3] + low_items[:2])[:8]
 
     today = _today()
     bookings = db.scalars(select(Booking)).all()
@@ -81,25 +108,27 @@ def get_dashboard(
     for b in bookings:
         vehicle = vehicles_by_id.get(b.vehicle_id)
         vehicle_ref = vehicle.public_ref if vehicle else ""
-        if b.starts_at.date() == today and b.status in ("reserved", "active"):
+        if _local_date(b.starts_at) == today and b.status in ("reserved", "active"):
             today_items.append(
                 TodayItem(
-                    kind="departure", booking_ref=b.public_ref, vehicle_ref=vehicle_ref,
+                    kind="departure",
+                    booking_ref=b.public_ref,
+                    vehicle_ref=vehicle_ref,
                     scheduled_at=b.starts_at,
                 )
             )
-        if b.ends_at.date() == today and b.status in ("active", "returned"):
+        if _local_date(b.ends_at) == today and b.status in ("active", "returned"):
             today_items.append(
                 TodayItem(
-                    kind="return", booking_ref=b.public_ref, vehicle_ref=vehicle_ref,
+                    kind="return",
+                    booking_ref=b.public_ref,
+                    vehicle_ref=vehicle_ref,
                     scheduled_at=b.ends_at,
                 )
             )
     today_items.sort(key=lambda item: item.scheduled_at)
 
-    recent = db.scalars(
+    recent = db.scalars(select(OutboxEvent).order_by(OutboxEvent.occurred_at.desc()).limit(5)).all()
-        select(OutboxEvent).order_by(OutboxEvent.occurred_at.desc()).limit(5)
-    ).all()
     recent_automation = [
         AutomationRunOut(
             event_id=str(r.event_id),
@@ -108,6 +137,7 @@ def get_dashboard(
             status=r.delivery_status,
             attempts=r.attempts,
             last_error=r.last_error,
+            last_error_code=r.last_error_code,
             occurred_at=r.occurred_at,
         )
         for r in recent

backend/app/api/routers/data_quality.py

@@ -1,7 +1,9 @@
 from __future__ import annotations
 
+from datetime import UTC, datetime
+
 from fastapi import APIRouter, Depends, HTTPException, Query
-from sqlalchemy import select
+from sqlalchemy import case, func, select
 from sqlalchemy.orm import Session
 
 from app.api.deps import get_db, require_operations_manager
@@ -9,23 +11,32 @@ from app.models.booking import Booking
 from app.models.customer import Customer
 from app.models.data_quality import DataQualityIssue
 from app.models.inspection import Inspection
+from app.models.user import User
 from app.models.vehicle import Vehicle
 from app.schemas import (
+    ApplyRecommendedStatusRequest,
     ApplyRecommendedStatusResult,
+    BulkDataQualityWorkRequest,
+    BulkDataQualityWorkResult,
     CurrentUser,
     DataQualityIssueDetailOut,
     DataQualityIssueOut,
+    DataQualityIssuePageOut,
     MergeCustomersRequest,
     MergeCustomersResult,
     ProvideFieldsRequest,
     ResolveOdometerRegressionRequest,
     ResolveOverlapRequest,
     ScanResultOut,
+    StatusRecommendationOut,
+    VehicleStatusFactsOut,
 )
+from app.services.audit import record_audit_event
 from app.services.data_quality import (
     apply_recommended_status,
     defer_issue,
     merge_customers,
+    preview_vehicle_status_recommendation,
     provide_missing_fields,
     reject_issue,
     resolve_booking_overlap,
@@ -37,6 +48,7 @@ router = APIRouter(prefix="/api/v1/data-quality", tags=["data-quality"])
 
 
 def _to_out(issue: DataQualityIssue) -> DataQualityIssueOut:
+    assignee = issue.assigned_to_user
     return DataQualityIssueOut(
         public_ref=issue.public_ref,
         rule_type=issue.rule_type,
@@ -46,27 +58,160 @@ def _to_out(issue: DataQualityIssue) -> DataQualityIssueOut:
         status=issue.status,
         evidence=issue.evidence_json,
         detected_at=issue.detected_at,
+        due_at=issue.due_at,
+        assigned_to_ref=assignee.public_ref if assignee else None,
+        assigned_to_name=assignee.display_name if assignee else None,
+        overdue=(
+            issue.status == "open" and issue.due_at is not None and issue.due_at < datetime.now(UTC)
+        ),
         resolved_at=issue.resolved_at,
     )
 
 
-@router.get("/issues", response_model=list[DataQualityIssueOut])
+@router.get("/issues", response_model=list[DataQualityIssueOut] | DataQualityIssuePageOut)
 def list_issues(
     status: str | None = Query(default=None),
     rule_type: str | None = Query(default=None),
     severity: str | None = Query(default=None),
+    assigned_to_ref: str | None = Query(default=None),
+    overdue: bool | None = Query(default=None),
+    demo_only: bool | None = Query(default=None),
+    page: int | None = Query(default=None, ge=1),
+    page_size: int = Query(default=25, ge=1, le=25),
     db: Session = Depends(get_db),
     _user: CurrentUser = Depends(require_operations_manager),
-) -> list[DataQualityIssueOut]:
+) -> list[DataQualityIssueOut] | DataQualityIssuePageOut:
-    stmt = select(DataQualityIssue).order_by(DataQualityIssue.detected_at.desc())
+    severity_order = case(
+        (DataQualityIssue.severity == "high", 0),
+        (DataQualityIssue.severity == "medium", 1),
+        else_=2,
+    )
+    stmt = select(DataQualityIssue).order_by(
+        DataQualityIssue.due_at.asc().nulls_last(),
+        severity_order,
+        DataQualityIssue.detected_at.desc(),
+    )
     if status:
         stmt = stmt.where(DataQualityIssue.status == status)
     if rule_type:
         stmt = stmt.where(DataQualityIssue.rule_type == rule_type)
     if severity:
         stmt = stmt.where(DataQualityIssue.severity == severity)
-    issues = db.scalars(stmt).all()
+    if assigned_to_ref == "unassigned":
-    return [_to_out(i) for i in issues]
+        stmt = stmt.where(DataQualityIssue.assigned_to_user_id.is_(None))
+    elif assigned_to_ref:
+        stmt = stmt.join(DataQualityIssue.assigned_to_user).where(
+            User.public_ref == assigned_to_ref
+        )
+    if overdue is True:
+        stmt = stmt.where(
+            DataQualityIssue.status == "open",
+            DataQualityIssue.due_at < datetime.now(UTC),
+        )
+    if demo_only is True:
+        # Server-side so the guided demo scenarios are found on any page, not only the
+        # 25 rows currently loaded in the browser.
+        stmt = stmt.where(DataQualityIssue.public_ref.like("DQ-DEMO-%"))
+    total = db.scalar(select(func.count()).select_from(stmt.subquery())) or 0
+    page_number = page or 1
+    issues = db.scalars(
+        stmt if page is None else stmt.offset((page_number - 1) * page_size).limit(page_size)
+    ).all()
+    items = [_to_out(i) for i in issues]
+    if page is None:
+        return items
+    total_pages = max(1, (total + page_size - 1) // page_size)
+    return DataQualityIssuePageOut(
+        items=items,
+        page=min(page_number, total_pages),
+        page_size=page_size,
+        total=total,
+        total_pages=total_pages,
+    )
+
+
+@router.post("/issues/bulk-work", response_model=BulkDataQualityWorkResult)
+def update_issue_work_queue(
+    body: BulkDataQualityWorkRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(require_operations_manager),
+) -> BulkDataQualityWorkResult:
+    refs = list(dict.fromkeys(body.issue_refs))
+    if (
+        body.assigned_to_ref is None
+        and not body.clear_assignment
+        and body.due_at is None
+        and not body.clear_due_at
+    ):
+        raise HTTPException(status_code=422, detail="No work queue change was requested")
+    if body.assigned_to_ref is not None and body.clear_assignment:
+        raise HTTPException(status_code=422, detail="Choose an assignee or clear assignment")
+    if body.due_at is not None and body.clear_due_at:
+        raise HTTPException(status_code=422, detail="Choose a due date or clear the due date")
+    if body.due_at is not None and body.due_at.tzinfo is None:
+        raise HTTPException(status_code=422, detail="Due date must include a timezone")
+
+    assignee = None
+    if body.assigned_to_ref is not None:
+        assignee = db.scalar(
+            select(User).where(
+                User.public_ref == body.assigned_to_ref,
+                User.active.is_(True),
+            )
+        )
+        if assignee is None:
+            raise HTTPException(status_code=422, detail="Active assignee not found")
+
+    issues = list(
+        db.scalars(
+            select(DataQualityIssue).where(DataQualityIssue.public_ref.in_(refs)).with_for_update()
+        ).all()
+    )
+    if len(issues) != len(refs):
+        found = {issue.public_ref for issue in issues}
+        missing = next(ref for ref in refs if ref not in found)
+        raise HTTPException(status_code=404, detail=f"Data quality issue {missing} not found")
+
+    for issue in issues:
+        if issue.status != "open":
+            raise HTTPException(
+                status_code=409,
+                detail=f"Data quality issue {issue.public_ref} is not open",
+            )
+        before = {
+            "assigned_to_ref": issue.assigned_to_user.public_ref
+            if issue.assigned_to_user
+            else None,
+            "due_at": issue.due_at.isoformat() if issue.due_at else None,
+        }
+        if body.assigned_to_ref is not None:
+            issue.assigned_to_user = assignee
+        elif body.clear_assignment:
+            issue.assigned_to_user = None
+        if body.due_at is not None:
+            issue.due_at = body.due_at
+        elif body.clear_due_at:
+            issue.due_at = None
+        after = {
+            "assigned_to_ref": assignee.public_ref
+            if body.assigned_to_ref is not None and assignee
+            else (None if body.clear_assignment else before["assigned_to_ref"]),
+            "due_at": issue.due_at.isoformat() if issue.due_at else None,
+        }
+        record_audit_event(
+            db,
+            actor_type="user",
+            actor_label=user.display_name,
+            action="data_quality_work_updated",
+            entity_type="data_quality_issue",
+            entity_id=issue.id,
+            before=before,
+            after=after,
+        )
+    db.commit()
+    for issue in issues:
+        db.refresh(issue)
+    return BulkDataQualityWorkResult(updated=[_to_out(issue) for issue in issues])
 
 
 # Every public reference in this system carries its entity type in its own prefix
@@ -111,6 +256,7 @@ def _snapshot(entity_type: str, ref: str, db: Session) -> dict | None:
         return {
             "entity_type": "vehicle",
             "public_ref": vehicle.public_ref,
+            "registration_number": vehicle.registration_number,
             "make": vehicle.make,
             "model": vehicle.model,
             "location": vehicle.location,
@@ -217,9 +363,7 @@ def provide_fields(
     return _to_out(issue)
 
 
-@router.post(
+@router.post("/issues/{public_ref}/resolve-odometer-regression", response_model=DataQualityIssueOut)
-    "/issues/{public_ref}/resolve-odometer-regression", response_model=DataQualityIssueOut
-)
 def resolve_odometer(
     public_ref: str,
     body: ResolveOdometerRegressionRequest,
@@ -241,17 +385,41 @@ def resolve_overlap(
     return _to_out(issue)
 
 
+@router.post("/issues/{public_ref}/status-recommendation", response_model=StatusRecommendationOut)
+def status_recommendation(
+    public_ref: str,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(require_operations_manager),
+) -> StatusRecommendationOut:
+    """Non-mutating preview: computes the recommendation without changing anything,
+    resolving no issue and writing no audit event. Safe to call repeatedly."""
+    _issue, _vehicle, recommendation, token = preview_vehicle_status_recommendation(db, public_ref)
+    return StatusRecommendationOut(
+        current_status=recommendation.current_status,
+        recommended_status=recommendation.recommended_status,
+        recommendation_code=recommendation.recommendation_code,
+        safe_to_apply=recommendation.safe_to_apply,
+        manual_review_required=recommendation.manual_review_required,
+        facts=VehicleStatusFactsOut(**recommendation.facts.as_dict()),
+        blocking_reasons=recommendation.blocking_reasons,
+        recommendation_token=token,
+    )
+
+
 @router.post(
     "/issues/{public_ref}/apply-recommended-status", response_model=ApplyRecommendedStatusResult
 )
 def apply_status(
     public_ref: str,
+    body: ApplyRecommendedStatusRequest,
     db: Session = Depends(get_db),
     user: CurrentUser = Depends(require_operations_manager),
 ) -> ApplyRecommendedStatusResult:
-    issue, applied_status, reason = apply_recommended_status(db, public_ref, user)
+    issue, applied_status, reason_code = apply_recommended_status(
+        db, public_ref, user, body.recommendation_token
+    )
     return ApplyRecommendedStatusResult(
-        issue=_to_out(issue), applied_status=applied_status, reason=reason
+        issue=_to_out(issue), applied_status=applied_status, reason_code=reason_code
     )
 
 

backend/app/api/routers/demo.py

@@ -1,10 +1,11 @@
 from __future__ import annotations
 
+import threading
 import time
 import uuid
 
 from fastapi import APIRouter, Depends, HTTPException, Request, Response, status
-from sqlalchemy import select
+from sqlalchemy import func, select
 from sqlalchemy.orm import Session
 
 from app.api.deps import get_current_user, get_db, require_operations_manager
@@ -15,13 +16,19 @@ from app.schemas import CurrentUser, DemoLoginRequest, DemoManifestOut
 from app.seed_loader import reset_and_seed
 from app.services.audit import record_audit_event
 from app.services.demo_manifest import build_demo_manifest, scenario_integrity_report
+from app.services.sessions import revoke_session
 
 router = APIRouter(prefix="/api/v1/demo", tags=["demo"])
 settings = get_settings()
+_reset_guard = threading.Lock()
+_last_reset_monotonic = 0.0
+_RESET_ADVISORY_LOCK_ID = 706_533_149
 
 
 @router.get("/manifest", response_model=DemoManifestOut)
 def demo_manifest(db: Session = Depends(get_db)) -> DemoManifestOut:
+    if not settings.mobilityops_demo_mode:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Demo mode is disabled")
     # Deliberately unauthenticated: the demo-entry screen and the permanent demo badge
     # both need this before any session exists. Nothing here is sensitive — it's the same
     # honest "what is this demo" summary a logged-in user would see.
@@ -32,6 +39,8 @@ def demo_manifest(db: Session = Depends(get_db)) -> DemoManifestOut:
 def demo_login(
     body: DemoLoginRequest, response: Response, db: Session = Depends(get_db)
 ) -> CurrentUser:
+    if not settings.mobilityops_demo_mode:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Demo mode is disabled")
     public_ref = "USR-OPS" if body.role == "operations_manager" else "USR-EMP"
     user = db.scalar(select(User).where(User.public_ref == public_ref))
     if user is None:
@@ -44,6 +53,7 @@ def demo_login(
             role=user.role,
             display_name=user.display_name,
             issued_at=int(time.time()),
+            session_id=str(uuid.uuid4()),
         )
     )
     response.set_cookie(
@@ -68,9 +78,7 @@ def demo_login(
 
 
 @router.get("/session", response_model=CurrentUser)
-def get_session(
+def get_session(response: Response, user: CurrentUser = Depends(get_current_user)) -> CurrentUser:
-    response: Response, user: CurrentUser = Depends(get_current_user)
-) -> CurrentUser:
     # Never let the browser (or an intermediary) cache an authentication check — a stale
     # cached 200 here would keep showing a logged-out browser as authenticated.
     response.headers["Cache-Control"] = "no-store"
@@ -81,7 +89,8 @@ def get_session(
 def demo_logout(request: Request, response: Response, db: Session = Depends(get_db)) -> dict:
     token = request.cookies.get(settings.session_cookie_name)
     payload = read_session_token(token) if token else None
-    if payload is not None:
+    if payload is not None and token is not None:
+        revoke_session(db, token, payload)
         record_audit_event(
             db,
             actor_type="user",
@@ -101,12 +110,31 @@ def demo_reset(
     db: Session = Depends(get_db),
     user: CurrentUser = Depends(require_operations_manager),
 ) -> dict:
+    global _last_reset_monotonic
+    if not settings.mobilityops_demo_mode:
+        # Outside demo mode the reset endpoint must not exist at all: it wipes
+        # operational data and replaces it with synthetic records.
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Demo mode is disabled")
     if not settings.demo_allow_reset:
         raise HTTPException(
             status_code=status.HTTP_403_FORBIDDEN,
             detail="Demo reset is disabled on this deployment.",
         )
-    result = reset_and_seed(db)
+    if not _reset_guard.acquire(blocking=False):
+        raise HTTPException(status_code=409, detail="A demo reset is already running.")
+    try:
+        elapsed = time.monotonic() - _last_reset_monotonic
+        if _last_reset_monotonic and elapsed < settings.demo_reset_cooldown_seconds:
+            retry_after = max(1, int(settings.demo_reset_cooldown_seconds - elapsed + 0.999))
+            raise HTTPException(
+                status_code=429,
+                detail=f"Demo reset is cooling down. Retry in {retry_after} seconds.",
+                headers={"Retry-After": str(retry_after)},
+            )
+        locked = db.scalar(select(func.pg_try_advisory_xact_lock(_RESET_ADVISORY_LOCK_ID)))
+        if not locked:
+            raise HTTPException(status_code=409, detail="A demo reset is already running.")
+        result = reset_and_seed(db, preserve_integration_telemetry=True)
         integrity = scenario_integrity_report(db)
         record_audit_event(
             db,
@@ -121,6 +149,9 @@ def demo_reset(
             },
         )
         db.commit()
+        _last_reset_monotonic = time.monotonic()
+    finally:
+        _reset_guard.release()
     response.delete_cookie(settings.session_cookie_name)
     return {
         "status": "reset",

backend/app/api/routers/integration_status.py

@@ -4,16 +4,10 @@ from fastapi import APIRouter, Depends
 from sqlalchemy.orm import Session
 
 from app.api.deps import get_db, require_operations_manager
-from app.core.config import get_settings
+from app.schemas import CurrentUser, IntegrationStatusOut
-from app.schemas import (
+from app.services.integration_status import derive_mcp_hub_status, derive_n8n_status
-    CurrentUser,
-    IntegrationStatusOut,
-    McpHubIntegrationStatus,
-)
-from app.services.integration_status import derive_n8n_status
 
 router = APIRouter(prefix="/api/v1/integrations", tags=["integrations"])
-settings = get_settings()
 
 
 @router.get("/status", response_model=IntegrationStatusOut)
@@ -23,8 +17,5 @@ def integration_status(
 ) -> IntegrationStatusOut:
     return IntegrationStatusOut(
         n8n=derive_n8n_status(db),
-        mcp_hub=McpHubIntegrationStatus(
+        mcp_hub=derive_mcp_hub_status(db),
-            registration_enabled=settings.mcp_hub_registration_enabled,
-            state="configured" if settings.mcp_hub_registration_enabled else "not_configured",
-        ),
     )

backend/app/api/routers/integrations.py

@@ -1,8 +1,9 @@
 from __future__ import annotations
 
+import hmac
 import uuid
 from datetime import UTC, datetime
-from typing import Any
+from pathlib import Path
 
 from fastapi import APIRouter, Depends, Header
 from sqlalchemy import select
@@ -13,23 +14,93 @@ from app.core.config import get_settings
 from app.core.errors import AppError
 from app.models.audit import AuditEvent
 from app.models.outbox import OutboxEvent
-from app.schemas import ScanResultOut
+from app.schemas import (
+    N8nHeartbeatIn,
+    N8nHeartbeatResult,
+    ProcedureDocumentOut,
+    ProcedureListOut,
+    ProcedureSyncResultIn,
+    ProcedureSyncResultResult,
+    ReturnCallbackIn,
+    ScanResultOut,
+    WorkflowErrorReportIn,
+    WorkflowErrorReportResult,
+)
 from app.services.audit import record_audit_event
 from app.services.data_quality import run_scan
+from app.services.knowledge.procedures import iter_procedure_documents
 
 router = APIRouter(prefix="/api/v1/integrations/n8n", tags=["integrations"])
 settings = get_settings()
 
+_CANONICAL_WORKFLOW_NAMES = frozenset(
+    {
+        "Fleet Ops — Vehicle Return Orchestration",
+        "Fleet Ops — Scheduled Data Quality Scan",
+        "Fleet Ops — RAGcore Procedure Sync",
+        "Fleet Ops — Workflow Error Handler",
+    }
+)
+
+
+def _require_service_token(service_token: str) -> None:
+    # Constant-time comparison: a plain ``!=`` leaks how many leading bytes matched.
+    if not hmac.compare_digest(
+        service_token.encode("utf-8"), settings.n8n_callback_token.encode("utf-8")
+    ):
+        raise AppError("UNAUTHORIZED_SERVICE", "Invalid service token.", status_code=401)
+
+
+@router.post("/heartbeat", response_model=N8nHeartbeatResult)
+def workflow_heartbeat(
+    body: N8nHeartbeatIn,
+    service_token: str = Header(..., alias="X-Service-Token"),
+    db: Session = Depends(get_db),
+) -> N8nHeartbeatResult:
+    """Authenticated, idempotent execution evidence from a canonical n8n workflow."""
+    _require_service_token(service_token)
+    if body.workflow_name not in _CANONICAL_WORKFLOW_NAMES:
+        raise AppError("UNKNOWN_WORKFLOW", "Unknown Fleet Ops workflow.", status_code=422)
+    already_recorded = (
+        db.scalar(
+            select(AuditEvent.id).where(
+                AuditEvent.action == "n8n_workflow_heartbeat",
+                AuditEvent.metadata_json["execution_id"].astext == body.execution_id,
+                AuditEvent.after_json["status"].astext == body.status,
+            )
+        )
+        is not None
+    )
+    if not already_recorded:
+        record_audit_event(
+            db,
+            actor_type="service",
+            actor_label="n8n workflow heartbeat",
+            action="n8n_workflow_heartbeat",
+            entity_type="automation",
+            after={
+                "workflow_id": body.workflow_id,
+                "workflow_name": body.workflow_name,
+                "status": body.status,
+            },
+            metadata={"execution_id": body.execution_id},
+        )
+        db.commit()
+    return N8nHeartbeatResult(
+        status="already_registered" if already_recorded else "registered",
+        execution_id=body.execution_id,
+        occurred_at=datetime.now(UTC),
+    )
+
 
 @router.post("/return-callback")
 def return_callback(
-    body: dict[str, Any],
+    body: ReturnCallbackIn,
     idempotency_key: str = Header(..., alias="Idempotency-Key"),
     service_token: str = Header(..., alias="X-Service-Token"),
     db: Session = Depends(get_db),
 ) -> dict:
-    if service_token != settings.n8n_callback_token:
+    _require_service_token(service_token)
-        raise AppError("UNAUTHORIZED_SERVICE", "Invalid service token.", status_code=401)
 
     try:
         event_id = uuid.UUID(idempotency_key)
@@ -60,10 +131,8 @@ def return_callback(
             actor_label="n8n",
             action="n8n_return_followup_recorded",
             entity_type="booking",
-            correlation_id=uuid.UUID(body.get("correlation_id"))
+            correlation_id=body.correlation_id,
-            if body.get("correlation_id")
+            after={"follow_up": body.follow_up, "summary": body.summary},
-            else None,
-            after={"follow_up": body.get("follow_up"), "summary": body.get("summary")},
             metadata={"event_id": str(event_id)},
         )
         db.commit()
@@ -84,8 +153,125 @@ def scheduled_scan(
     safe to call repeatedly: run_scan() only ever creates an issue for a condition that
     doesn't already have one open, so a duplicate or overlapping trigger does no
     duplicate domain work -- it just reports zero new issues for anything already known."""
-    if service_token != settings.n8n_callback_token:
+    _require_service_token(service_token)
-        raise AppError("UNAUTHORIZED_SERVICE", "Invalid service token.", status_code=401)
 
     result = run_scan(db, actor_label="n8n scheduled scan", actor_type="service")
     return ScanResultOut(created=result.created)
+
+
+@router.post("/workflow-error", response_model=WorkflowErrorReportResult)
+def workflow_error(
+    body: WorkflowErrorReportIn,
+    service_token: str = Header(..., alias="X-Service-Token"),
+    db: Session = Depends(get_db),
+) -> WorkflowErrorReportResult:
+    """Receives a bounded, secret-free failure report from the central n8n "Fleet Ops --
+    Workflow Error Handler" workflow, which is attached as the Error Workflow on every
+    other Fleet Ops n8n workflow. Idempotent on execution_id: n8n may redeliver the same
+    error report (e.g. after a timed-out response), so this must not double-record."""
+    _require_service_token(service_token)
+
+    already_recorded = (
+        db.scalar(
+            select(AuditEvent.id).where(
+                AuditEvent.action == "n8n_workflow_failure_registered",
+                AuditEvent.metadata_json["execution_id"].astext == body.execution_id,
+            )
+        )
+        is not None
+    )
+    if not already_recorded:
+        correlation_id: uuid.UUID | None = None
+        if body.correlation_id:
+            try:
+                correlation_id = uuid.UUID(body.correlation_id)
+            except ValueError:
+                correlation_id = None
+        record_audit_event(
+            db,
+            actor_type="service",
+            actor_label="n8n error handler",
+            action="n8n_workflow_failure_registered",
+            entity_type="automation",
+            correlation_id=correlation_id,
+            after={
+                "workflow_id": body.workflow_id,
+                "workflow_name": body.workflow_name,
+                "error_category": body.error_category,
+                "error_summary": body.error_summary,
+                "trigger_context": body.trigger_context,
+                "attempt": body.attempt,
+                "retry_action": body.retry_action,
+                "failed_at": body.failed_at.isoformat(),
+            },
+            metadata={"execution_id": body.execution_id},
+        )
+        db.commit()
+
+    return WorkflowErrorReportResult(
+        status="already_registered" if already_recorded else "registered",
+        execution_id=body.execution_id,
+        occurred_at=datetime.now(UTC),
+    )
+
+
+@router.get("/procedures", response_model=ProcedureListOut)
+def list_procedures(service_token: str = Header(..., alias="X-Service-Token")) -> ProcedureListOut:
+    """Read-only source list for the RAGcore Procedure Sync workflow: every procedure
+    Markdown file Fleet Ops ships, across every supported language, with a stable
+    per-document id (source_id) and a content hash so the caller can detect changes
+    without re-fetching content it already has."""
+    _require_service_token(service_token)
+
+    documents = [
+        ProcedureDocumentOut(
+            id=doc.source_id,
+            language=doc.language,
+            document_id=doc.document_id,
+            title=doc.title,
+            version=doc.version,
+            content=doc.content,
+            content_hash=doc.content_hash,
+        )
+        for doc in iter_procedure_documents(Path(settings.knowledge_dir))
+    ]
+    return ProcedureListOut(documents=documents)
+
+
+@router.post("/procedures-sync-result", response_model=ProcedureSyncResultResult)
+def procedures_sync_result(
+    body: ProcedureSyncResultIn,
+    service_token: str = Header(..., alias="X-Service-Token"),
+    db: Session = Depends(get_db),
+) -> ProcedureSyncResultResult:
+    """Receives a summary (counts only, no document content) from the n8n "Fleet Ops --
+    RAGcore Procedure Sync" workflow once it finishes uploading procedures to RAGcore.
+    Idempotent on execution_id, matching the workflow-error and return-callback pattern."""
+    _require_service_token(service_token)
+
+    already_recorded = (
+        db.scalar(
+            select(AuditEvent.id).where(
+                AuditEvent.action == "n8n_procedures_synced",
+                AuditEvent.metadata_json["execution_id"].astext == body.execution_id,
+            )
+        )
+        is not None
+    )
+    if not already_recorded:
+        record_audit_event(
+            db,
+            actor_type="service",
+            actor_label="n8n procedure sync",
+            action="n8n_procedures_synced",
+            entity_type="automation",
+            after={"synced": body.synced, "failed": body.failed},
+            metadata={"execution_id": body.execution_id},
+        )
+        db.commit()
+
+    return ProcedureSyncResultResult(
+        status="already_registered" if already_recorded else "registered",
+        execution_id=body.execution_id,
+        occurred_at=datetime.now(UTC),
+    )

backend/app/api/routers/knowledge.py

@@ -1,18 +1,32 @@
 from __future__ import annotations
 
+import hashlib
 import uuid
 from typing import Literal
 
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter, Depends, HTTPException, Request, status
 from pydantic import BaseModel, Field
+from sqlalchemy import select
 from sqlalchemy.orm import Session
 
 from app.api.deps import get_current_user, get_db
+from app.core.config import get_settings
+from app.core.ratelimit import SlidingWindowLimiter
+from app.models.audit import AuditEvent
 from app.schemas import CurrentUser
 from app.services.audit import record_audit_event
 from app.services.knowledge import GroundedAnswer, KnowledgeHealth, get_knowledge_provider
 
 router = APIRouter(prefix="/api/v1/knowledge", tags=["knowledge"])
+settings = get_settings()
+_question_limiter = (
+    SlidingWindowLimiter(
+        max_requests=settings.knowledge_max_requests,
+        window_seconds=settings.knowledge_rate_limit_window_seconds,
+    )
+    if settings.knowledge_max_requests > 0
+    else None
+)
 
 SupportedLanguage = Literal["nl-BE", "en-GB", "fr-BE"]
 
@@ -22,12 +36,37 @@ class AskQuestionRequest(BaseModel):
     language: SupportedLanguage = "en-GB"
 
 
+class KnowledgeFeedbackRequest(BaseModel):
+    correlation_id: uuid.UUID
+    helpful: bool
+
+
 @router.post("/questions", response_model=GroundedAnswer)
 def ask_question(
     body: AskQuestionRequest,
+    request: Request,
     db: Session = Depends(get_db),
     user: CurrentUser = Depends(get_current_user),
 ) -> GroundedAnswer:
+    if _question_limiter is not None:
+        forwarded = request.headers.get("x-forwarded-for", "")
+        client_ip = (
+            forwarded.split(",")[-1].strip()
+            if forwarded
+            else request.client.host if request.client else "unknown"
+        )
+        token = request.cookies.get(settings.session_cookie_name, "")
+        session_key = hashlib.sha256(token.encode("utf-8")).hexdigest()
+        retry_after = max(
+            _question_limiter.consume(f"ip:{client_ip}"),
+            _question_limiter.consume(f"session:{session_key}"),
+        )
+        if retry_after:
+            raise HTTPException(
+                status_code=status.HTTP_429_TOO_MANY_REQUESTS,
+                detail="Too many knowledge questions. Try again later.",
+                headers={"Retry-After": str(retry_after)},
+            )
     correlation_id = str(uuid.uuid4())
     provider = get_knowledge_provider()
     answer = provider.ask(body.question, correlation_id, body.language)
@@ -51,9 +90,78 @@ def ask_question(
     return answer
 
 
+@router.post("/feedback")
+def record_feedback(
+    body: KnowledgeFeedbackRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(get_current_user),
+) -> dict[str, str]:
+    question_event = db.scalar(
+        select(AuditEvent.id).where(
+            AuditEvent.action == "knowledge_question_asked",
+            AuditEvent.correlation_id == body.correlation_id,
+            AuditEvent.actor_label == user.display_name,
+        )
+    )
+    if question_event is None:
+        raise HTTPException(status_code=404, detail="Knowledge exchange not found")
+
+    existing = db.scalar(
+        select(AuditEvent).where(
+            AuditEvent.action == "knowledge_feedback_recorded",
+            AuditEvent.correlation_id == body.correlation_id,
+            AuditEvent.actor_label == user.display_name,
+        )
+    )
+    if existing is not None:
+        existing.metadata_json = {"helpful": body.helpful}
+    else:
+        record_audit_event(
+            db,
+            actor_type="user",
+            actor_label=user.display_name,
+            action="knowledge_feedback_recorded",
+            entity_type="knowledge",
+            correlation_id=body.correlation_id,
+            metadata={"helpful": body.helpful},
+        )
+    db.commit()
+    return {"status": "recorded"}
+
+
 @router.get("/status", response_model=KnowledgeHealth)
 def knowledge_status(
     language: SupportedLanguage = "en-GB",
+    db: Session = Depends(get_db),
     _user: CurrentUser = Depends(get_current_user),
 ) -> KnowledgeHealth:
-    return get_knowledge_provider().health(language)
+    health = get_knowledge_provider().health(language)
+    if health.provider != "ragcore":
+        return health
+
+    latest_sync = db.scalar(
+        select(AuditEvent)
+        .where(AuditEvent.action == "n8n_procedures_synced")
+        .order_by(AuditEvent.occurred_at.desc())
+        .limit(1)
+    )
+    if latest_sync is None:
+        return health
+
+    reported = latest_sync.after_json or {}
+    synced = reported.get("synced")
+    failed = reported.get("failed")
+    return health.model_copy(
+        update={
+            "reported_synced_document_count": synced if isinstance(synced, int) else None,
+            "reported_failed_document_count": failed if isinstance(failed, int) else None,
+            "last_sync_at": latest_sync.occurred_at,
+            # A persisted sync callback is useful additional provenance, but must not
+            # downgrade stronger provider-side verification to merely "reported".
+            "statistics_state": (
+                health.statistics_state
+                if health.statistics_state == "verified"
+                else "sync_reported"
+            ),
+        }
+    )

backend/app/api/routers/mcp_integrations.py

@@ -3,11 +3,11 @@ from __future__ import annotations
 import uuid
 from datetime import date
 
-from fastapi import APIRouter, Depends, Query
+from fastapi import APIRouter, Depends, Header, Query, Response
 from sqlalchemy import select
 from sqlalchemy.orm import Session
 
-from app.api.deps import get_db, require_mcp_service_token
+from app.api.deps import McpClientContext, get_db, require_mcp_service_token
 from app.core.config import get_settings
 from app.core.errors import AppError
 from app.models.booking import Booking
@@ -27,44 +27,94 @@ router = APIRouter(prefix="/api/v1/integrations/mcp", tags=["mcp"])
 settings = get_settings()
 
 
-def _audit_service_request(db: Session, *, client_id: str, tool: str, status_label: str) -> None:
+def get_correlation_id(
+    x_correlation_id: str | None = Header(default=None, alias="X-Correlation-Id"),
+) -> str:
+    """Preserve the Hub's own inbound correlation ID through MCP client -> Hub -> Fleet
+    Ops -> RAGcore -> Fleet Ops Audit; only mint a fresh one when none was supplied or
+    it isn't a valid UUID (per the task's own correlation-propagation contract)."""
+    if x_correlation_id:
+        try:
+            return str(uuid.UUID(x_correlation_id))
+        except ValueError:
+            pass
+    return str(uuid.uuid4())
+
+
+def _audit_service_request(
+    db: Session,
+    *,
+    reported_client_id: str,
+    tool: str,
+    status_label: str,
+    correlation_id: str,
+    metadata: dict[str, object] | None = None,
+) -> None:
     record_audit_event(
         db,
         actor_type="service",
-        actor_label=client_id,
+        # The shared service token authenticates the Hub, not the caller identity that
+        # the Hub reports in a header. Keep attribution authoritative and retain the
+        # reported value only as explicitly non-authenticated diagnostic metadata.
+        actor_label="itworx-mcp-hub",
         action="mcp_tool_request",
         entity_type="mcp_tool",
-        correlation_id=uuid.uuid4(),
+        correlation_id=uuid.UUID(correlation_id),
-        metadata={"tool": tool, "status": status_label},
+        metadata={
+            "tool": tool,
+            "status": status_label,
+            "reported_client_id": reported_client_id,
+            **(metadata or {}),
+        },
     )
     db.commit()
 
 
+def _set_trace_headers(response: Response, correlation_id: str, tenant: str) -> None:
+    response.headers["X-Correlation-Id"] = correlation_id
+    response.headers["X-Tenant-Id"] = tenant
+    response.headers["Cache-Control"] = "no-store"
+
+
 @router.get("/operations-summary", response_model=OperationsSummaryOut)
 def operations_summary(
+    response: Response,
     db: Session = Depends(get_db),
-    client_id: str = Depends(require_mcp_service_token),
+    client: McpClientContext = Depends(require_mcp_service_token),
+    correlation_id: str = Depends(get_correlation_id),
 ) -> OperationsSummaryOut:
     metrics = compute_metrics(db)
+    _set_trace_headers(response, correlation_id, client.tenant)
     _audit_service_request(
-        db, client_id=client_id, tool="mobilityops_get_operations_summary", status_label="ok"
+        db,
+        reported_client_id=client.reported_client_id,
+        tool="fleet_ops_get_operations_summary",
+        status_label="ok",
+        correlation_id=correlation_id,
     )
-    return OperationsSummaryOut(tenant=settings.ragcore_tenant, metrics=metrics)
+    return OperationsSummaryOut(tenant=client.tenant, metrics=metrics)
 
 
 @router.get("/attention-vehicles", response_model=list[AttentionVehicleOut])
 def attention_vehicles(
+    response: Response,
     minimum_severity: str = Query(default="medium", pattern="^(low|medium|high)$"),
     date_filter: date | None = Query(default=None, alias="date"),
     limit: int = Query(default=20, ge=1, le=50),
     db: Session = Depends(get_db),
-    client_id: str = Depends(require_mcp_service_token),
+    client: McpClientContext = Depends(require_mcp_service_token),
+    correlation_id: str = Depends(get_correlation_id),
 ) -> list[AttentionVehicleOut]:
     results = list_attention_vehicles(
         db, minimum_severity=minimum_severity, on_or_before=date_filter, limit=limit
     )
+    _set_trace_headers(response, correlation_id, client.tenant)
     _audit_service_request(
-        db, client_id=client_id, tool="mobilityops_list_attention_vehicles", status_label="ok"
+        db,
+        reported_client_id=client.reported_client_id,
+        tool="fleet_ops_list_attention_vehicles",
+        status_label="ok",
+        correlation_id=correlation_id,
     )
     return [AttentionVehicleOut(**r) for r in results]
 
@@ -72,16 +122,20 @@ def attention_vehicles(
 @router.get("/vehicles/{vehicle_ref}", response_model=McpVehicleDetailOut)
 def vehicle_details(
     vehicle_ref: str,
+    response: Response,
     db: Session = Depends(get_db),
-    client_id: str = Depends(require_mcp_service_token),
+    client: McpClientContext = Depends(require_mcp_service_token),
+    correlation_id: str = Depends(get_correlation_id),
 ) -> McpVehicleDetailOut:
+    _set_trace_headers(response, correlation_id, client.tenant)
     vehicle = db.scalar(select(Vehicle).where(Vehicle.public_ref == vehicle_ref))
     if vehicle is None:
         _audit_service_request(
             db,
-            client_id=client_id,
+            reported_client_id=client.reported_client_id,
-            tool="mobilityops_get_vehicle_details",
+            tool="fleet_ops_get_vehicle_details",
             status_label="not_found",
+            correlation_id=correlation_id,
         )
         raise AppError("VEHICLE_NOT_FOUND", "Vehicle not found.", status_code=404)
 
@@ -99,7 +153,11 @@ def vehicle_details(
     )
 
     _audit_service_request(
-        db, client_id=client_id, tool="mobilityops_get_vehicle_details", status_label="ok"
+        db,
+        reported_client_id=client.reported_client_id,
+        tool="fleet_ops_get_vehicle_details",
+        status_label="ok",
+        correlation_id=correlation_id,
     )
     return McpVehicleDetailOut(
         public_ref=vehicle.public_ref,
@@ -118,17 +176,29 @@ def vehicle_details(
 @router.post("/search-knowledge", response_model=GroundedAnswer)
 def search_knowledge(
     body: McpKnowledgeSearchRequest,
+    response: Response,
     db: Session = Depends(get_db),
-    client_id: str = Depends(require_mcp_service_token),
+    client: McpClientContext = Depends(require_mcp_service_token),
+    correlation_id: str = Depends(get_correlation_id),
 ) -> GroundedAnswer:
     provider = get_knowledge_provider()
-    correlation_id = str(uuid.uuid4())
+    answer = provider.ask(body.question, correlation_id, language=body.locale)
-    answer = provider.ask(body.question, correlation_id)
+    source_count_available = len(answer.sources)
     answer.sources = answer.sources[: body.max_sources]
+    _set_trace_headers(response, correlation_id, client.tenant)
+    response.headers["X-Sources-Available"] = str(source_count_available)
+    response.headers["X-Sources-Returned"] = str(len(answer.sources))
     _audit_service_request(
         db,
-        client_id=client_id,
+        reported_client_id=client.reported_client_id,
-        tool="mobilityops_search_knowledge",
+        tool="fleet_ops_search_knowledge",
         status_label=answer.evidence_state,
+        correlation_id=correlation_id,
+        metadata={
+            "tenant": client.tenant,
+            "locale": body.locale,
+            "sources_available": source_count_available,
+            "sources_returned": len(answer.sources),
+        },
     )
     return answer

backend/app/api/routers/observability.py

@@ -0,0 +1,46 @@
+from __future__ import annotations
+
+import hmac
+
+from fastapi import APIRouter, Header, HTTPException
+from fastapi.responses import Response
+from prometheus_client import CONTENT_TYPE_LATEST, generate_latest
+from sqlalchemy import func, select
+
+from app.core.config import get_settings
+from app.core.db import SessionLocal
+from app.core.observability import OUTBOX_EVENTS
+from app.models.outbox import DELIVERY_STATUSES, DEMO_SCENARIO_ERROR_CODE, OutboxEvent
+
+router = APIRouter(tags=["observability"])
+settings = get_settings()
+
+
+def _refresh_database_metrics() -> None:
+    with SessionLocal() as db:
+        rows = db.execute(
+            select(
+                OutboxEvent.delivery_status,
+                (OutboxEvent.last_error_code == DEMO_SCENARIO_ERROR_CODE).label("demo"),
+                func.count(),
+            ).group_by(OutboxEvent.delivery_status, "demo")
+        ).all()
+    OUTBOX_EVENTS.clear()
+    # Keep every time series present even when a state currently contains no rows.
+    # Stable zero-valued series make dashboards and alerts deterministic after resets,
+    # restores and fresh installations instead of turning "zero" into "no data".
+    for scenario in ("synthetic", "operational"):
+        for status in DELIVERY_STATUSES:
+            OUTBOX_EVENTS.labels(scenario, status).set(0)
+    for status, is_demo, count in rows:
+        OUTBOX_EVENTS.labels("synthetic" if is_demo else "operational", str(status)).set(count)
+
+
+@router.get("/metrics", include_in_schema=False)
+def metrics(authorization: str | None = Header(default=None)) -> Response:
+    if settings.metrics_bearer_token:
+        supplied = authorization.removeprefix("Bearer ") if authorization else ""
+        if not hmac.compare_digest(supplied, settings.metrics_bearer_token):
+            raise HTTPException(status_code=401, detail="Metrics token required")
+    _refresh_database_metrics()
+    return Response(content=generate_latest(), media_type=CONTENT_TYPE_LATEST)

backend/app/api/routers/privacy.py

@@ -0,0 +1,166 @@
+from __future__ import annotations
+
+from datetime import UTC, datetime, timedelta
+
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import JSONResponse
+from sqlalchemy import func, or_, select
+from sqlalchemy.orm import Session
+
+from app.api.deps import get_db, require_operations_manager
+from app.core.config import get_settings
+from app.models.booking import Booking
+from app.models.customer import Customer
+from app.schemas import (
+    CurrentUser,
+    CustomerAnonymizeRequest,
+    CustomerAnonymizeResult,
+    PrivacyRetentionOut,
+)
+from app.services.audit import record_audit_event
+
+router = APIRouter(prefix="/api/v1/privacy", tags=["privacy"])
+settings = get_settings()
+
+
+def _retention_cutoff() -> datetime:
+    return datetime.now(UTC) - timedelta(days=settings.privacy_minimum_booking_retention_days)
+
+
+def _customer_is_eligible(db: Session, customer_id) -> bool:
+    blocking = db.scalar(
+        select(func.count())
+        .select_from(Booking)
+        .where(
+            Booking.customer_id == customer_id,
+            or_(
+                Booking.status.in_(("reserved", "active")),
+                Booking.ends_at > _retention_cutoff(),
+            ),
+        )
+    )
+    return not blocking
+
+
+@router.get("/retention", response_model=PrivacyRetentionOut)
+def retention_status(
+    db: Session = Depends(get_db),
+    _user: CurrentUser = Depends(require_operations_manager),
+) -> PrivacyRetentionOut:
+    customers = db.scalars(select(Customer)).all()
+    return PrivacyRetentionOut(
+        minimum_booking_retention_days=settings.privacy_minimum_booking_retention_days,
+        audit_retention_days=settings.privacy_audit_retention_days,
+        customers_total=len(customers),
+        customers_anonymized=sum(customer.anonymized_at is not None for customer in customers),
+        customers_eligible=sum(
+            customer.anonymized_at is None and _customer_is_eligible(db, customer.id)
+            for customer in customers
+        ),
+    )
+
+
+@router.get("/customers/{public_ref}/export")
+def export_customer_data(
+    public_ref: str,
+    db: Session = Depends(get_db),
+    actor: CurrentUser = Depends(require_operations_manager),
+) -> JSONResponse:
+    customer = db.scalar(select(Customer).where(Customer.public_ref == public_ref))
+    if customer is None:
+        raise HTTPException(status_code=404, detail="Customer not found")
+    bookings = db.scalars(
+        select(Booking).where(Booking.customer_id == customer.id).order_by(Booking.starts_at)
+    ).all()
+    payload = {
+        "generated_at": datetime.now(UTC).isoformat(),
+        "customer": {
+            "public_ref": customer.public_ref,
+            "first_name": customer.first_name,
+            "last_name": customer.last_name,
+            "email": customer.email,
+            "phone": customer.phone,
+            "postal_code": customer.postal_code,
+            "city": customer.city,
+            "date_of_birth": customer.date_of_birth.isoformat() if customer.date_of_birth else None,
+            "anonymized_at": customer.anonymized_at.isoformat() if customer.anonymized_at else None,
+        },
+        "bookings": [
+            {
+                "public_ref": booking.public_ref,
+                "starts_at": booking.starts_at.isoformat(),
+                "ends_at": booking.ends_at.isoformat(),
+                "status": booking.status,
+            }
+            for booking in bookings
+        ],
+    }
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=actor.display_name,
+        action="privacy_customer_exported",
+        entity_type="customer",
+        entity_id=customer.id,
+        metadata={"customer_ref": customer.public_ref, "booking_count": len(bookings)},
+    )
+    db.commit()
+    return JSONResponse(
+        payload,
+        headers={"Content-Disposition": f'attachment; filename="{public_ref}-privacy.json"'},
+    )
+
+
+@router.post("/customers/{public_ref}/anonymize", response_model=CustomerAnonymizeResult)
+def anonymize_customer(
+    public_ref: str,
+    body: CustomerAnonymizeRequest,
+    db: Session = Depends(get_db),
+    actor: CurrentUser = Depends(require_operations_manager),
+) -> CustomerAnonymizeResult:
+    customer = db.scalar(
+        select(Customer).where(Customer.public_ref == public_ref).with_for_update()
+    )
+    if customer is None:
+        raise HTTPException(status_code=404, detail="Customer not found")
+    if body.confirmation != public_ref:
+        raise HTTPException(
+            status_code=422, detail="Customer reference confirmation does not match"
+        )
+    if customer.anonymized_at is not None:
+        return CustomerAnonymizeResult(
+            public_ref=public_ref,
+            anonymized_at=customer.anonymized_at,
+            status="already_anonymized",
+        )
+    if not _customer_is_eligible(db, customer.id):
+        raise HTTPException(
+            status_code=409,
+            detail="Customer has an active/recent booking within the minimum retention period",
+        )
+    anonymized_at = datetime.now(UTC)
+    customer.first_name = "Anoniem"
+    customer.last_name = public_ref
+    customer.email = None
+    customer.phone = None
+    customer.postal_code = None
+    customer.city = None
+    customer.date_of_birth = None
+    customer.anonymized_at = anonymized_at
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=actor.display_name,
+        action="privacy_customer_anonymized",
+        entity_type="customer",
+        entity_id=customer.id,
+        before={"anonymized": False},
+        after={"anonymized": True},
+        metadata={"reason": body.reason, "customer_ref": public_ref},
+    )
+    db.commit()
+    return CustomerAnonymizeResult(
+        public_ref=public_ref,
+        anonymized_at=anonymized_at,
+        status="anonymized",
+    )

backend/app/api/routers/search.py

@@ -12,53 +12,87 @@ from app.schemas import CurrentUser, SearchResponse, SearchResultItem
 
 router = APIRouter(prefix="/api/v1/search", tags=["search"])
 
-# Static application sections. Manager-only sections are filtered by role, mirroring the
+# Static application sections. `id` is a stable code matching navigation.json's
-# same nav visibility rule Layout.tsx applies -- search must never surface a destination
+# `items.*` keys -- the frontend localizes both the section label and its one-line
-# the current role can't actually reach.
+# detail from `id`, so no English prose is sent over the wire (search.sections.<id> in
+# every locale; see docs/fleet-ops-correction/i18n-inventory.md). Manager-only sections
+# are filtered by role, mirroring the same nav visibility rule Layout.tsx applies --
+# search must never surface a destination the current role can't actually reach.
 _SECTIONS: list[dict] = [
     {
-        "label": "Overview",
+        "id": "overview",
-        "detail": "Operations dashboard",
         "link": "/dashboard",
-        "terms": ["overview", "dashboard", "readiness"],
+        # Search terms deliberately span all three supported UI languages (not just
+        # English) so a query never depends on the operator's selected locale.
+        "terms": ["overview", "dashboard", "readiness", "overzicht", "aperçu", "tableau de bord"],
     },
     {
-        "label": "Fleet",
+        "id": "fleet",
-        "detail": "Vehicle registry",
         "link": "/vehicles",
-        "terms": ["fleet", "vehicle", "vehicles"],
+        "terms": ["fleet", "vehicle", "vehicles", "wagenpark", "voertuig", "flotte", "véhicule"],
     },
     {
-        "label": "Bookings",
+        "id": "bookings",
-        "detail": "Rental bookings",
         "link": "/bookings",
-        "terms": ["booking", "bookings", "rental"],
+        "terms": [
+            "booking",
+            "bookings",
+            "rental",
+            "boeking",
+            "boekingen",
+            "verhuur",
+            "réservation",
+            "réservations",
+            "location",
+        ],
     },
     {
-        "label": "Data quality",
+        "id": "quality",
-        "detail": "Quality workbench",
         "link": "/data-quality",
-        "terms": ["quality", "data quality", "issues"],
+        "terms": [
+            "quality",
+            "data quality",
+            "issues",
+            "kwaliteit",
+            "datakwaliteit",
+            "problemen",
+            "qualité",
+            "problèmes",
+        ],
         "role": "operations_manager",
     },
     {
-        "label": "Knowledge",
+        "id": "knowledge",
-        "detail": "Procedure assistant",
         "link": "/knowledge",
-        "terms": ["knowledge", "procedures"],
+        "terms": ["knowledge", "procedures", "kennis", "procedures", "connaissances", "procédures"],
     },
     {
-        "label": "Integrations",
+        "id": "integrations",
-        "detail": "Automation and integration status",
         "link": "/automation",
-        "terms": ["automation", "integrations", "systems", "n8n"],
+        "terms": [
+            "automation",
+            "integrations",
+            "systems",
+            "n8n",
+            "automatisering",
+            "integraties",
+            "systemen",
+            "automatisation",
+            "intégrations",
+            "systèmes",
+        ],
         "role": "operations_manager",
     },
     {
-        "label": "Audit trail",
+        "id": "audit",
-        "detail": "Audit history",
         "link": "/audit",
-        "terms": ["audit", "history"],
+        "terms": ["audit", "history", "geschiedenis", "historique"],
+        "role": "operations_manager",
+    },
+    {
+        "id": "privacy",
+        "link": "/privacy",
+        "terms": ["privacy", "retention", "anonymise", "anonimiseren", "confidentialité"],
         "role": "operations_manager",
     },
 ]
@@ -83,8 +117,8 @@ def search(
             results.append(
                 SearchResultItem(
                     type="section",
-                    label=section["label"],
+                    label=section["id"],
-                    detail=section["detail"],
+                    detail_code=section["id"],
                     link=section["link"],
                 )
             )
@@ -108,19 +142,23 @@ def search(
             SearchResultItem(
                 type="vehicle",
                 label=v.public_ref,
-                detail=f"{v.make} {v.model} · {v.location}",
+                detail_code="vehicleSummary",
+                detail_params={"make": v.make, "model": v.model, "location": v.location},
                 link=f"/vehicles/{v.public_ref}",
             )
         )
 
     for b in db.scalars(
-        select(Booking).where(Booking.public_ref.ilike(like)).order_by(Booking.starts_at.desc()).limit(5)
+        select(Booking)
+        .where(Booking.public_ref.ilike(like))
+        .order_by(Booking.starts_at.desc())
+        .limit(5)
     ).all():
         results.append(
             SearchResultItem(
                 type="booking",
                 label=b.public_ref,
-                detail=b.status,
+                detail_code=b.status,
                 link=f"/bookings/{b.public_ref}",
             )
         )
@@ -138,7 +176,7 @@ def search(
                 SearchResultItem(
                     type="data_quality_issue",
                     label=i.public_ref,
-                    detail=i.rule_type.replace("_", " "),
+                    detail_code=i.rule_type,
                     link=f"/data-quality/{i.public_ref}",
                 )
             )

backend/app/api/routers/users.py

@@ -0,0 +1,100 @@
+import uuid
+
+from fastapi import APIRouter, Depends, HTTPException
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from app.api.deps import get_db, require_operations_manager
+from app.core.security import hash_password
+from app.models.user import User
+from app.schemas import CreateUserRequest, CurrentUser, UpdateUserRequest, UserOut
+from app.services.audit import record_audit_event
+
+router = APIRouter(prefix="/api/v1/users", tags=["users"])
+
+
+def _to_out(user: User) -> UserOut:
+    return UserOut(
+        public_ref=user.public_ref,
+        email=user.email,
+        display_name=user.display_name,
+        role=user.role,  # type: ignore[arg-type]
+        active=user.active,
+    )
+
+
+@router.get("", response_model=list[UserOut])
+def list_users(
+    db: Session = Depends(get_db),
+    _manager: CurrentUser = Depends(require_operations_manager),
+) -> list[UserOut]:
+    return [_to_out(user) for user in db.scalars(select(User).order_by(User.display_name)).all()]
+
+
+@router.post("", response_model=UserOut, status_code=201)
+def create_user(
+    body: CreateUserRequest,
+    db: Session = Depends(get_db),
+    manager: CurrentUser = Depends(require_operations_manager),
+) -> UserOut:
+    email = body.email.strip().lower()
+    if db.scalar(select(User.id).where(User.email == email)) is not None:
+        raise HTTPException(status_code=409, detail="A user with this email already exists")
+    user = User(
+        public_ref=f"USR-{uuid.uuid4().hex[:8].upper()}",
+        email=email,
+        password_hash=hash_password(body.password),
+        display_name=body.display_name.strip(),
+        role=body.role,
+        active=True,
+    )
+    db.add(user)
+    db.flush()
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=manager.display_name,
+        action="user_created",
+        entity_type="user",
+        entity_id=user.id,
+        after={"public_ref": user.public_ref, "role": user.role, "active": user.active},
+    )
+    db.commit()
+    return _to_out(user)
+
+
+@router.patch("/{public_ref}", response_model=UserOut)
+def update_user(
+    public_ref: str,
+    body: UpdateUserRequest,
+    db: Session = Depends(get_db),
+    manager: CurrentUser = Depends(require_operations_manager),
+) -> UserOut:
+    user = db.scalar(select(User).where(User.public_ref == public_ref).with_for_update())
+    if user is None:
+        raise HTTPException(status_code=404, detail="User not found")
+    if user.public_ref == manager.public_ref and body.active is False:
+        raise HTTPException(status_code=409, detail="You cannot deactivate your own account")
+    if user.public_ref == manager.public_ref and body.role not in (None, "operations_manager"):
+        raise HTTPException(status_code=409, detail="You cannot remove your own manager role")
+    before = {"display_name": user.display_name, "role": user.role, "active": user.active}
+    if body.display_name is not None:
+        user.display_name = body.display_name.strip()
+    if body.role is not None:
+        user.role = body.role
+    if body.active is not None:
+        user.active = body.active
+    if body.password is not None:
+        user.password_hash = hash_password(body.password)
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=manager.display_name,
+        action="user_updated",
+        entity_type="user",
+        entity_id=user.id,
+        before=before,
+        after={"display_name": user.display_name, "role": user.role, "active": user.active},
+    )
+    db.commit()
+    return _to_out(user)

backend/app/api/routers/vehicles.py

@@ -1,10 +1,13 @@
 from __future__ import annotations
 
+import uuid
+from datetime import UTC, datetime
+
 from fastapi import APIRouter, Depends, HTTPException, Query
-from sqlalchemy import select
+from sqlalchemy import func, or_, select
 from sqlalchemy.orm import Session
 
-from app.api.deps import get_current_user, get_db
+from app.api.deps import get_current_user, get_db, require_operations_manager
 from app.models.booking import Booking
 from app.models.customer import Customer
 from app.models.data_quality import DataQualityIssue
@@ -13,13 +16,17 @@ from app.models.maintenance import MaintenanceRecord
 from app.models.vehicle import Vehicle
 from app.schemas import (
     BookingSummaryOut,
+    CreateMaintenanceRequest,
     CurrentUser,
     DataQualityIssueOut,
     InspectionOut,
     MaintenanceOut,
+    ReleaseVehicleRequest,
     VehicleDetailOut,
     VehicleOut,
+    VehiclePageOut,
 )
+from app.services.audit import record_audit_event
 
 router = APIRouter(prefix="/api/v1/vehicles", tags=["vehicles"])
 
@@ -34,19 +41,61 @@ def _attention_vehicle_ids(db: Session) -> set:
     return set(rows)
 
 
-@router.get("", response_model=list[VehicleOut])
+@router.get("", response_model=list[VehicleOut] | VehiclePageOut)
 def list_vehicles(
     status: str | None = Query(default=None),
     attention_only: bool = Query(default=False),
+    location: str | None = Query(default=None, min_length=1, max_length=120),
+    query: str | None = Query(default=None, min_length=1, max_length=100),
+    page: int | None = Query(default=None, ge=1),
+    page_size: int = Query(default=25, ge=1, le=25),
     db: Session = Depends(get_db),
     _user: CurrentUser = Depends(get_current_user),
-) -> list[VehicleOut]:
+) -> list[VehicleOut] | VehiclePageOut:
     stmt = select(Vehicle).order_by(Vehicle.public_ref)
     if status:
         stmt = stmt.where(Vehicle.operational_status == status)
-    vehicles = db.scalars(stmt).all()
+    if location:
+        stmt = stmt.where(Vehicle.location.ilike(location.strip()))
+    if query:
+        term = f"%{query.strip()}%"
+        stmt = stmt.where(
+            or_(
+                Vehicle.public_ref.ilike(term),
+                Vehicle.make.ilike(term),
+                Vehicle.model.ilike(term),
+                Vehicle.location.ilike(term),
+                Vehicle.registration_number.ilike(term),
+            )
+        )
     attention_ids = _attention_vehicle_ids(db)
-    out = [
+    if attention_only:
+        stmt = stmt.where(
+            or_(
+                Vehicle.id.in_(attention_ids),
+                Vehicle.operational_status == "blocked",
+                Vehicle.next_service_km <= Vehicle.odometer_km,
+            )
+        )
+    total = db.scalar(select(func.count()).select_from(stmt.subquery())) or 0
+    page_number = page or 1
+    vehicles = db.scalars(
+        stmt if page is None else stmt.offset((page_number - 1) * page_size).limit(page_size)
+    ).all()
+    vehicle_ids = [vehicle.id for vehicle in vehicles]
+    next_bookings: dict[uuid.UUID, Booking] = {}
+    if vehicle_ids:
+        for booking in db.scalars(
+            select(Booking)
+            .where(
+                Booking.vehicle_id.in_(vehicle_ids),
+                Booking.status == "reserved",
+                Booking.starts_at >= datetime.now(UTC),
+            )
+            .order_by(Booking.starts_at.asc())
+        ).all():
+            next_bookings.setdefault(booking.vehicle_id, booking)
+    items = [
         VehicleOut(
             public_ref=v.public_ref,
             make=v.make,
@@ -58,13 +107,36 @@ def list_vehicles(
             odometer_km=v.odometer_km,
             next_service_km=v.next_service_km,
             active=v.active,
-            attention=v.id in attention_ids or v.operational_status == "blocked",
+            attention=(
+                v.id in attention_ids
+                or v.operational_status == "blocked"
+                or v.next_service_km <= v.odometer_km
+            ),
+            attention_reason=(
+                "blocked_status"
+                if v.operational_status == "blocked"
+                else "service_due"
+                if v.next_service_km <= v.odometer_km
+                else "data_quality"
+                if v.id in attention_ids
+                else None
+            ),
+            service_remaining_km=v.next_service_km - v.odometer_km,
+            next_booking_ref=(next_bookings[v.id].public_ref if v.id in next_bookings else None),
+            next_booking_at=(next_bookings[v.id].starts_at if v.id in next_bookings else None),
         )
         for v in vehicles
     ]
-    if attention_only:
+    if page is None:
-        out = [v for v in out if v.attention]
+        return items
-    return out
+    total_pages = max(1, (total + page_size - 1) // page_size)
+    return VehiclePageOut(
+        items=items,
+        page=min(page_number, total_pages),
+        page_size=page_size,
+        total=total,
+        total_pages=total_pages,
+    )
 
 
 @router.get("/{public_ref}", response_model=VehicleDetailOut)
@@ -98,6 +170,14 @@ def get_vehicle(
     ).all()
 
     booking_by_id = {b.id: b.public_ref for b in bookings}
+    next_booking = next(
+        (
+            booking
+            for booking in sorted(bookings, key=lambda item: item.starts_at)
+            if booking.status == "reserved" and booking.starts_at >= datetime.now(UTC)
+        ),
+        None,
+    )
 
     attention_ids = _attention_vehicle_ids(db)
     return VehicleDetailOut(
@@ -111,7 +191,23 @@ def get_vehicle(
         odometer_km=vehicle.odometer_km,
         next_service_km=vehicle.next_service_km,
         active=vehicle.active,
-        attention=vehicle.id in attention_ids or vehicle.operational_status == "blocked",
+        attention=(
+            vehicle.id in attention_ids
+            or vehicle.operational_status == "blocked"
+            or vehicle.next_service_km <= vehicle.odometer_km
+        ),
+        attention_reason=(
+            "blocked_status"
+            if vehicle.operational_status == "blocked"
+            else "service_due"
+            if vehicle.next_service_km <= vehicle.odometer_km
+            else "data_quality"
+            if vehicle.id in attention_ids
+            else None
+        ),
+        service_remaining_km=vehicle.next_service_km - vehicle.odometer_km,
+        next_booking_ref=next_booking.public_ref if next_booking else None,
+        next_booking_at=next_booking.starts_at if next_booking else None,
         bookings=[
             BookingSummaryOut(
                 public_ref=b.public_ref,
@@ -157,8 +253,121 @@ def get_vehicle(
                 status=q.status,
                 evidence=q.evidence_json,
                 detected_at=q.detected_at,
+                due_at=q.due_at,
+                assigned_to_ref=(q.assigned_to_user.public_ref if q.assigned_to_user else None),
+                assigned_to_name=(q.assigned_to_user.display_name if q.assigned_to_user else None),
+                overdue=(
+                    q.status == "open" and q.due_at is not None and q.due_at < datetime.now(UTC)
+                ),
                 resolved_at=q.resolved_at,
             )
             for q in issues
         ],
     )
+
+
+@router.post("/{public_ref}/maintenance", response_model=MaintenanceOut, status_code=201)
+def create_maintenance_record(
+    public_ref: str,
+    body: CreateMaintenanceRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(require_operations_manager),
+) -> MaintenanceOut:
+    vehicle = db.scalar(select(Vehicle).where(Vehicle.public_ref == public_ref).with_for_update())
+    if vehicle is None:
+        raise HTTPException(status_code=404, detail="Vehicle not found")
+    record = MaintenanceRecord(
+        public_ref=f"MAINT-{uuid.uuid4().hex[:8].upper()}",
+        vehicle_id=vehicle.id,
+        occurred_at=body.occurred_at,
+        odometer_km=body.odometer_km,
+        category=body.category,
+        summary=body.summary.strip(),
+    )
+    db.add(record)
+    vehicle.odometer_km = max(vehicle.odometer_km, body.odometer_km)
+    if body.next_service_km is not None:
+        if body.next_service_km < vehicle.odometer_km:
+            raise HTTPException(status_code=422, detail="Next service must not be below odometer")
+        vehicle.next_service_km = body.next_service_km
+    if body.mark_maintenance:
+        vehicle.operational_status = "maintenance"
+    vehicle.version += 1
+    db.flush()
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=user.display_name,
+        action="maintenance_record_created",
+        entity_type="vehicle",
+        entity_id=vehicle.id,
+        after={"maintenance_ref": record.public_ref, "status": vehicle.operational_status},
+    )
+    db.commit()
+    return MaintenanceOut(
+        public_ref=record.public_ref,
+        occurred_at=record.occurred_at,
+        odometer_km=record.odometer_km,
+        category=record.category,
+        summary=record.summary,
+    )
+
+
+@router.post("/{public_ref}/release", response_model=VehicleOut)
+def release_vehicle(
+    public_ref: str,
+    body: ReleaseVehicleRequest,
+    db: Session = Depends(get_db),
+    user: CurrentUser = Depends(require_operations_manager),
+) -> VehicleOut:
+    vehicle = db.scalar(select(Vehicle).where(Vehicle.public_ref == public_ref).with_for_update())
+    if vehicle is None:
+        raise HTTPException(status_code=404, detail="Vehicle not found")
+    if vehicle.operational_status not in {"cleaning", "maintenance", "blocked"}:
+        raise HTTPException(status_code=409, detail="Vehicle does not require release")
+    active_booking = db.scalar(
+        select(Booking.id).where(Booking.vehicle_id == vehicle.id, Booking.status == "active")
+    )
+    open_high_issue = db.scalar(
+        select(DataQualityIssue.id).where(
+            DataQualityIssue.entity_type == "vehicle",
+            DataQualityIssue.entity_id == vehicle.id,
+            DataQualityIssue.status == "open",
+            DataQualityIssue.severity == "high",
+        )
+    )
+    if active_booking is not None or open_high_issue is not None:
+        raise HTTPException(status_code=409, detail="Vehicle still has a blocking condition")
+    before = {"status": vehicle.operational_status}
+    vehicle.operational_status = "available"
+    vehicle.version += 1
+    record_audit_event(
+        db,
+        actor_type="user",
+        actor_label=user.display_name,
+        action="vehicle_released",
+        entity_type="vehicle",
+        entity_id=vehicle.id,
+        before=before,
+        after={"status": "available", "reason": body.reason.strip()},
+    )
+    db.commit()
+    return VehicleOut(
+        public_ref=vehicle.public_ref,
+        make=vehicle.make,
+        model=vehicle.model,
+        model_year=vehicle.model_year,
+        registration_number=vehicle.registration_number,
+        location=vehicle.location,
+        operational_status=vehicle.operational_status,
+        odometer_km=vehicle.odometer_km,
+        next_service_km=vehicle.next_service_km,
+        active=vehicle.active,
+        attention=vehicle.next_service_km <= vehicle.odometer_km,
+        attention_reason=(
+            "service_due" if vehicle.next_service_km <= vehicle.odometer_km else None
+        ),
+        service_remaining_km=vehicle.next_service_km - vehicle.odometer_km,
+        next_booking_ref=None,
+        next_booking_at=None,
+    )

backend/app/api/routers/workflows.py

@@ -8,7 +8,7 @@ from sqlalchemy.orm import Session
 
 from app.api.deps import get_db, require_operations_manager
 from app.core.errors import AppError
-from app.models.outbox import OutboxEvent
+from app.models.outbox import OutboxEvent, is_demo_scenario_failure
 from app.schemas import AutomationRunOut, CurrentUser
 from app.services.audit import record_audit_event
 
@@ -23,6 +23,8 @@ def _to_out(event: OutboxEvent) -> AutomationRunOut:
         status=event.delivery_status,
         attempts=event.attempts,
         last_error=event.last_error,
+        last_error_code=event.last_error_code,
+        is_demo_scenario=is_demo_scenario_failure(event),
         occurred_at=event.occurred_at,
     )
 
@@ -62,15 +64,27 @@ def retry_workflow(
             status_code=409,
         )
 
+    # Captured before the status flips, so the audit records what was actually retried.
+    was_demo_scenario = is_demo_scenario_failure(event)
+
     event.delivery_status = "pending"
     event.next_attempt_at = None
+    # The retry itself is real either way: the event goes back on the outbox and the
+    # dispatcher delivers it to the configured n8n webhook like any other. The only
+    # difference recorded here is *what* was retried -- a staged demo failure or a real
+    # one -- so the audit trail never implies a production incident was resolved when a
+    # prop was.
     record_audit_event(
         db,
         actor_type="user",
         actor_label=user.display_name,
         action="workflow_retry",
         entity_type="outbox_event",
-        metadata={"event_id": event_id, "previous_attempts": event.attempts},
+        metadata={
+            "event_id": event_id,
+            "previous_attempts": event.attempts,
+            "demo_scenario": was_demo_scenario,
+        },
     )
     db.commit()
     return _to_out(event)

backend/app/core/config.py

@@ -2,6 +2,12 @@ from functools import lru_cache
 
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
+# The visible product name is fixed and never translated or configured per-deployment --
+# see docs/fleet-ops-correction/current-gap-audit.md section 1. Internal identifiers
+# (package name, Compose project, database name, repository) intentionally remain
+# "mobilityops"; this constant is only for user-facing surfaces (e.g. the OpenAPI title).
+PRODUCT_NAME = "Fleet Ops"
+
 
 class Settings(BaseSettings):
     model_config = SettingsConfigDict(env_file=".env", extra="ignore")
@@ -15,8 +21,15 @@ class Settings(BaseSettings):
     ragcore_workspace: str = "mobilityops"
     ragcore_collection: str = "internal-procedures"
     ragcore_api_token: str = ""
+    ragcore_space_id: str = ""
     ragcore_http_timeout_seconds: float = 5.0
+    ragcore_answers_circuit_breaker_seconds: float = 60.0
+    # Search fallback is only labelled grounded above this explicit retrieval threshold.
+    # RAGcore's fused score is reciprocal-rank based (top ranks are ~1/61), so this
+    # accepts only leading results while still rejecting absent and low-ranked evidence.
+    ragcore_min_search_score: float = 0.016
     n8n_webhook_url: str = "http://n8n:5678/webhook/mobilityops-return"
+    n8n_webhook_trigger_token: str = "replace-me-n8n-webhook-trigger-token"
     n8n_callback_token: str = "replace-me-n8n-callback-token"
     n8n_dispatch_enabled: bool = True
     n8n_dispatch_interval_seconds: float = 3.0
@@ -31,12 +44,81 @@ class Settings(BaseSettings):
     knowledge_dir: str = "/app/knowledge/procedures"
     mcp_hub_service_token: str = "replace-me-mcp-hub-token"
     mcp_hub_registration_enabled: bool = False
+    # MCP Hub's own registration is catalog-driven on the Hub side (the Hub reconciles
+    # its catalog into the gateway; Fleet Ops never pushes a registration call), so
+    # these are only used for an honest reachability health check, not self-registration.
+    mcp_hub_base_url: str = ""
+    mcp_provider_id: str = "fleet-ops"
     cors_allow_origins: str = "http://localhost:1228"
     demo_organization_name: str = "Northstar Mobility"
     demo_timezone: str = "Europe/Brussels"
     demo_allow_reset: bool = True
+    demo_reset_cooldown_seconds: int = 60
+    mcp_hub_health_cache_seconds: int = 60
+    initial_admin_email: str = ""
+    initial_admin_password: str = ""
+    initial_admin_display_name: str = "Operations Manager"
+    mobilityops_public_url: str = "http://localhost:1228"
+    oidc_enabled: bool = False
+    oidc_provider_name: str = "Organisatieaccount"
+    oidc_issuer_url: str = ""
+    oidc_client_id: str = ""
+    oidc_client_secret: str = ""
+    oidc_redirect_uri: str = ""
+    oidc_allowed_email_domains: str = ""
+    oidc_auto_provision: bool = True
+    oidc_default_role: str = "rental_employee"
+    log_level: str = "INFO"
+    # Failed password logins per client IP before a temporary 429 (0 disables).
+    login_max_failures: int = 10
+    login_failure_window_seconds: int = 900
+    knowledge_max_requests: int = 30
+    knowledge_rate_limit_window_seconds: int = 60
+    metrics_bearer_token: str = ""
+    privacy_minimum_booking_retention_days: int = 30
+    privacy_audit_retention_days: int = 2555
+    privacy_audit_export_max_rows: int = 10000
+
+
+# Secrets that guard *inbound* trust (session cookies, service callbacks). Running
+# production with any of these at their placeholder value means forged sessions or
+# unauthenticated writes, so startup refuses.
+INSECURE_DEFAULT_SECRETS: tuple[tuple[str, str], ...] = (
+    ("app_secret", "replace-in-production"),
+    ("n8n_callback_token", "replace-me-n8n-callback-token"),
+    ("mcp_hub_service_token", "replace-me-mcp-hub-token"),
+)
+
+
+def insecure_default_secrets(settings: "Settings") -> list[str]:
+    """Return the names of secret settings that still carry their placeholder value.
+
+    MCP routes are always mounted, independently of the Hub reachability-status flag, so
+    their inbound token must always be non-placeholder in production.
+    """
+    insecure: list[str] = []
+    for name, placeholder in INSECURE_DEFAULT_SECRETS:
+        value = getattr(settings, name)
+        if not value or value == placeholder or value.startswith("replace-me"):
+            insecure.append(name)
+    return insecure
 
 
 @lru_cache
 def get_settings() -> Settings:
-    return Settings()
+    settings = Settings()
+    if settings.mobilityops_env.lower() == "production":
+        insecure = insecure_default_secrets(settings)
+        if insecure:
+            # Refuse to boot rather than run production with forgeable session cookies
+            # or guessable service tokens. Development/test/demo keep the defaults.
+            raise RuntimeError(
+                "Refusing to start in production with placeholder secrets: "
+                + ", ".join(insecure)
+                + ". Set real values in the environment (see .env.example)."
+            )
+        if not settings.mobilityops_public_url.lower().startswith("https://"):
+            raise RuntimeError("Production MOBILITYOPS_PUBLIC_URL must use HTTPS.")
+        if not settings.session_cookie_secure:
+            raise RuntimeError("Production SESSION_COOKIE_SECURE must be true.")
+    return settings

backend/app/core/observability.py

@@ -0,0 +1,113 @@
+from __future__ import annotations
+
+import json
+import logging
+import time
+import uuid
+from contextvars import ContextVar
+from datetime import UTC, datetime
+
+from fastapi import Request
+from prometheus_client import Counter, Gauge, Histogram
+
+correlation_id_context: ContextVar[str] = ContextVar("correlation_id", default="")
+
+HTTP_REQUESTS = Counter(
+    "mobilityops_http_requests_total",
+    "Completed MobilityOps HTTP requests.",
+    ("method", "route", "status"),
+)
+HTTP_DURATION = Histogram(
+    "mobilityops_http_request_duration_seconds",
+    "MobilityOps HTTP request duration.",
+    ("method", "route"),
+    buckets=(0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10),
+)
+HTTP_IN_PROGRESS = Gauge(
+    "mobilityops_http_requests_in_progress",
+    "MobilityOps HTTP requests currently executing.",
+)
+OUTBOX_EVENTS = Gauge(
+    "mobilityops_outbox_events",
+    "Persisted outbox events by state and scenario type.",
+    ("scenario", "status"),
+)
+DATABASE_READY = Gauge(
+    "mobilityops_database_ready",
+    "Whether the canonical PostgreSQL database answered the most recent readiness probe.",
+)
+KNOWLEDGE_PROVIDER_REQUESTS = Counter(
+    "mobilityops_knowledge_provider_requests_total",
+    "RAGcore adapter requests by stage and outcome.",
+    ("stage", "outcome"),
+)
+KNOWLEDGE_RETRIEVAL_SCORE = Histogram(
+    "mobilityops_knowledge_retrieval_score",
+    "Observed RAGcore fused/rerank retrieval scores.",
+    buckets=(0.005, 0.01, 0.015, 0.016, 0.0162, 0.0164, 0.02, 0.05, 0.1, 0.5, 1.0),
+)
+
+
+class JsonFormatter(logging.Formatter):
+    def format(self, record: logging.LogRecord) -> str:
+        payload: dict[str, object] = {
+            "timestamp": datetime.now(UTC).isoformat(),
+            "level": record.levelname.lower(),
+            "logger": record.name,
+            "message": record.getMessage(),
+        }
+        correlation_id = correlation_id_context.get()
+        if correlation_id:
+            payload["correlation_id"] = correlation_id
+        for key in ("method", "path", "status_code", "duration_ms", "client_ip"):
+            value = getattr(record, key, None)
+            if value is not None:
+                payload[key] = value
+        if record.exc_info:
+            payload["exception"] = self.formatException(record.exc_info)
+        return json.dumps(payload, separators=(",", ":"), default=str)
+
+
+def configure_logging(level: str) -> None:
+    handler = logging.StreamHandler()
+    handler.setFormatter(JsonFormatter())
+    root = logging.getLogger()
+    root.handlers = [handler]
+    root.setLevel(level.upper())
+
+
+def correlation_id_for(request: Request) -> str:
+    candidate = request.headers.get("X-Correlation-Id", "").strip()
+    try:
+        return str(uuid.UUID(candidate)) if candidate else str(uuid.uuid4())
+    except ValueError:
+        return str(uuid.uuid4())
+
+
+UNMATCHED_ROUTE_LABEL = "<unmatched>"
+
+
+def route_label(request: Request) -> str:
+    """Return the route *template* for metrics labels.
+
+    Unmatched paths (404 probes, scanners) must not become their own label value:
+    every distinct URL would otherwise create a new Prometheus time series and the
+    metric cardinality would grow without bound.
+    """
+    route = request.scope.get("route")
+    path = getattr(route, "path", None)
+    return str(path) if path else UNMATCHED_ROUTE_LABEL
+
+
+def request_started() -> float:
+    HTTP_IN_PROGRESS.inc()
+    return time.perf_counter()
+
+
+def request_finished(request: Request, status_code: int, started_at: float) -> float:
+    duration = time.perf_counter() - started_at
+    route = route_label(request)
+    HTTP_REQUESTS.labels(request.method, route, str(status_code)).inc()
+    HTTP_DURATION.labels(request.method, route).observe(duration)
+    HTTP_IN_PROGRESS.dec()
+    return duration

backend/app/core/ratelimit.py

@@ -0,0 +1,72 @@
+"""Small in-process failed-attempt limiter for credential endpoints.
+
+Fleet Ops runs as a single API process per deployment, so an in-memory sliding window
+is sufficient to blunt online password guessing (and the scrypt CPU amplification that
+comes with it) without adding Redis. Only *failed* attempts count, so legitimate users
+and the automated test suite are never throttled.
+"""
+
+from __future__ import annotations
+
+import threading
+import time
+from collections import deque
+
+
+class FailedAttemptLimiter:
+    def __init__(self, *, max_failures: int, window_seconds: float) -> None:
+        self.max_failures = max_failures
+        self.window_seconds = window_seconds
+        self._failures: dict[str, deque[float]] = {}
+        self._lock = threading.Lock()
+
+    def _prune(self, key: str, now: float) -> deque[float]:
+        bucket = self._failures.setdefault(key, deque())
+        cutoff = now - self.window_seconds
+        while bucket and bucket[0] <= cutoff:
+            bucket.popleft()
+        if not bucket:
+            self._failures.pop(key, None)
+        return bucket
+
+    def retry_after_seconds(self, key: str) -> int:
+        """Return >0 seconds to wait when the key is currently blocked, else 0."""
+        now = time.monotonic()
+        with self._lock:
+            bucket = self._prune(key, now)
+            if len(bucket) < self.max_failures:
+                return 0
+            return max(1, int(bucket[0] + self.window_seconds - now + 0.999))
+
+    def record_failure(self, key: str) -> None:
+        now = time.monotonic()
+        with self._lock:
+            self._prune(key, now)
+            self._failures.setdefault(key, deque()).append(now)
+
+    def reset(self, key: str) -> None:
+        with self._lock:
+            self._failures.pop(key, None)
+
+
+class SlidingWindowLimiter:
+    """Thread-safe request limiter where every accepted request consumes capacity."""
+
+    def __init__(self, *, max_requests: int, window_seconds: float) -> None:
+        self.max_requests = max_requests
+        self.window_seconds = window_seconds
+        self._requests: dict[str, deque[float]] = {}
+        self._lock = threading.Lock()
+
+    def consume(self, key: str) -> int:
+        """Record an accepted request, or return the seconds until capacity is available."""
+        now = time.monotonic()
+        with self._lock:
+            bucket = self._requests.setdefault(key, deque())
+            cutoff = now - self.window_seconds
+            while bucket and bucket[0] <= cutoff:
+                bucket.popleft()
+            if len(bucket) >= self.max_requests:
+                return max(1, int(bucket[0] + self.window_seconds - now + 0.999))
+            bucket.append(now)
+            return 0

backend/app/core/security.py

@@ -4,6 +4,7 @@ import base64
 import hashlib
 import hmac
 import json
+import os
 import time
 from dataclasses import dataclass
 
@@ -19,6 +20,7 @@ class SessionPayload:
     role: str
     display_name: str
     issued_at: int
+    session_id: str = ""
 
 
 def _sign(data: bytes) -> str:
@@ -26,6 +28,11 @@ def _sign(data: bytes) -> str:
     return base64.urlsafe_b64encode(digest).decode().rstrip("=")
 
 
+def session_token_hash(token: str) -> str:
+    """Return a non-reversible identifier safe to persist for token revocation."""
+    return hashlib.sha256(token.encode()).hexdigest()
+
+
 def create_session_token(payload: SessionPayload) -> str:
     body = json.dumps(payload.__dict__, separators=(",", ":")).encode()
     encoded_body = base64.urlsafe_b64encode(body).decode().rstrip("=")
@@ -50,3 +57,31 @@ def read_session_token(token: str) -> SessionPayload | None:
     if time.time() - payload.issued_at > settings.session_ttl_seconds:
         return None
     return payload
+
+
+def hash_password(password: str) -> str:
+    salt = os.urandom(16)
+    derived = hashlib.scrypt(password.encode(), salt=salt, n=2**14, r=8, p=1, dklen=32)
+    encoded_salt = base64.urlsafe_b64encode(salt).decode()
+    encoded_hash = base64.urlsafe_b64encode(derived).decode()
+    return f"scrypt$16384$8$1${encoded_salt}${encoded_hash}"
+
+
+def verify_password(password: str, encoded: str | None) -> bool:
+    if not encoded:
+        return False
+    try:
+        algorithm, n, r, p, salt, expected = encoded.split("$")
+        if algorithm != "scrypt":
+            return False
+        derived = hashlib.scrypt(
+            password.encode(),
+            salt=base64.urlsafe_b64decode(salt.encode()),
+            n=int(n),
+            r=int(r),
+            p=int(p),
+            dklen=32,
+        )
+        return hmac.compare_digest(derived, base64.urlsafe_b64decode(expected.encode()))
+    except (ValueError, TypeError):
+        return False

backend/app/main.py

@@ -1,13 +1,18 @@
+import logging
 import uuid
 from contextlib import asynccontextmanager
 
 from fastapi import FastAPI, HTTPException, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import JSONResponse
+from sqlalchemy import text
+from starlette.middleware.sessions import SessionMiddleware
 
 from app.api.routers import (
     audit,
+    auth,
     bookings,
+    customers,
     dashboard,
     data_quality,
     demo,
@@ -15,25 +20,92 @@ from app.api.routers import (
     integrations,
     knowledge,
     mcp_integrations,
+    observability,
+    privacy,
     search,
+    users,
     vehicles,
     workflows,
 )
-from app.core.config import get_settings
+from app.api.routers.auth import bootstrap_initial_admin
+from app.core.config import PRODUCT_NAME, get_settings
+from app.core.db import SessionLocal
 from app.core.errors import AppError, error_body
+from app.core.observability import (
+    DATABASE_READY,
+    configure_logging,
+    correlation_id_context,
+    correlation_id_for,
+    request_finished,
+    request_started,
+)
 from app.services.dispatcher import start_background_dispatcher, stop_background_dispatcher
 
 settings = get_settings()
+configure_logging(settings.log_level)
+request_logger = logging.getLogger("mobilityops.request")
 
 
 @asynccontextmanager
 async def lifespan(_app: FastAPI):
+    with SessionLocal() as db:
+        bootstrap_initial_admin(db)
     start_background_dispatcher()
     yield
     stop_background_dispatcher()
 
 
-app = FastAPI(title="MobilityOps API", version="0.1.0", lifespan=lifespan)
+production = settings.mobilityops_env.lower() == "production"
+app = FastAPI(
+    title=f"{PRODUCT_NAME} API",
+    version="0.1.0",
+    description=(
+        "Generated contract for Fleet Ops. The visible product name is Fleet Ops; "
+        "MobilityOps remains the technical repository and service identifier."
+    ),
+    servers=[{"url": "http://localhost:8128"}],
+    lifespan=lifespan,
+    docs_url=None if production else "/docs",
+    redoc_url=None if production else "/redoc",
+    openapi_url=None if production else "/openapi.json",
+)
+
+app.add_middleware(
+    SessionMiddleware,
+    secret_key=settings.app_secret,
+    session_cookie="mobilityops_oidc_state",
+    max_age=600,
+    same_site="lax",
+    https_only=settings.session_cookie_secure,
+)
+
+
+@app.middleware("http")
+async def request_observability(request: Request, call_next):
+    correlation_id = correlation_id_for(request)
+    request.state.correlation_id = correlation_id
+    token = correlation_id_context.set(correlation_id)
+    started_at = request_started()
+    status_code = 500
+    try:
+        response = await call_next(request)
+        status_code = response.status_code
+        response.headers["X-Correlation-Id"] = correlation_id
+        return response
+    finally:
+        duration = request_finished(request, status_code, started_at)
+        request_logger.info(
+            "request_completed",
+            extra={
+                "method": request.method,
+                "path": request.url.path,
+                "status_code": status_code,
+                "duration_ms": round(duration * 1000, 2),
+                "client_ip": request.client.host if request.client else None,
+            },
+        )
+        correlation_id_context.reset(token)
+
 
 app.add_middleware(
     CORSMiddleware,
@@ -45,23 +117,29 @@ app.add_middleware(
 
 
 @app.exception_handler(AppError)
-def handle_app_error(_request: Request, exc: AppError) -> JSONResponse:
+def handle_app_error(request: Request, exc: AppError) -> JSONResponse:
     return JSONResponse(
         status_code=exc.status_code,
-        content=error_body(exc.code, exc.message, exc.correlation_id, exc.details),
+        content=error_body(
+            exc.code,
+            exc.message,
+            request.state.correlation_id,
+            exc.details,
+        ),
     )
 
 
 @app.exception_handler(HTTPException)
-def handle_http_exception(_request: Request, exc: HTTPException) -> JSONResponse:
+def handle_http_exception(request: Request, exc: HTTPException) -> JSONResponse:
     return JSONResponse(
         status_code=exc.status_code,
         content=error_body(
             code=str(exc.status_code),
             message=str(exc.detail),
-            correlation_id=str(uuid.uuid4()),
+            correlation_id=getattr(request.state, "correlation_id", str(uuid.uuid4())),
             details={},
         ),
+        headers=exc.headers,
     )
 
 
@@ -70,6 +148,28 @@ def health() -> dict[str, str]:
     return {"status": "ok", "service": "mobilityops-api"}
 
 
+@app.get("/health/live")
+def liveness() -> dict[str, str]:
+    """Process liveness only; external dependencies deliberately do not affect it."""
+    return {"status": "ok", "service": "mobilityops-api"}
+
+
+@app.get("/health/ready")
+def readiness() -> JSONResponse:
+    """Traffic readiness: the API is useful only while its canonical database responds."""
+    try:
+        with SessionLocal() as db:
+            db.execute(text("SELECT 1"))
+    except Exception:  # noqa: BLE001 -- readiness must convert infrastructure errors to 503
+        DATABASE_READY.set(0)
+        return JSONResponse(
+            status_code=503,
+            content={"status": "not_ready", "service": "mobilityops-api", "database": "down"},
+        )
+    DATABASE_READY.set(1)
+    return JSONResponse(content={"status": "ready", "service": "mobilityops-api", "database": "up"})
+
+
 @app.get("/api/v1/system/status")
 def system_status() -> dict[str, object]:
     return {
@@ -77,13 +177,17 @@ def system_status() -> dict[str, object]:
         "environment": settings.mobilityops_env,
         "demo_mode": settings.mobilityops_demo_mode,
         "knowledge_provider": settings.knowledge_provider,
+        "oidc_enabled": auth.oidc_status().enabled,
+        "oidc_provider_name": auth.oidc_status().provider_name,
     }
 
 
 app.include_router(demo.router)
+app.include_router(auth.router)
 app.include_router(dashboard.router)
 app.include_router(vehicles.router)
 app.include_router(bookings.router)
+app.include_router(customers.router)
 app.include_router(audit.router)
 app.include_router(data_quality.router)
 app.include_router(workflows.router)
@@ -92,3 +196,6 @@ app.include_router(knowledge.router)
 app.include_router(mcp_integrations.router)
 app.include_router(search.router)
 app.include_router(integration_status.router)
+app.include_router(users.router)
+app.include_router(observability.router)
+app.include_router(privacy.router)

backend/app/models/__init__.py

@@ -7,6 +7,7 @@ from app.models.idempotency import IdempotencyRecord
 from app.models.inspection import Inspection
 from app.models.maintenance import MaintenanceRecord
 from app.models.outbox import OutboxEvent
+from app.models.revoked_session import RevokedSession
 from app.models.user import User
 from app.models.vehicle import Vehicle
 
@@ -20,6 +21,7 @@ __all__ = [
     "Inspection",
     "MaintenanceRecord",
     "OutboxEvent",
+    "RevokedSession",
     "User",
     "Vehicle",
 ]

backend/app/models/audit.py

@@ -1,7 +1,7 @@
 import uuid
 from datetime import datetime
 
-from sqlalchemy import DateTime, String
+from sqlalchemy import CheckConstraint, DateTime, Index, String
 from sqlalchemy.dialects.postgresql import JSONB, UUID
 from sqlalchemy.orm import Mapped, mapped_column
 
@@ -13,6 +13,11 @@ ACTOR_TYPES = ("user", "service", "system")
 
 class AuditEvent(UUIDPrimaryKeyMixin, Base):
     __tablename__ = "audit_events"
+    __table_args__ = (
+        CheckConstraint("actor_type IN ('user','service','system')", name="ck_audit_actor_type"),
+        Index("ix_audit_action_occurred", "action", "occurred_at"),
+        Index("ix_audit_entity", "entity_type", "entity_id"),
+    )
 
     actor_type: Mapped[str] = mapped_column(String(20), nullable=False)
     actor_id: Mapped[uuid.UUID | None] = mapped_column(UUID(as_uuid=True))

backend/app/models/booking.py

@@ -1,7 +1,7 @@
 import uuid
 from datetime import datetime
 
-from sqlalchemy import Boolean, DateTime, ForeignKey, Integer, String
+from sqlalchemy import Boolean, CheckConstraint, DateTime, ForeignKey, Index, Integer, String
 from sqlalchemy.dialects.postgresql import UUID
 from sqlalchemy.orm import Mapped, mapped_column
 
@@ -13,6 +13,22 @@ BOOKING_STATUSES = ("reserved", "active", "returned", "cancelled", "blocked")
 
 class Booking(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     __tablename__ = "bookings"
+    __table_args__ = (
+        CheckConstraint(
+            "status IN ('reserved','active','returned','cancelled','blocked')",
+            name="ck_bookings_status",
+        ),
+        CheckConstraint("ends_at > starts_at", name="ck_bookings_time_window"),
+        CheckConstraint(
+            "start_odometer_km IS NULL OR start_odometer_km >= 0",
+            name="ck_bookings_start_odometer",
+        ),
+        CheckConstraint(
+            "end_odometer_km IS NULL OR end_odometer_km >= 0",
+            name="ck_bookings_end_odometer",
+        ),
+        Index("ix_bookings_vehicle_status_window", "vehicle_id", "status", "starts_at", "ends_at"),
+    )
 
     public_ref: Mapped[str] = mapped_column(String(20), unique=True, nullable=False)
     customer_id: Mapped[uuid.UUID] = mapped_column(
@@ -26,4 +42,4 @@ class Booking(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     status: Mapped[str] = mapped_column(String(20), nullable=False)
     start_odometer_km: Mapped[int | None] = mapped_column(Integer)
     end_odometer_km: Mapped[int | None] = mapped_column(Integer)
-    requirements_complete: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
+    requirements_complete: Mapped[bool] = mapped_column(Boolean, nullable=False, default=False)

backend/app/models/customer.py

@@ -1,7 +1,7 @@
 import uuid
-from datetime import date
+from datetime import date, datetime
 
-from sqlalchemy import Date, ForeignKey, String
+from sqlalchemy import Date, DateTime, ForeignKey, String
 from sqlalchemy.dialects.postgresql import UUID
 from sqlalchemy.orm import Mapped, mapped_column
 
@@ -23,3 +23,4 @@ class Customer(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     merged_into_customer_id: Mapped[uuid.UUID | None] = mapped_column(
         UUID(as_uuid=True), ForeignKey("customers.id")
     )
+    anonymized_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), index=True)

backend/app/models/data_quality.py

@@ -1,13 +1,17 @@
 import uuid
 from datetime import datetime
+from typing import TYPE_CHECKING
 
-from sqlalchemy import DateTime, String
+from sqlalchemy import CheckConstraint, DateTime, ForeignKey, Index, String, text
 from sqlalchemy.dialects.postgresql import JSONB, UUID
-from sqlalchemy.orm import Mapped, mapped_column
+from sqlalchemy.orm import Mapped, mapped_column, relationship
 
 from app.core.db import Base
 from app.models.mixins import TimestampMixin, UUIDPrimaryKeyMixin
 
+if TYPE_CHECKING:
+    from app.models.user import User
+
 RULE_TYPES = (
     "possible_duplicate_customer",
     "missing_required_field",
@@ -21,6 +25,27 @@ ISSUE_STATUSES = ("open", "deferred", "resolved", "rejected")
 
 class DataQualityIssue(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     __tablename__ = "data_quality_issues"
+    __table_args__ = (
+        CheckConstraint(
+            "rule_type IN ('possible_duplicate_customer','missing_required_field',"
+            "'odometer_regression','booking_overlap','vehicle_status_conflict')",
+            name="ck_data_quality_rule_type",
+        ),
+        CheckConstraint("severity IN ('low','medium','high')", name="ck_data_quality_severity"),
+        CheckConstraint(
+            "status IN ('open','deferred','resolved','rejected')",
+            name="ck_data_quality_status",
+        ),
+        Index("ix_data_quality_work_queue", "status", "due_at", "severity"),
+        Index(
+            "uq_data_quality_one_open_condition",
+            "rule_type",
+            "entity_type",
+            "entity_id",
+            unique=True,
+            postgresql_where=text("status = 'open'"),
+        ),
+    )
 
     public_ref: Mapped[str] = mapped_column(String(20), unique=True, nullable=False)
     rule_type: Mapped[str] = mapped_column(String(40), nullable=False)
@@ -31,5 +56,10 @@ class DataQualityIssue(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     evidence_json: Mapped[dict] = mapped_column(JSONB, nullable=False, default=dict)
     proposed_action_json: Mapped[dict] = mapped_column(JSONB, nullable=False, default=dict)
     detected_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), nullable=False)
+    due_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), index=True)
+    assigned_to_user_id: Mapped[uuid.UUID | None] = mapped_column(
+        UUID(as_uuid=True), ForeignKey("users.id", ondelete="SET NULL"), index=True
+    )
+    assigned_to_user: Mapped["User | None"] = relationship(lazy="selectin")
     resolved_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
     resolved_by: Mapped[str | None] = mapped_column(String(120))

backend/app/models/idempotency.py

@@ -15,5 +15,8 @@ class IdempotencyRecord(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     booking_id: Mapped[uuid.UUID] = mapped_column(
         UUID(as_uuid=True), ForeignKey("bookings.id"), nullable=False
     )
+    # SHA-256 of the canonical request body. Replaying a key with a *different* body is
+    # a client bug and must be rejected instead of silently answered with the old result.
+    request_fingerprint: Mapped[str | None] = mapped_column(String(64), nullable=True)
     response_status: Mapped[int] = mapped_column(Integer, nullable=False)
     response_body: Mapped[dict] = mapped_column(JSONB, nullable=False)

backend/app/models/mixins.py

@@ -7,9 +7,7 @@ from sqlalchemy.orm import Mapped, mapped_column
 
 
 class UUIDPrimaryKeyMixin:
-    id: Mapped[uuid.UUID] = mapped_column(
+    id: Mapped[uuid.UUID] = mapped_column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
-        UUID(as_uuid=True), primary_key=True, default=uuid.uuid4
-    )
 
 
 class TimestampMixin:

backend/app/models/outbox.py

@@ -1,7 +1,7 @@
 import uuid
 from datetime import datetime
 
-from sqlalchemy import DateTime, Integer, String, Text
+from sqlalchemy import CheckConstraint, DateTime, Index, Integer, String, Text
 from sqlalchemy.dialects.postgresql import JSONB, UUID
 from sqlalchemy.orm import Mapped, mapped_column
 
@@ -10,9 +10,36 @@ from app.models.mixins import TimestampMixin
 
 DELIVERY_STATUSES = ("pending", "delivering", "succeeded", "failed")
 
+# The one delivery failure the demo seed deliberately plants (BK-H-0020, see
+# seed/workflow_runs.csv). It exists to show retry and audit working, so it must never
+# be read as an integration-health problem: it is a scripted prop, not evidence that
+# n8n is unhealthy. A dedicated error code -- rather than the generic
+# "connectionError" a real timeout produces -- is what lets every reader tell the two
+# apart without guessing from the message text.
+#
+# It is deliberately a `last_error_code` value and not a new column: the code is
+# already persisted, already surfaced to the UI, and already localizable, so no schema
+# change or migration is needed. A genuine later failure of this same event overwrites
+# the code with the real one, which is exactly right -- from that moment it *is* a real
+# failure.
+DEMO_SCENARIO_ERROR_CODE = "demoScenarioTimeout"
+
+
+def is_demo_scenario_failure(event: "OutboxEvent") -> bool:
+    """True for the prepared demo failure, false for every real one."""
+    return event.delivery_status == "failed" and event.last_error_code == DEMO_SCENARIO_ERROR_CODE
+
 
 class OutboxEvent(TimestampMixin, Base):
     __tablename__ = "outbox_events"
+    __table_args__ = (
+        CheckConstraint(
+            "delivery_status IN ('pending','delivering','succeeded','failed')",
+            name="ck_outbox_delivery_status",
+        ),
+        CheckConstraint("attempts >= 0", name="ck_outbox_attempts"),
+        Index("ix_outbox_delivery_next_attempt", "delivery_status", "next_attempt_at"),
+    )
 
     event_id: Mapped[uuid.UUID] = mapped_column(
         UUID(as_uuid=True), primary_key=True, default=uuid.uuid4
@@ -26,4 +53,9 @@ class OutboxEvent(TimestampMixin, Base):
     attempts: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
     next_attempt_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
     last_error: Mapped[str | None] = mapped_column(Text)
+    # Stable, localizable classification of last_error -- the frontend renders a
+    # localized summary from this code as the primary text and shows last_error itself
+    # only under "Technical details" (section 10 of docs/fleet-ops-correction/
+    # current-gap-audit.md). Kept alongside the raw message for backward compatibility.
+    last_error_code: Mapped[str | None] = mapped_column(String(60))
     external_run_id: Mapped[str | None] = mapped_column(String(120))

backend/app/models/revoked_session.py

@@ -0,0 +1,16 @@
+from datetime import datetime
+
+from sqlalchemy import DateTime, String
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.core.db import Base
+from app.models.mixins import TimestampMixin, UUIDPrimaryKeyMixin
+
+
+class RevokedSession(UUIDPrimaryKeyMixin, TimestampMixin, Base):
+    __tablename__ = "revoked_sessions"
+
+    token_hash: Mapped[str] = mapped_column(String(64), unique=True, index=True, nullable=False)
+    expires_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, index=True
+    )

backend/app/models/user.py

@@ -1,4 +1,4 @@
-from sqlalchemy import Boolean, String
+from sqlalchemy import Boolean, String, UniqueConstraint
 from sqlalchemy.orm import Mapped, mapped_column
 
 from app.core.db import Base
@@ -9,8 +9,15 @@ ROLES = ("operations_manager", "rental_employee")
 
 class User(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     __tablename__ = "users"
+    __table_args__ = (
+        UniqueConstraint("identity_provider", "external_subject", name="uq_user_external_identity"),
+    )
 
     public_ref: Mapped[str] = mapped_column(String(20), unique=True, nullable=False)
+    email: Mapped[str | None] = mapped_column(String(320), unique=True, nullable=True)
+    password_hash: Mapped[str | None] = mapped_column(String(512), nullable=True)
     display_name: Mapped[str] = mapped_column(String(120), nullable=False)
     role: Mapped[str] = mapped_column(String(30), nullable=False)
     active: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
+    identity_provider: Mapped[str | None] = mapped_column(String(80), nullable=True)
+    external_subject: Mapped[str | None] = mapped_column(String(255), nullable=True)

backend/app/models/vehicle.py

@@ -1,4 +1,4 @@
-from sqlalchemy import Boolean, Integer, String
+from sqlalchemy import Boolean, CheckConstraint, Integer, String
 from sqlalchemy.orm import Mapped, mapped_column
 
 from app.core.db import Base
@@ -9,6 +9,16 @@ OPERATIONAL_STATUSES = ("available", "rented", "cleaning", "maintenance", "block
 
 class Vehicle(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     __tablename__ = "vehicles"
+    __table_args__ = (
+        CheckConstraint(
+            "operational_status IN ('available','rented','cleaning','maintenance','blocked')",
+            name="ck_vehicles_operational_status",
+        ),
+        CheckConstraint("model_year BETWEEN 1900 AND 2100", name="ck_vehicles_model_year"),
+        CheckConstraint("odometer_km >= 0", name="ck_vehicles_odometer"),
+        CheckConstraint("next_service_km >= 0", name="ck_vehicles_next_service"),
+        CheckConstraint("version >= 1", name="ck_vehicles_version"),
+    )
 
     public_ref: Mapped[str] = mapped_column(String(20), unique=True, nullable=False)
     make: Mapped[str] = mapped_column(String(80), nullable=False)

backend/app/schemas.py

@@ -1,9 +1,10 @@
 from __future__ import annotations
 
+import uuid
 from datetime import datetime
 from typing import Annotated, Any, Literal
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, ConfigDict, Field
 
 Role = Literal["operations_manager", "rental_employee"]
 
@@ -12,6 +13,33 @@ class DemoLoginRequest(BaseModel):
     role: Role
 
 
+class PasswordLoginRequest(BaseModel):
+    email: str = Field(min_length=3, max_length=320)
+    password: str = Field(min_length=8, max_length=256)
+
+
+class UserOut(BaseModel):
+    public_ref: str
+    email: str | None
+    display_name: str
+    role: Role
+    active: bool
+
+
+class CreateUserRequest(BaseModel):
+    email: str = Field(min_length=3, max_length=320)
+    display_name: str = Field(min_length=2, max_length=120)
+    role: Role
+    password: str = Field(min_length=8, max_length=256)
+
+
+class UpdateUserRequest(BaseModel):
+    display_name: str | None = Field(default=None, min_length=2, max_length=120)
+    role: Role | None = None
+    active: bool | None = None
+    password: str | None = Field(default=None, min_length=8, max_length=256)
+
+
 class CurrentUser(BaseModel):
     public_ref: str
     display_name: str
@@ -30,6 +58,18 @@ class VehicleOut(BaseModel):
     next_service_km: int
     active: bool
     attention: bool = False
+    attention_reason: str | None = None
+    service_remaining_km: int
+    next_booking_ref: str | None = None
+    next_booking_at: datetime | None = None
+
+
+class VehiclePageOut(BaseModel):
+    items: list[VehicleOut]
+    page: int
+    page_size: int
+    total: int
+    total_pages: int
 
 
 class BookingSummaryOut(BaseModel):
@@ -48,6 +88,70 @@ class BookingOut(BookingSummaryOut):
     customer_name: str
 
 
+class CreateBookingRequest(BaseModel):
+    customer_ref: str = Field(min_length=3, max_length=20)
+    vehicle_ref: str = Field(min_length=3, max_length=20)
+    starts_at: datetime
+    ends_at: datetime
+    requirements_complete: bool = False
+
+
+class CompleteBookingRequirementsRequest(BaseModel):
+    confirmation: str = Field(min_length=3, max_length=500)
+
+
+class RescheduleBookingRequest(BaseModel):
+    starts_at: datetime
+    ends_at: datetime
+    reason: str = Field(min_length=3, max_length=500)
+
+
+class CustomerOptionOut(BaseModel):
+    public_ref: str
+    display_name: str
+    email: str | None
+
+
+class AvailableVehicleOut(BaseModel):
+    public_ref: str
+    make: str
+    model: str
+    registration_number: str
+    location: str
+    operational_status: str
+
+
+class CancelBookingRequest(BaseModel):
+    reason: str = Field(min_length=3, max_length=500)
+
+
+class CheckoutBookingRequest(BaseModel):
+    start_odometer_km: Annotated[int, Field(ge=0)]
+    fuel_level_percent: Annotated[int, Field(ge=0, le=100)]
+    cleanliness_ok: bool
+    damage_reported: bool = False
+    technical_warning: bool = False
+    notes: str | None = Field(default=None, max_length=2000)
+
+
+class CheckoutBookingResult(BaseModel):
+    booking_ref: str
+    vehicle_ref: str
+    inspection_ref: str
+    booking_status: str
+    resulting_vehicle_status: str
+    activated: bool
+    attention_reasons: list[str]
+
+
+class BookingPageOut(BaseModel):
+    items: list[BookingOut]
+    page: int
+    page_size: int
+    total: int
+    total_pages: int
+
+
 class RegisterReturnRequest(BaseModel):
     end_odometer_km: Annotated[int, Field(ge=0)]
     fuel_level_percent: Annotated[int, Field(ge=0, le=100)]
@@ -71,6 +175,7 @@ class RegisterReturnResult(BaseModel):
     odometer_regression: bool
     quality_issue_ref: str | None
     workflow_event_id: str
+    correlation_id: str
     next_booking_risk: NextBookingRisk | None
 
 
@@ -83,6 +188,8 @@ class ReturnPreviewResult(BaseModel):
     resulting_odometer_km: int
     resulting_vehicle_status: str
     status_reason: str
+    status_reason_code: str
+    status_reason_params: dict[str, str | int] = {}
     would_create_quality_issue: bool
     attention_reasons: list[str]
     next_booking_risk: NextBookingRisk | None
@@ -108,6 +215,19 @@ class MaintenanceOut(BaseModel):
     summary: str
 
 
+class CreateMaintenanceRequest(BaseModel):
+    occurred_at: datetime
+    odometer_km: Annotated[int, Field(ge=0)]
+    category: Literal["periodic_service", "repair", "inspection", "tyres", "other"]
+    summary: str = Field(min_length=3, max_length=2000)
+    next_service_km: Annotated[int | None, Field(default=None, ge=0)]
+    mark_maintenance: bool = True
+
+
+class ReleaseVehicleRequest(BaseModel):
+    reason: str = Field(min_length=3, max_length=500)
+
+
 class DataQualityIssueOut(BaseModel):
     public_ref: str
     rule_type: str
@@ -117,14 +237,38 @@ class DataQualityIssueOut(BaseModel):
     status: str
     evidence: dict[str, Any]
     detected_at: datetime
+    due_at: datetime | None = None
+    assigned_to_ref: str | None = None
+    assigned_to_name: str | None = None
+    overdue: bool = False
     resolved_at: datetime | None = None
 
 
+class DataQualityIssuePageOut(BaseModel):
+    items: list[DataQualityIssueOut]
+    page: int
+    page_size: int
+    total: int
+    total_pages: int
+
+
 class DataQualityIssueDetailOut(DataQualityIssueOut):
     entity_snapshot: dict[str, Any] | None = None
     related_snapshots: list[dict[str, Any]] = Field(default_factory=list)
 
 
+class BulkDataQualityWorkRequest(BaseModel):
+    issue_refs: list[str] = Field(min_length=1, max_length=25)
+    assigned_to_ref: str | None = Field(default=None, min_length=3, max_length=20)
+    clear_assignment: bool = False
+    due_at: datetime | None = None
+    clear_due_at: bool = False
+
+
+class BulkDataQualityWorkResult(BaseModel):
+    updated: list[DataQualityIssueOut]
+
+
 class MergeCustomersRequest(BaseModel):
     survivor_ref: str
     field_overrides: dict[str, str] | None = None
@@ -141,6 +285,53 @@ class ScanResultOut(BaseModel):
     created: dict[str, int]
 
 
+class WorkflowErrorReportIn(BaseModel):
+    workflow_id: str = Field(max_length=120)
+    workflow_name: str = Field(max_length=200)
+    execution_id: str = Field(max_length=120)
+    failed_at: datetime
+    error_category: Literal[
+        "timeout", "authError", "connectionError", "httpError", "validationError", "unknown"
+    ]
+    error_summary: str = Field(max_length=500)
+    trigger_context: str | None = Field(default=None, max_length=200)
+    correlation_id: str | None = None
+    attempt: int = Field(default=1, ge=1, le=1000)
+    retry_action: str | None = Field(default=None, max_length=200)
+
+
+class WorkflowErrorReportResult(BaseModel):
+    status: Literal["registered", "already_registered"]
+    execution_id: str
+    occurred_at: datetime
+
+
+class ProcedureDocumentOut(BaseModel):
+    id: str
+    language: str
+    document_id: str
+    title: str
+    version: str
+    content: str
+    content_hash: str
+
+
+class ProcedureListOut(BaseModel):
+    documents: list[ProcedureDocumentOut]
+
+
+class ProcedureSyncResultIn(BaseModel):
+    execution_id: str = Field(max_length=120)
+    synced: int = Field(ge=0)
+    failed: int = Field(default=0, ge=0)
+
+
+class ProcedureSyncResultResult(BaseModel):
+    status: Literal["registered", "already_registered"]
+    execution_id: str
+    occurred_at: datetime
+
+
 class ProvideFieldsRequest(BaseModel):
     fields: dict[str, str]
 
@@ -157,16 +348,41 @@ class ResolveOverlapRequest(BaseModel):
     note: str | None = Field(default=None, max_length=500)
 
 
+class VehicleStatusFactsOut(BaseModel):
+    active_booking_refs: list[str]
+    overlapping_booking_pairs: list[list[str]]
+    service_threshold_reached: bool
+    odometer_km: int
+    next_service_km: int
+    open_booking_overlap_issue_ref: str | None = None
+
+
+class StatusRecommendationOut(BaseModel):
+    current_status: str
+    recommended_status: str | None
+    recommendation_code: str
+    safe_to_apply: bool
+    manual_review_required: bool
+    facts: VehicleStatusFactsOut
+    blocking_reasons: list[str]
+    recommendation_token: str
+
+
+class ApplyRecommendedStatusRequest(BaseModel):
+    recommendation_token: str
+
+
 class ApplyRecommendedStatusResult(BaseModel):
     issue: DataQualityIssueOut
     applied_status: str
-    reason: str
+    reason_code: str
 
 
 class SearchResultItem(BaseModel):
     type: Literal["vehicle", "booking", "data_quality_issue", "section"]
     label: str
-    detail: str
+    detail_code: str
+    detail_params: dict[str, str] = {}
     link: str
 
 
@@ -175,21 +391,106 @@ class SearchResponse(BaseModel):
     results: list[SearchResultItem]
 
 
+class OidcStatusOut(BaseModel):
+    enabled: bool
+    provider_name: str | None = None
+
+
+class PrivacyRetentionOut(BaseModel):
+    minimum_booking_retention_days: int
+    audit_retention_days: int
+    customers_total: int
+    customers_anonymized: int
+    customers_eligible: int
+
+
+class CustomerAnonymizeRequest(BaseModel):
+    confirmation: str = Field(min_length=1, max_length=20)
+    reason: str = Field(min_length=8, max_length=500)
+
+
+class CustomerAnonymizeResult(BaseModel):
+    public_ref: str
+    anonymized_at: datetime
+    status: Literal["anonymized", "already_anonymized"]
+
+
+class N8nWorkflowEvidence(BaseModel):
+    name: str
+    built: bool
+    last_seen_at: datetime | None
+    state: Literal["no_evidence", "healthy", "stale", "failed"] = "no_evidence"
+    last_status: Literal["succeeded", "failed"] | None = None
+    last_execution_id: str | None = None
+
+
+class ReturnCallbackIn(BaseModel):
+    """Body of the n8n return follow-up callback.
+
+    n8n forwards its whole item (``JSON.stringify($json)``), so unknown keys are ignored;
+    only the fields we persist are validated and bounded.
+    """
+
+    model_config = ConfigDict(extra="ignore")
+
+    correlation_id: uuid.UUID | None = None
+    follow_up: str | None = Field(default=None, max_length=200)
+    summary: str | None = Field(default=None, max_length=2000)
+
+
+class N8nHeartbeatIn(BaseModel):
+    workflow_id: str = Field(min_length=1, max_length=120)
+    workflow_name: str = Field(min_length=1, max_length=200)
+    execution_id: str = Field(min_length=1, max_length=120)
+    status: Literal["succeeded", "failed"]
+
+
+class N8nHeartbeatResult(BaseModel):
+    status: Literal["registered", "already_registered"]
+    execution_id: str
+    occurred_at: datetime
+
+
+class N8nErrorHandlerStatus(BaseModel):
+    total_failures_registered: int
+    latest_failure_at: datetime | None
+    latest_failure_workflow: str | None
+
+
 class N8nIntegrationStatus(BaseModel):
     configured: bool
     dispatch_enabled: bool
     state: Literal["disabled", "unavailable", "degraded", "operational", "no_evidence"]
     pending: int
-    delivering: int
+    #: Every failed delivery, staged and real together -- the number a viewer sees in
+    #: the run list.
     failed: int
+    #: Failures that were not planted by the demo seed. This is the only failure count
+    #: that may influence `state`.
+    unexpected_failed: int = 0
+    #: Prepared demo failures (see `app.models.outbox.DEMO_SCENARIO_ERROR_CODE`).
+    #: Present so the UI can label them instead of implying the automation is broken.
+    demo_scenario_failed: int = 0
+    delivering: int
     succeeded: int
     latest_success_at: datetime | None
+    #: Most recent *real* failure; a staged one never sets this.
     latest_failure_at: datetime | None
+    latest_demo_scenario_at: datetime | None = None
+    expected_workflow_count: int
+    known_workflow_count: int
+    workflows: list[N8nWorkflowEvidence]
+    error_handler: N8nErrorHandlerStatus
 
 
 class McpHubIntegrationStatus(BaseModel):
     registration_enabled: bool
-    state: Literal["not_configured", "configured"]
+    state: Literal["not_configured", "no_evidence", "operational"]
+    total_calls: int
+    last_tool: str | None = None
+    last_client: str | None = None
+    last_called_at: datetime | None = None
+    hub_reachable: bool | None = None
 
 
 class IntegrationStatusOut(BaseModel):
@@ -245,11 +546,16 @@ class DashboardMetrics(BaseModel):
     pending_or_failed_workflows: int
 
 
+class EvidenceSignalOut(BaseModel):
+    code: str
+    params: dict[str, Any] = Field(default_factory=dict)
+
+
 class AttentionItem(BaseModel):
     kind: Literal["quality_issue", "vehicle"]
     severity: str
     rule_type: str
-    detail: str
+    evidence_signals: list[EvidenceSignalOut] = Field(default_factory=list)
     link_type: Literal["vehicle", "booking", "customer"]
     link_ref: str
     issue_ref: str | None = None
@@ -269,6 +575,11 @@ class AutomationRunOut(BaseModel):
     status: str
     attempts: int
     last_error: str | None
+    last_error_code: str | None
+    #: True for the deliberately seeded demo failure. The UI uses this to label the run
+    #: as a prepared scenario and to offer the demo retry, instead of presenting it as
+    #: an unexplained production error.
+    is_demo_scenario: bool = False
     occurred_at: datetime
 
 
@@ -308,6 +619,7 @@ class McpVehicleDetailOut(BaseModel):
 class McpKnowledgeSearchRequest(BaseModel):
     question: str = Field(min_length=3, max_length=1000)
     max_sources: int = Field(default=4, ge=1, le=8)
+    locale: Literal["nl-BE", "en-GB", "fr-BE"] = "en-GB"
 
 
 class AuditEventOut(BaseModel):
@@ -324,3 +636,11 @@ class AuditEventOut(BaseModel):
     before: dict[str, Any] | None = None
     after: dict[str, Any] | None = None
     metadata: dict[str, Any] | None = None
+
+
+class AuditEventPageOut(BaseModel):
+    items: list[AuditEventOut]
+    page: int
+    page_size: int
+    total: int
+    total_pages: int

backend/app/seed_loader.py

@@ -4,6 +4,7 @@ import csv
 import uuid
 from dataclasses import dataclass
 from datetime import UTC, date, datetime, timedelta
+from difflib import SequenceMatcher
 from pathlib import Path
 
 from sqlalchemy import delete, insert, update
@@ -17,7 +18,7 @@ from app.models.data_quality import DataQualityIssue
 from app.models.idempotency import IdempotencyRecord
 from app.models.inspection import Inspection
 from app.models.maintenance import MaintenanceRecord
-from app.models.outbox import OutboxEvent
+from app.models.outbox import DEMO_SCENARIO_ERROR_CODE, OutboxEvent
 from app.models.user import User
 from app.models.vehicle import Vehicle
 from app.services.audit import record_audit_event
@@ -79,9 +80,20 @@ def _read_csv(name: str) -> list[dict[str, str]]:
         return list(csv.DictReader(handle))
 
 
-def clear_all(db: Session) -> None:
+_PERSISTENT_TELEMETRY_ACTIONS = (
+    "mcp_tool_request",
+    "n8n_return_followup_recorded",
+    "n8n_workflow_failure_registered",
+    "n8n_procedures_synced",
+    "n8n_workflow_heartbeat",
+    "knowledge_question_asked",
+)
+
+
+def clear_all(db: Session, *, preserve_integration_telemetry: bool = False) -> None:
+    # RevokedSession is intentionally NOT cleared: it has no FK to users and wiping it
+    # would silently re-validate cookies that were logged out before the reset.
     for model in (
-        AuditEvent,
         OutboxEvent,
         IdempotencyRecord,
         DataQualityIssue,
@@ -93,6 +105,12 @@ def clear_all(db: Session) -> None:
         User,
     ):
         db.execute(delete(model))
+    if preserve_integration_telemetry:
+        db.execute(
+            delete(AuditEvent).where(AuditEvent.action.not_in(_PERSISTENT_TELEMETRY_ACTIONS))
+        )
+    else:
+        db.execute(delete(AuditEvent))
 
 
 def load_seed(db: Session) -> SeedResult:
@@ -100,19 +118,17 @@ def load_seed(db: Session) -> SeedResult:
     today = datetime.now(UTC).date()
     shift = _seed_anchor_shift(today)
 
-    user_rows = [
+    user_rows = [{"id": uuid.uuid4(), **user, "active": True} for user in DEMO_USERS]
-        {"id": uuid.uuid4(), **user, "active": True} for user in DEMO_USERS
-    ]
     db.execute(insert(User), user_rows)
     counts["users"] = len(user_rows)
 
     customer_id_by_ref: dict[str, uuid.UUID] = {}
     customer_rows = []
+    customer_row_by_ref: dict[str, dict] = {}
     for row in _read_csv("customers.csv"):
         cid = uuid.uuid4()
         customer_id_by_ref[row["public_ref"]] = cid
-        customer_rows.append(
+        customer_row = {
-            {
             "id": cid,
             "public_ref": row["public_ref"],
             "first_name": row["first_name"],
@@ -122,7 +138,8 @@ def load_seed(db: Session) -> SeedResult:
             "postal_code": row["postal_code"] or None,
             "city": row["city"] or None,
         }
-        )
+        customer_rows.append(customer_row)
+        customer_row_by_ref[row["public_ref"]] = customer_row
     db.execute(insert(Customer), customer_rows)
     counts["customers"] = len(customer_rows)
     # Second pass for merged_into (self-referencing FK) since target must exist first.
@@ -137,11 +154,11 @@ def load_seed(db: Session) -> SeedResult:
 
     vehicle_id_by_ref: dict[str, uuid.UUID] = {}
     vehicle_rows = []
+    vehicle_row_by_ref: dict[str, dict] = {}
     for row in _read_csv("vehicles.csv"):
         vid = uuid.uuid4()
         vehicle_id_by_ref[row["public_ref"]] = vid
-        vehicle_rows.append(
+        vehicle_row = {
-            {
             "id": vid,
             "public_ref": row["public_ref"],
             "make": row["make"],
@@ -155,17 +172,18 @@ def load_seed(db: Session) -> SeedResult:
             "active": _parse_bool(row["active"]),
             "version": 1,
         }
-        )
+        vehicle_rows.append(vehicle_row)
+        vehicle_row_by_ref[row["public_ref"]] = vehicle_row
     db.execute(insert(Vehicle), vehicle_rows)
     counts["vehicles"] = len(vehicle_rows)
 
     booking_id_by_ref: dict[str, uuid.UUID] = {}
     booking_rows = []
+    booking_row_by_ref: dict[str, dict] = {}
     for row in _read_csv("bookings.csv"):
         bid = uuid.uuid4()
         booking_id_by_ref[row["public_ref"]] = bid
-        booking_rows.append(
+        booking_row = {
-            {
             "id": bid,
             "public_ref": row["public_ref"],
             "customer_id": customer_id_by_ref[row["customer_ref"]],
@@ -177,7 +195,8 @@ def load_seed(db: Session) -> SeedResult:
             "end_odometer_km": _parse_optional_int(row["end_odometer_km"]),
             "requirements_complete": _parse_bool(row["requirements_complete"]),
         }
-        )
+        booking_rows.append(booking_row)
+        booking_row_by_ref[row["public_ref"]] = booking_row
     db.execute(insert(Booking), booking_rows)
     counts["bookings"] = len(booking_rows)
 
@@ -223,11 +242,123 @@ def load_seed(db: Session) -> SeedResult:
             return "customer", customer_id_by_ref[entity_ref]
         return "vehicle", vehicle_id_by_ref[entity_ref]
 
+    def _vehicle_conflict_facts(vehicle_ref: str, *, service_threshold_reached: bool) -> dict:
+        # Mirrors app.services.vehicle_status.VehicleStatusFacts.as_dict() for the
+        # handful of seed-only rows below -- none of them carry an active rental or a
+        # real booking conflict (verified against the fixed seed dataset), only a
+        # genuinely-crossed service threshold or none at all, so those two fields are
+        # the only ones that vary per vehicle.
+        vehicle = vehicle_row_by_ref[vehicle_ref]
+        return {
+            "active_booking_refs": [],
+            "overlapping_booking_pairs": [],
+            "service_threshold_reached": service_threshold_reached,
+            "odometer_km": vehicle["odometer_km"],
+            "next_service_km": vehicle["next_service_km"],
+            "open_booking_overlap_issue_ref": None,
+        }
+
+    def _odometer_regression_signal(later_ref: str, earlier_ref: str) -> list[dict]:
+        later = booking_row_by_ref[later_ref]
+        earlier = booking_row_by_ref[earlier_ref]
+        return [
+            {
+                "code": "odometer.regression",
+                "params": {
+                    "later_ref": later_ref,
+                    "later_km": later["end_odometer_km"],
+                    "earlier_ref": earlier_ref,
+                    "earlier_km": earlier["end_odometer_km"],
+                },
+            }
+        ]
+
+    def _missing_field_signal(field: str) -> list[dict]:
+        return [{"code": "missing_field", "params": {"field": field}}]
+
+    # Every seed-only row below (i.e. not one of the four named DQ-DEMO-* scenarios)
+    # used to carry no structured signal at all -- just the placeholder summary
+    # "Synthetic deterministic seed issue". Each now cites a real fact about its actual
+    # entity (a genuinely-crossed service threshold, a genuinely-blank field, or a real
+    # pair of booking odometer readings engineered into seed/bookings.csv), using the
+    # exact same signal vocabulary the live scan (app.services.data_quality) already
+    # renders through -- see docs/fleet-ops-correction/current-gap-audit.md §6.
+    _SEED_SIGNALS_BY_REF: dict[str, list[dict]] = {
+        "DQ-0005": [
+            {
+                "code": "vehicle.service_threshold_reached",
+                "params": _vehicle_conflict_facts("MO-036", service_threshold_reached=True),
+            }
+        ],
+        "DQ-0006": _missing_field_signal("location"),
+        "DQ-0007": _odometer_regression_signal("BK-H-0007", "BK-H-0057"),
+        "DQ-0008": [
+            {
+                "code": "vehicle.rental_ended",
+                "params": _vehicle_conflict_facts("MO-007", service_threshold_reached=False),
+            }
+        ],
+        "DQ-0009": _missing_field_signal("location"),
+        "DQ-0010": _odometer_regression_signal("BK-H-0010", "BK-H-0060"),
+        "DQ-0011": [
+            {
+                "code": "vehicle.service_threshold_reached",
+                "params": _vehicle_conflict_facts("MO-028", service_threshold_reached=True),
+            }
+        ],
+        "DQ-0012": _missing_field_signal("registration_number"),
+        "DQ-0013": _missing_field_signal("location"),
+        "DQ-0014": _missing_field_signal("registration_number"),
+        "DQ-0015": _missing_field_signal("location"),
+        "DQ-0016": _missing_field_signal("location"),
+        "DQ-0017": _missing_field_signal("location"),
+        "DQ-0018": _missing_field_signal("registration_number"),
+        "DQ-0019": _missing_field_signal("location"),
+        "DQ-0020": _missing_field_signal("location"),
+        "DQ-0021": _missing_field_signal("location"),
+    }
+
+    def _seed_signals(public_ref: str, entity_ref: str, related_refs: list[str]) -> list[dict]:
+        # The four named DQ-DEMO-* rows anchor the guided demo's scripted scenarios, so
+        # they carry real, accurate structured signals (not just a legacy English
+        # sentence) -- the frontend renders these as the primary, localized evidence;
+        # see docs/fleet-ops-correction/current-gap-audit.md §6.
+        if public_ref == "DQ-DEMO-DUPLICATE":
+            a = customer_row_by_ref[entity_ref]
+            b = customer_row_by_ref[related_refs[0]]
+            name_a = f"{a['first_name']} {a['last_name']}".strip().lower()
+            name_b = f"{b['first_name']} {b['last_name']}".strip().lower()
+            ratio = SequenceMatcher(None, name_a, name_b).ratio()
+            return [
+                {"code": "duplicate.exact_email"},
+                {"code": "duplicate.exact_phone"},
+                {"code": "duplicate.same_postal_code"},
+                {"code": "duplicate.similar_name", "params": {"score": round(ratio, 2)}},
+            ]
+        if public_ref == "DQ-DEMO-OVERLAP":
+            return [{"code": "overlap.reserved_bookings", "params": {"refs": related_refs}}]
+        if public_ref == "DQ-DEMO-STATUS":
+            return [{"code": "vehicle.booking_conflict"}]
+        if public_ref == "DQ-DEMO-ATTENTION":
+            return [
+                {
+                    "code": "attention.upcoming_booking_missing_inspection",
+                    "params": {"booking_ref": related_refs[0] if related_refs else ""},
+                }
+            ]
+        return _SEED_SIGNALS_BY_REF.get(public_ref, [])
+
     dq_rows = []
     now = datetime.now(UTC)
     for row in _read_csv("data_quality_issues.csv"):
         entity_type, entity_id = resolve_entity(row["entity_ref"])
         related_ref = row.get("related_ref") or ""
+        related_refs = related_ref.split("|") if related_ref else []
+        severity_due_delta = {
+            "high": timedelta(hours=4),
+            "medium": timedelta(days=1),
+            "low": timedelta(days=3),
+        }.get(row["severity"], timedelta(days=1))
         dq_rows.append(
             {
                 "id": uuid.uuid4(),
@@ -240,10 +371,12 @@ def load_seed(db: Session) -> SeedResult:
                 "evidence_json": {
                     "summary": row["evidence"],
                     "entity_ref": row["entity_ref"],
-                    "related_refs": related_ref.split("|") if related_ref else [],
+                    "related_refs": related_refs,
+                    "signals": _seed_signals(row["public_ref"], row["entity_ref"], related_refs),
                 },
                 "proposed_action_json": {},
                 "detected_at": now,
+                "due_at": now + severity_due_delta if row["status"] == "open" else None,
                 "resolved_at": now if row["status"] == "resolved" else None,
                 "resolved_by": "USR-OPS" if row["status"] == "resolved" else None,
             }
@@ -287,6 +420,12 @@ def load_seed(db: Session) -> SeedResult:
                 "attempts": int(row["attempts"]),
                 "next_attempt_at": None,
                 "last_error": row["last_error"] or None,
+                # The seed dataset's one synthetic failure (BK-H-0020) models a
+                # connection-timeout-style delivery failure -- see workflow_runs.csv.
+                # It is coded as a *prepared demo scenario*, not as a real
+                # connectionError, so integration health never degrades because of a
+                # prop and a viewer is told plainly that this failure is staged.
+                "last_error_code": DEMO_SCENARIO_ERROR_CODE if row["last_error"] else None,
                 "external_run_id": None,
             }
         )
@@ -310,10 +449,10 @@ def load_seed(db: Session) -> SeedResult:
     return SeedResult(counts=counts, anchor_date=today, seeded_at=seeded_at)
 
 
-def reset_and_seed(db: Session) -> SeedResult:
+def reset_and_seed(db: Session, *, preserve_integration_telemetry: bool = False) -> SeedResult:
     from app.services.data_quality import run_scan
 
-    clear_all(db)
+    clear_all(db, preserve_integration_telemetry=preserve_integration_telemetry)
     result = load_seed(db)
     db.commit()
     scan = run_scan(db)

backend/app/services/data_quality.py

@@ -2,10 +2,9 @@ from __future__ import annotations
 
 import uuid
 from dataclasses import dataclass, field
-from datetime import UTC, datetime
+from datetime import UTC, datetime, timedelta
-from difflib import SequenceMatcher
 
-from sqlalchemy import select, update
+from sqlalchemy import func, select, update
 from sqlalchemy.orm import Session
 
 from app.core.errors import AppError
@@ -15,10 +14,27 @@ from app.models.data_quality import DataQualityIssue
 from app.models.vehicle import Vehicle
 from app.schemas import CurrentUser, ResolveOdometerRegressionRequest
 from app.services.audit import record_audit_event
+from app.services.data_quality_duplicate_scan import scan_duplicate_customers
+from app.services.vehicle_status import (
+    RECOMMENDATION_CODE_NO_CONFLICT,
+    VehicleStatusRecommendation,
+    compute_recommendation_token,
+    evaluate_vehicle_status,
+    gather_vehicle_status_facts,
+)
 
 REQUIRED_CUSTOMER_FIELDS = ("first_name", "last_name")
 REQUIRED_VEHICLE_FIELDS = ("registration_number", "make", "model", "location")
-DUPLICATE_THRESHOLD = 70
+DATA_QUALITY_SCAN_LOCK_ID = 6_138_493_717_091_029_491
+
+
+def issue_due_at(detected_at: datetime, severity: str) -> datetime:
+    """Return the local operational SLA deadline for a newly detected issue."""
+    return detected_at + {
+        "high": timedelta(hours=4),
+        "medium": timedelta(days=1),
+        "low": timedelta(days=3),
+    }.get(severity, timedelta(days=1))
 
 
 @dataclass
@@ -29,10 +45,6 @@ class ScanResult:
         self.created[rule_type] = self.created.get(rule_type, 0) + 1
 
 
-def _normalize(value: str | None) -> str:
-    return (value or "").strip().lower()
-
-
 def _has_open_issue(db: Session, rule_type: str, entity_type: str, entity_id: uuid.UUID) -> bool:
     return (
         db.scalar(
@@ -47,15 +59,9 @@ def _has_open_issue(db: Session, rule_type: str, entity_type: str, entity_id: uu
     )
 
 
-def _next_public_ref(db: Session, prefix: str) -> str:
+def _new_scan_ref(prefix: str) -> str:
-    existing = db.execute(select(DataQualityIssue.public_ref)).scalars().all()
+    """Generate a stable human-readable prefix with a concurrent-safe suffix."""
-    numbers = [
+    return f"{prefix}-{uuid.uuid4().hex[:10].upper()}"
-        int(ref.rsplit("-", 1)[-1])
-        for ref in existing
-        if ref.startswith(f"{prefix}-") and ref.rsplit("-", 1)[-1].isdigit()
-    ]
-    next_number = (max(numbers) + 1) if numbers else 1
-    return f"{prefix}-{next_number:04d}"
 
 
 def _open_issue(
@@ -69,6 +75,7 @@ def _open_issue(
     summary: str,
     entity_ref: str,
     related_refs: list[str],
+    signals: list[dict] | None = None,
 ) -> None:
     if _has_open_issue(db, rule_type, entity_type, entity_id):
         return
@@ -87,17 +94,21 @@ def _open_issue(
         )
         .order_by(DataQualityIssue.detected_at.desc())
     )
+    # `summary` is kept as a technical-fallback string (shown only under "Technical
+    # details"); `signals` is the stable, localizable structure the frontend renders as
+    # the primary evidence -- see docs/fleet-ops-correction/current-gap-audit.md §2/§6.
     evidence: dict = {
         "summary": summary,
         "entity_ref": entity_ref,
         "related_refs": related_refs,
+        "signals": signals or [],
     }
     if previous is not None:
         evidence["reopened_from"] = previous.public_ref
         evidence["previous_decision"] = previous.status
 
     issue = DataQualityIssue(
-        public_ref=_next_public_ref(db, "DQ-SCAN"),
+        public_ref=_new_scan_ref("DQ-SCAN"),
         rule_type=rule_type,
         entity_type=entity_type,
         entity_id=entity_id,
@@ -106,55 +117,21 @@ def _open_issue(
         evidence_json=evidence,
         proposed_action_json={},
         detected_at=now,
+        due_at=issue_due_at(now, severity),
     )
     db.add(issue)
     db.flush()
     scan.bump(rule_type)
 
 
-def _scan_duplicate_customers(db: Session, scan: ScanResult) -> None:
-    customers = list(
-        db.scalars(select(Customer).where(Customer.merged_into_customer_id.is_(None))).all()
-    )
-    customers.sort(key=lambda c: c.public_ref)
-
-    for i, a in enumerate(customers):
-        for b in customers[i + 1 :]:
-            score = 0
-            signals = []
-            if _normalize(a.email) and _normalize(a.email) == _normalize(b.email):
-                score += 60
-                signals.append("exact email")
-            if _normalize(a.phone) and _normalize(a.phone) == _normalize(b.phone):
-                score += 50
-                signals.append("exact phone")
-            if _normalize(a.postal_code) and _normalize(a.postal_code) == _normalize(b.postal_code):
-                score += 10
-                signals.append("exact postal code")
-            name_a = f"{_normalize(a.first_name)} {_normalize(a.last_name)}"
-            name_b = f"{_normalize(b.first_name)} {_normalize(b.last_name)}"
-            ratio = SequenceMatcher(None, name_a, name_b).ratio()
-            if ratio >= 0.5:
-                score += round(ratio * 30)
-                signals.append("similar name")
-
-            if score >= DUPLICATE_THRESHOLD:
-                _open_issue(
-                    db,
-                    scan,
-                    rule_type="possible_duplicate_customer",
-                    entity_type="customer",
-                    entity_id=a.id,
-                    severity="high",
-                    summary="; ".join(signals) + f" (score {score})",
-                    entity_ref=a.public_ref,
-                    related_refs=[b.public_ref],
-                )
-
-
 def _scan_missing_required_fields(db: Session, scan: ScanResult) -> None:
+    # Anonymised customers have had their contact data removed on purpose; flagging
+    # them as "missing required field" would only be resolvable by re-entering PII.
     for customer in db.scalars(
-        select(Customer).where(Customer.merged_into_customer_id.is_(None))
+        select(Customer).where(
+            Customer.merged_into_customer_id.is_(None),
+            Customer.anonymized_at.is_(None),
+        )
     ).all():
         missing = [f for f in REQUIRED_CUSTOMER_FIELDS if not getattr(customer, f)]
         if not customer.email and not customer.phone:
@@ -170,6 +147,7 @@ def _scan_missing_required_fields(db: Session, scan: ScanResult) -> None:
                 summary=f"Missing: {', '.join(missing)}",
                 entity_ref=customer.public_ref,
                 related_refs=[],
+                signals=[{"code": "missing_field", "params": {"field": f}} for f in missing],
             )
 
     for vehicle in db.scalars(select(Vehicle).where(Vehicle.active.is_(True))).all():
@@ -185,6 +163,7 @@ def _scan_missing_required_fields(db: Session, scan: ScanResult) -> None:
                 summary=f"Missing: {', '.join(missing)}",
                 entity_ref=vehicle.public_ref,
                 related_refs=[],
+                signals=[{"code": "missing_field", "params": {"field": f}} for f in missing],
             )
 
 
@@ -213,39 +192,32 @@ def _scan_booking_overlaps(db: Session, scan: ScanResult) -> None:
                         summary=f"Overlapping bookings {first.public_ref} and {second.public_ref}",
                         entity_ref=vehicle.public_ref,
                         related_refs=[first.public_ref, second.public_ref],
+                        signals=[
+                            {
+                                "code": "overlap.reserved_bookings",
+                                "params": {"refs": [first.public_ref, second.public_ref]},
+                            }
+                        ],
                     )
 
 
 def _scan_vehicle_status_conflicts(db: Session, scan: ScanResult) -> None:
+    # Uses the same shared evaluator as the preview/apply flow (app.services.vehicle_status)
+    # so detection and resolution can never structurally disagree -- see
+    # docs/fleet-ops-correction/vehicle-status-decision-table.md.
     vehicles = db.scalars(select(Vehicle)).all()
-    active_by_vehicle: dict[uuid.UUID, list[Booking]] = {}
-    for booking in db.scalars(select(Booking).where(Booking.status == "active")).all():
-        active_by_vehicle.setdefault(booking.vehicle_id, []).append(booking)
-
-    open_high_by_vehicle = {
-        row[0]
-        for row in db.execute(
-            select(DataQualityIssue.entity_id).where(
-                DataQualityIssue.entity_type == "vehicle",
-                DataQualityIssue.status == "open",
-                DataQualityIssue.severity == "high",
-            )
-        ).all()
-    }
-
     for vehicle in vehicles:
-        has_active_booking = vehicle.id in active_by_vehicle
+        facts = gather_vehicle_status_facts(db, vehicle)
-        reason = None
+        recommendation = evaluate_vehicle_status(vehicle, facts)
-        if vehicle.operational_status == "available" and has_active_booking:
+        if recommendation.recommendation_code == RECOMMENDATION_CODE_NO_CONFLICT:
-            reason = "marked available while an active booking exists"
+            continue
-        elif vehicle.operational_status == "rented" and not has_active_booking:
-            reason = "marked rented without an active booking"
-        elif vehicle.operational_status == "available" and vehicle.id in open_high_by_vehicle:
-            reason = "marked available while a high-severity quality issue is open"
-        elif vehicle.operational_status == "maintenance" and has_active_booking:
-            reason = "marked maintenance while an active booking exists"
 
-        if reason:
+        signals = [{"code": recommendation.recommendation_code, "params": facts.as_dict()}]
+        summary = (
+            f"Recommended status: {recommendation.recommended_status}"
+            if recommendation.recommended_status
+            else "Manual review required: active rental conflicts with a blocking condition"
+        )
         _open_issue(
             db,
             scan,
@@ -253,9 +225,13 @@ def _scan_vehicle_status_conflicts(db: Session, scan: ScanResult) -> None:
             entity_type="vehicle",
             entity_id=vehicle.id,
             severity="high",
-                summary=f"Vehicle {reason}",
+            summary=summary,
             entity_ref=vehicle.public_ref,
-                related_refs=[],
+            related_refs=[
+                *facts.active_booking_refs,
+                *(ref for pair in facts.overlapping_booking_pairs for ref in pair),
+            ],
+            signals=signals,
         )
 
 
@@ -267,9 +243,7 @@ def _scan_odometer_regressions(db: Session, scan: ScanResult) -> None:
     vehicles = {v.id: v for v in db.scalars(select(Vehicle)).all()}
     bookings_by_vehicle: dict[uuid.UUID, list[Booking]] = {}
     for booking in db.scalars(
-        select(Booking).where(
+        select(Booking).where(Booking.status == "returned", Booking.end_odometer_km.is_not(None))
-            Booking.status == "returned", Booking.end_odometer_km.is_not(None)
-        )
     ).all():
         bookings_by_vehicle.setdefault(booking.vehicle_id, []).append(booking)
 
@@ -295,6 +269,17 @@ def _scan_odometer_regressions(db: Session, scan: ScanResult) -> None:
                     ),
                     entity_ref=vehicle.public_ref,
                     related_refs=[earlier.public_ref, later.public_ref],
+                    signals=[
+                        {
+                            "code": "odometer.regression",
+                            "params": {
+                                "later_ref": later.public_ref,
+                                "later_km": later.end_odometer_km,
+                                "earlier_ref": earlier.public_ref,
+                                "earlier_km": earlier.end_odometer_km,
+                            },
+                        }
+                    ],
                 )
                 break
 
@@ -302,8 +287,11 @@ def _scan_odometer_regressions(db: Session, scan: ScanResult) -> None:
 def run_scan(
     db: Session, *, actor_label: str | None = None, actor_type: str = "user"
 ) -> ScanResult:
+    # The check-then-insert work below spans several rules. Serialise whole scans at the
+    # database boundary so API and n8n triggers cannot both observe an empty condition.
+    db.scalar(select(func.pg_advisory_xact_lock(DATA_QUALITY_SCAN_LOCK_ID)))
     scan = ScanResult()
-    _scan_duplicate_customers(db, scan)
+    scan_duplicate_customers(db, scan, _open_issue)
     _scan_missing_required_fields(db, scan)
     _scan_odometer_regressions(db, scan)
     _scan_booking_overlaps(db, scan)
@@ -321,10 +309,13 @@ def run_scan(
     return scan
 
 
-def _load_open_issue(db: Session, public_ref: str) -> DataQualityIssue:
+def _load_open_issue(
-    issue = db.scalar(
+    db: Session, public_ref: str, *, lock: bool = True
-        select(DataQualityIssue).where(DataQualityIssue.public_ref == public_ref)
+) -> DataQualityIssue:
-    )
+    statement = select(DataQualityIssue).where(DataQualityIssue.public_ref == public_ref)
+    if lock:
+        statement = statement.with_for_update()
+    issue = db.scalar(statement)
     if issue is None:
         raise AppError("ISSUE_NOT_FOUND", "Data quality issue not found.", status_code=404)
     if issue.status != "open":
@@ -475,6 +466,30 @@ def resolve_odometer_regression(
             "This issue is not an odometer_regression issue.",
             status_code=409,
         )
+    # Lock order is booking -> vehicle everywhere (checkout, return, reschedule); taking
+    # the vehicle lock first here would be a deadlock waiting to happen under concurrency.
+    booking: Booking | None = None
+    if body.decision != "retain_canonical":
+        related_refs = issue.evidence_json.get("related_refs", [])
+        if body.booking_ref not in related_refs:
+            raise AppError(
+                "INVALID_BOOKING_REFERENCE",
+                "booking_ref must be one of this issue's related bookings.",
+                status_code=422,
+            )
+        if body.corrected_odometer_km is None:
+            raise AppError(
+                "CORRECTED_VALUE_REQUIRED",
+                "corrected_odometer_km is required when correcting a reading.",
+                status_code=422,
+            )
+        booking = db.scalar(
+            select(Booking).where(Booking.public_ref == body.booking_ref).with_for_update()
+        )
+        if booking is None:
+            raise AppError(
+                "BOOKING_NOT_FOUND", "The booking to correct was not found.", status_code=404
+            )
     vehicle = db.scalar(select(Vehicle).where(Vehicle.id == issue.entity_id).with_for_update())
     if vehicle is None:
         raise AppError(
@@ -495,19 +510,7 @@ def resolve_odometer_regression(
             metadata={"issue_ref": issue.public_ref, "canonical_odometer_km": vehicle.odometer_km},
         )
     else:
-        related_refs = issue.evidence_json.get("related_refs", [])
+        assert booking is not None and body.corrected_odometer_km is not None
-        if body.booking_ref not in related_refs:
-            raise AppError(
-                "INVALID_BOOKING_REFERENCE",
-                "booking_ref must be one of this issue's related bookings.",
-                status_code=422,
-            )
-        if body.corrected_odometer_km is None:
-            raise AppError(
-                "CORRECTED_VALUE_REQUIRED",
-                "corrected_odometer_km is required when correcting a reading.",
-                status_code=422,
-            )
         # Never silently lower the canonical odometer: a correction must be at or above
         # the current canonical value, otherwise it would just create a new regression.
         if body.corrected_odometer_km < vehicle.odometer_km:
@@ -519,13 +522,6 @@ def resolve_odometer_regression(
                 ),
                 status_code=422,
             )
-        booking = db.scalar(
-            select(Booking).where(Booking.public_ref == body.booking_ref).with_for_update()
-        )
-        if booking is None:
-            raise AppError(
-                "BOOKING_NOT_FOUND", "The booking to correct was not found.", status_code=404
-            )
 
         before = {
             "booking_end_odometer_km": booking.end_odometer_km,
@@ -650,76 +646,85 @@ def resolve_booking_overlap(
     return issue
 
 
-def _recommend_vehicle_status(
+def _load_vehicle_status_conflict_issue(
-    operational_status: str, has_active_booking: bool, has_open_high_issue: bool
+    db: Session, public_ref: str, *, lock: bool = True
-) -> tuple[str, str] | None:
+) -> DataQualityIssue:
-    """The single authoritative recommendation function for vehicle_status_conflict,
+    issue = _load_open_issue(db, public_ref, lock=lock)
-    mirroring the exact conditions `_scan_vehicle_status_conflicts` flags."""
-    if operational_status == "available" and has_active_booking:
-        return "rented", "An active booking exists; the vehicle should be marked rented."
-    if operational_status == "rented" and not has_active_booking:
-        return "available", "No active booking exists; the vehicle should be marked available."
-    if operational_status == "available" and has_open_high_issue:
-        return "blocked", "A high-severity quality issue is open; the vehicle should be blocked."
-    if operational_status == "maintenance" and has_active_booking:
-        return (
-            "rented",
-            "An active booking exists despite the maintenance status; it should be rented.",
-        )
-    return None
-
-
-def apply_recommended_status(
-    db: Session, public_ref: str, actor: CurrentUser
-) -> tuple[DataQualityIssue, str, str]:
-    issue = _load_open_issue(db, public_ref)
     if issue.rule_type != "vehicle_status_conflict":
         raise AppError(
             "NOT_A_STATUS_CONFLICT_ISSUE",
             "This issue is not a vehicle_status_conflict issue.",
             status_code=409,
         )
+    return issue
+
+
+def preview_vehicle_status_recommendation(
+    db: Session, public_ref: str
+) -> tuple[DataQualityIssue, Vehicle, VehicleStatusRecommendation, str]:
+    """Non-mutating: computes and returns the recommendation only. Never resolves the
+    issue, never writes an audit event, never queues automation -- safe to call as often
+    as the UI needs (e.g. every time the panel is opened) with zero side effects."""
+    issue = _load_vehicle_status_conflict_issue(db, public_ref, lock=False)
+    vehicle = db.scalar(select(Vehicle).where(Vehicle.id == issue.entity_id))
+    if vehicle is None:
+        raise AppError(
+            "VEHICLE_NOT_FOUND", "The vehicle for this issue was not found.", status_code=404
+        )
+    facts = gather_vehicle_status_facts(db, vehicle, exclude_issue_id=issue.id)
+    recommendation = evaluate_vehicle_status(vehicle, facts)
+    token = compute_recommendation_token(vehicle, facts)
+    return issue, vehicle, recommendation, token
+
+
+def apply_recommended_status(
+    db: Session, public_ref: str, actor: CurrentUser, expected_token: str
+) -> tuple[DataQualityIssue, str, str]:
+    issue = _load_vehicle_status_conflict_issue(db, public_ref)
+    # Lock the vehicle row for the remainder of this transaction so a concurrent apply
+    # (or return/checkout) can't race between our fact-gathering and the write below.
     vehicle = db.scalar(select(Vehicle).where(Vehicle.id == issue.entity_id).with_for_update())
     if vehicle is None:
         raise AppError(
             "VEHICLE_NOT_FOUND", "The vehicle for this issue was not found.", status_code=404
         )
 
-    has_active_booking = (
+    facts = gather_vehicle_status_facts(db, vehicle, exclude_issue_id=issue.id)
-        db.scalar(
+    recommendation = evaluate_vehicle_status(vehicle, facts)
-            select(Booking.id).where(Booking.vehicle_id == vehicle.id, Booking.status == "active")
+    current_token = compute_recommendation_token(vehicle, facts)
+
+    if current_token != expected_token:
+        raise AppError(
+            "RECOMMENDATION_STALE",
+            "The underlying facts changed since this recommendation was shown; "
+            "review the recommendation again before applying it.",
+            status_code=409,
         )
-        is not None
+    if recommendation.manual_review_required or not recommendation.safe_to_apply:
+        raise AppError(
+            "MANUAL_REVIEW_REQUIRED",
+            "This vehicle's state requires manual review; no automatic status change is safe.",
+            status_code=409,
         )
-    has_open_high_issue = (
+    if recommendation.recommended_status is None:
-        db.scalar(
-            select(DataQualityIssue.id).where(
-                DataQualityIssue.entity_type == "vehicle",
-                DataQualityIssue.entity_id == vehicle.id,
-                DataQualityIssue.status == "open",
-                DataQualityIssue.severity == "high",
-                DataQualityIssue.id != issue.id,
-            )
-        )
-        is not None
-    )
-    recommendation = _recommend_vehicle_status(
-        vehicle.operational_status, has_active_booking, has_open_high_issue
-    )
-    if recommendation is None:
         raise AppError(
             "NO_CONFLICT_DETECTED",
             "The current vehicle state no longer conflicts; nothing to apply.",
             status_code=409,
         )
-    new_status, reason = recommendation
+    new_status = recommendation.recommended_status
+    reason_code = recommendation.recommendation_code
 
     before = {"operational_status": vehicle.operational_status}
     vehicle.operational_status = new_status
     vehicle.version += 1
 
-    # Re-validate: the same recommendation function must find no further conflict.
+    # Re-validate against the same shared evaluator, over freshly-gathered facts, that
-    if _recommend_vehicle_status(new_status, has_active_booking, has_open_high_issue) is not None:
+    # applying this change actually leaves no conflict -- never trust the pre-computed
+    # recommendation alone for the post-condition.
+    post_facts = gather_vehicle_status_facts(db, vehicle, exclude_issue_id=issue.id)
+    post_check = evaluate_vehicle_status(vehicle, post_facts)
+    if post_check.recommendation_code not in (RECOMMENDATION_CODE_NO_CONFLICT,):
         raise AppError(
             "CONFLICT_STILL_PRESENT",
             "Applying the recommended status did not resolve the conflict.",
@@ -737,7 +742,7 @@ def apply_recommended_status(
         correlation_id=correlation_id,
         before=before,
         after={"operational_status": vehicle.operational_status},
-        metadata={"issue_ref": issue.public_ref, "reason": reason},
+        metadata={"issue_ref": issue.public_ref, "reason_code": reason_code},
     )
 
     issue.status = "resolved"
@@ -755,10 +760,20 @@ def apply_recommended_status(
         after={"status": "resolved"},
     )
     db.commit()
-    return issue, new_status, reason
+    return issue, new_status, reason_code
 
 
 MERGEABLE_FIELDS = ("first_name", "last_name", "email", "phone", "postal_code", "city")
+# Mirrors the column lengths in app/models/customer.py so an override can never fail with
+# a database DataError (500) instead of a validation error.
+_MERGEABLE_FIELD_MAX_LENGTH = {
+    "first_name": 80,
+    "last_name": 80,
+    "email": 200,
+    "phone": 40,
+    "postal_code": 20,
+    "city": 120,
+}
 
 
 def merge_customers(
@@ -788,12 +803,26 @@ def merge_customers(
         )
     loser_ref = next(ref for ref in candidate_refs if ref != survivor_ref)
 
-    survivor = db.scalar(select(Customer).where(Customer.public_ref == survivor_ref))
+    # Lock both rows in a deterministic order (by public_ref) so two concurrent merges
-    loser = db.scalar(select(Customer).where(Customer.public_ref == loser_ref))
+    # touching the same customers serialise instead of deadlocking or double-merging.
+    survivor = None
+    loser = None
+    for ref in sorted((survivor_ref, loser_ref)):
+        customer = db.scalar(select(Customer).where(Customer.public_ref == ref).with_for_update())
+        if ref == survivor_ref:
+            survivor = customer
+        else:
+            loser = customer
     if survivor is None or loser is None:
         raise AppError(
             "CUSTOMER_NOT_FOUND", "One of the customers could not be found.", status_code=404
         )
+    if survivor.merged_into_customer_id is not None or loser.merged_into_customer_id is not None:
+        raise AppError(
+            "CUSTOMER_ALREADY_MERGED",
+            "One of the customers has already been merged into another record.",
+            status_code=409,
+        )
 
     before = {
         "survivor": {f: getattr(survivor, f) for f in MERGEABLE_FIELDS},
@@ -805,7 +834,15 @@ def merge_customers(
             raise AppError(
                 "INVALID_FIELD_OVERRIDE", f"Field '{field_name}' cannot be merged.", status_code=422
             )
-        setattr(survivor, field_name, value)
+        cleaned = value.strip() if isinstance(value, str) else value
+        max_length = _MERGEABLE_FIELD_MAX_LENGTH[field_name]
+        if not cleaned or len(cleaned) > max_length:
+            raise AppError(
+                "INVALID_FIELD_OVERRIDE",
+                f"Field '{field_name}' must be 1 to {max_length} characters.",
+                status_code=422,
+            )
+        setattr(survivor, field_name, cleaned)
 
     rewired = db.execute(
         update(Booking).where(Booking.customer_id == loser.id).values(customer_id=survivor.id)

backend/app/services/data_quality_duplicate_scan.py

@@ -0,0 +1,110 @@
+from __future__ import annotations
+
+import uuid
+from difflib import SequenceMatcher
+from typing import Protocol, TypeVar
+
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from app.models.customer import Customer
+
+DUPLICATE_THRESHOLD = 70
+
+
+class ScanAccumulator(Protocol):
+    def bump(self, rule_type: str) -> None: ...
+
+
+ScanTypeContra = TypeVar("ScanTypeContra", bound=ScanAccumulator, contravariant=True)
+
+
+class OpenIssue(Protocol[ScanTypeContra]):
+    def __call__(
+        self,
+        db: Session,
+        scan: ScanTypeContra,
+        *,
+        rule_type: str,
+        entity_type: str,
+        entity_id: uuid.UUID,
+        severity: str,
+        summary: str,
+        entity_ref: str,
+        related_refs: list[str],
+        signals: list[dict] | None = None,
+    ) -> None: ...
+
+
+def _normalize(value: str | None) -> str:
+    return (value or "").strip().lower()
+
+
+def scan_duplicate_customers[ScanType: ScanAccumulator](
+    db: Session, scan: ScanType, open_issue: OpenIssue[ScanType]
+) -> None:
+    customers = list(
+        db.scalars(
+            select(Customer).where(
+                Customer.merged_into_customer_id.is_(None),
+                Customer.anonymized_at.is_(None),
+            )
+        ).all()
+    )
+    customers.sort(key=lambda customer: customer.public_ref)
+
+    # The threshold cannot be reached without an exact email (60 points) or phone
+    # (50 points). Block on normalized identifiers so this remains linear for the
+    # overwhelmingly common case and only scores plausible pairs.
+    candidate_pairs: set[tuple[int, int]] = set()
+    for attribute in ("email", "phone"):
+        blocks: dict[str, list[int]] = {}
+        for index, customer in enumerate(customers):
+            key = _normalize(getattr(customer, attribute))
+            if key:
+                blocks.setdefault(key, []).append(index)
+        for indices in blocks.values():
+            for offset, left in enumerate(indices):
+                candidate_pairs.update((left, right) for right in indices[offset + 1 :])
+
+    for left, right in sorted(candidate_pairs):
+        a = customers[left]
+        b = customers[right]
+        score = 0
+        signals: list[dict] = []
+        summary_parts: list[str] = []
+        if _normalize(a.email) and _normalize(a.email) == _normalize(b.email):
+            score += 60
+            signals.append({"code": "duplicate.exact_email"})
+            summary_parts.append("exact email")
+        if _normalize(a.phone) and _normalize(a.phone) == _normalize(b.phone):
+            score += 50
+            signals.append({"code": "duplicate.exact_phone"})
+            summary_parts.append("exact phone")
+        if _normalize(a.postal_code) and _normalize(a.postal_code) == _normalize(b.postal_code):
+            score += 10
+            signals.append({"code": "duplicate.same_postal_code"})
+            summary_parts.append("exact postal code")
+        name_a = f"{_normalize(a.first_name)} {_normalize(a.last_name)}"
+        name_b = f"{_normalize(b.first_name)} {_normalize(b.last_name)}"
+        ratio = SequenceMatcher(None, name_a, name_b).ratio()
+        if ratio >= 0.5:
+            score += round(ratio * 30)
+            signals.append(
+                {"code": "duplicate.similar_name", "params": {"score": round(ratio, 2)}}
+            )
+            summary_parts.append("similar name")
+
+        if score >= DUPLICATE_THRESHOLD:
+            open_issue(
+                db,
+                scan,
+                rule_type="possible_duplicate_customer",
+                entity_type="customer",
+                entity_id=a.id,
+                severity="high",
+                summary="; ".join(summary_parts) + f" (score {score})",
+                entity_ref=a.public_ref,
+                related_refs=[b.public_ref],
+                signals=signals,
+            )

backend/app/services/demo_manifest.py

@@ -11,7 +11,7 @@ from app.models.booking import Booking
 from app.models.data_quality import DataQualityIssue
 from app.models.outbox import OutboxEvent
 from app.schemas import DemoIntegrationSummaryOut, DemoManifestOut, DemoScenarioOut
-from app.services.integration_status import derive_n8n_status
+from app.services.integration_status import derive_mcp_hub_status, derive_n8n_status
 from app.services.knowledge import get_knowledge_provider
 
 settings = get_settings()
@@ -39,18 +39,14 @@ def _scenarios(db: Session) -> list[DemoScenarioOut]:
     overlap_issue = db.scalar(
         select(DataQualityIssue).where(DataQualityIssue.public_ref == "DQ-DEMO-OVERLAP")
     )
-    failed_run = db.scalar(
+    failed_run = db.scalar(select(OutboxEvent).where(OutboxEvent.event_id == _FAILED_DEMO_EVENT_ID))
-        select(OutboxEvent).where(OutboxEvent.event_id == _FAILED_DEMO_EVENT_ID)
-    )
     knowledge_health = get_knowledge_provider().health()
 
     # Human copy (title, problem statement, "demonstrates" summary) lives entirely in the
     # frontend's demo.json (scenarios.items.<id>.*) so it's available in all three UI
     # languages. This service only emits stable identifiers and message codes -- never
     # display prose -- per the message_code + params architecture used across the app.
-    return_ready = bool(
+    return_ready = bool(booking and booking.status == "active" and booking.end_odometer_km is None)
-        booking and booking.status == "active" and booking.end_odometer_km is None
-    )
     duplicate_ready = bool(duplicate_issue and duplicate_issue.status == "open")
     overlap_ready = bool(overlap_issue and overlap_issue.status == "open")
     automation_ready = bool(failed_run and failed_run.delivery_status == "failed")
@@ -65,7 +61,9 @@ def _scenarios(db: Session) -> list[DemoScenarioOut]:
             blocked_reason_code=(
                 None
                 if return_ready
-                else "bookingNotFound" if booking is None else "bookingAlreadyProcessed"
+                else "bookingNotFound"
+                if booking is None
+                else "bookingAlreadyProcessed"
             ),
         ),
         DemoScenarioOut(
@@ -81,7 +79,9 @@ def _scenarios(db: Session) -> list[DemoScenarioOut]:
             blocked_reason_code=(
                 None
                 if duplicate_ready
-                else "duplicateIssueNotFound" if duplicate_issue is None else "issueAlreadyResolved"
+                else "duplicateIssueNotFound"
+                if duplicate_issue is None
+                else "issueAlreadyResolved"
             ),
         ),
         DemoScenarioOut(
@@ -95,7 +95,9 @@ def _scenarios(db: Session) -> list[DemoScenarioOut]:
             blocked_reason_code=(
                 None
                 if overlap_ready
-                else "overlapIssueNotFound" if overlap_issue is None else "issueAlreadyResolved"
+                else "overlapIssueNotFound"
+                if overlap_issue is None
+                else "issueAlreadyResolved"
             ),
         ),
         DemoScenarioOut(
@@ -107,7 +109,9 @@ def _scenarios(db: Session) -> list[DemoScenarioOut]:
             blocked_reason_code=(
                 None
                 if automation_ready
-                else "failedEventNotFound" if failed_run is None else "eventAlreadyRecovered"
+                else "failedEventNotFound"
+                if failed_run is None
+                else "eventAlreadyRecovered"
             ),
         ),
         DemoScenarioOut(
@@ -124,6 +128,7 @@ def _scenarios(db: Session) -> list[DemoScenarioOut]:
 def _integrations(db: Session) -> list[DemoIntegrationSummaryOut]:
     n8n = derive_n8n_status(db)
     knowledge_health = get_knowledge_provider().health()
+    mcp_hub = derive_mcp_hub_status(db)
 
     return [
         DemoIntegrationSummaryOut(
@@ -138,20 +143,35 @@ def _integrations(db: Session) -> list[DemoIntegrationSummaryOut]:
         ),
         DemoIntegrationSummaryOut(
             key="ragcore",
-            status_code="operational" if knowledge_health.provider == "ragcore" else "demoMode",
+            status_code=(
-            detail_code="ragcoreDetail",
+                "operational"
+                if knowledge_health.provider == "ragcore" and knowledge_health.available
+                else "demoMode"
+                if knowledge_health.provider == "demo" and knowledge_health.available
+                else "unavailable"
+            ),
+            detail_code=(
+                "ragcoreDetail"
+                if knowledge_health.document_count is not None
+                else "ragcoreSourceDetail"
+            ),
             detail_params={
-                "count": knowledge_health.document_count,
+                "count": knowledge_health.document_count or 0,
+                "source_count": knowledge_health.source_document_count,
                 "collection": knowledge_health.collection,
             },
         ),
         DemoIntegrationSummaryOut(
             key="mcp_hub",
-            status_code="operational" if settings.mcp_hub_registration_enabled else "notConnected",
+            # `MCP_HUB_REGISTRATION_ENABLED` on its own proves nothing: registration is
+            # catalog-driven on the Hub's side, so the flag only says Fleet Ops expects
+            # to be called. Only real recorded `mcp_tool_request` calls make this
+            # "operational" -- same evidence rule the integration status page uses.
+            status_code="operational" if mcp_hub.state == "operational" else "notConnected",
             detail_code=(
-                "mcpDetailEnabled"
+                "mcpDetailOperational"
-                if settings.mcp_hub_registration_enabled
+                if mcp_hub.state == "operational"
-                else "mcpDetailNotConnected"
+                else "mcpDetailPrepared"
             ),
             detail_params={},
         ),
@@ -165,9 +185,7 @@ def scenario_integrity_report(db: Session) -> dict:
     overview already use, so this can never drift from what a visitor actually sees."""
     scenarios = _scenarios(db)
     not_ready = [
-        {"id": s.id, "reason_code": s.blocked_reason_code}
+        {"id": s.id, "reason_code": s.blocked_reason_code} for s in scenarios if not s.ready
-        for s in scenarios
-        if not s.ready
     ]
     return {"all_ready": len(not_ready) == 0, "not_ready": not_ready}
 

backend/app/services/dispatcher.py

@@ -48,6 +48,7 @@ def _reclaim_stale_deliveries(batch_size: int = 10) -> int:
                 f"(no outcome recorded within {settings.n8n_delivery_lease_seconds:.0f}s; "
                 f"the process likely crashed mid-delivery). attempts preserved at {row.attempts}."
             )[:2000]
+            row.last_error_code = "staleLeaseRecovered"
         db.commit()
         return len(rows)
     finally:
@@ -111,22 +112,42 @@ def _deliver_one(event_id: uuid.UUID) -> None:
     finally:
         db.close()
 
+    error_code: str | None
     if wire_event is None:
         success, error, body = False, payload_error, None
+        error_code = "malformedPayload"
     else:
         try:
             response = httpx.post(
                 settings.n8n_webhook_url,
                 json=wire_event,
+                headers={"X-Fleet-Ops-Trigger-Token": settings.n8n_webhook_trigger_token},
                 timeout=settings.n8n_http_timeout_seconds,
             )
             response.raise_for_status()
+            try:
                 body = response.json()
+            except ValueError:
+                body = None
+            if isinstance(body, dict):
                 success = bool(body.get("ok", True))
                 error = None if success else f"n8n reported failure: {body}"
+                error_code = None if success else "remoteReportedFailure"
+            else:
+                # A 2xx status with a non-object (or unparsable) body means the workflow
+                # itself errored before its "Respond to Webhook" node ran -- n8n's default
+                # error response still carries a 2xx-looking status here. Treat it as a
+                # failure so the event is retried rather than lost or wrongly marked
+                # succeeded.
+                success = False
+                error = (
+                    f"Unexpected non-JSON-object response from n8n (status {response.status_code})"
+                )
+                error_code = "malformedResponse"
         except httpx.HTTPError as exc:
             success = False
             error = f"{type(exc).__name__}: {exc}"
+            error_code = "connectionError"
             body = None
 
     db = SessionLocal()
@@ -138,10 +159,12 @@ def _deliver_one(event_id: uuid.UUID) -> None:
         if success:
             event.delivery_status = "succeeded"
             event.last_error = None
+            event.last_error_code = None
             event.next_attempt_at = None
             event.external_run_id = str((body or {}).get("event_id", event_id))
         else:
             event.last_error = (error or "delivery failed")[:2000]
+            event.last_error_code = error_code or "unknownError"
             if event.attempts >= settings.n8n_max_attempts:
                 event.delivery_status = "failed"
                 event.next_attempt_at = None

backend/app/services/integration_status.py

@@ -1,15 +1,58 @@
 from __future__ import annotations
 
-from typing import Literal
+import threading
+import time
+from collections.abc import Sequence
+from datetime import UTC, datetime, timedelta
+from typing import Any, Literal
 
-from sqlalchemy import func, select
+import httpx
+from sqlalchemy import Row, func, select
 from sqlalchemy.orm import Session
 
 from app.core.config import get_settings
-from app.models.outbox import OutboxEvent
+from app.models.audit import AuditEvent
-from app.schemas import N8nIntegrationStatus
+from app.models.outbox import DEMO_SCENARIO_ERROR_CODE, OutboxEvent
+from app.schemas import (
+    McpHubIntegrationStatus,
+    N8nErrorHandlerStatus,
+    N8nIntegrationStatus,
+    N8nWorkflowEvidence,
+)
 
 settings = get_settings()
+_hub_health_lock = threading.Lock()
+_hub_health_cached_at = 0.0
+_hub_health_cached_value: bool | None = None
+
+# The 4 canonical Fleet Ops n8n workflows (see n8n/workflows/MANIFEST.md). All 4 are
+# built (all with their full node set saved).
+_CANONICAL_WORKFLOWS = (
+    "Fleet Ops — Vehicle Return Orchestration",
+    "Fleet Ops — Scheduled Data Quality Scan",
+    "Fleet Ops — RAGcore Procedure Sync",
+    "Fleet Ops — Workflow Error Handler",
+)
+_STALE_AFTER = {
+    "Fleet Ops — Scheduled Data Quality Scan": timedelta(hours=2, minutes=30),
+    "Fleet Ops — RAGcore Procedure Sync": timedelta(hours=30),
+}
+
+
+def _latest_rows_per_workflow(db: Session, action: str) -> Sequence[Row[Any]]:
+    """Return the most recent audit row per workflow name for one action.
+
+    Heartbeats arrive on every scheduled run, so loading *all* rows and picking the
+    latest in Python would grow linearly with deployment age. ``DISTINCT ON`` lets
+    PostgreSQL return exactly one (latest) row per workflow instead.
+    """
+    workflow_name = AuditEvent.after_json["workflow_name"].astext
+    return db.execute(
+        select(AuditEvent.occurred_at, AuditEvent.after_json, AuditEvent.metadata_json)
+        .where(AuditEvent.action == action)
+        .distinct(workflow_name)
+        .order_by(workflow_name, AuditEvent.occurred_at.desc())
+    ).all()
 
 
 def derive_n8n_status(db: Session) -> N8nIntegrationStatus:
@@ -23,25 +66,163 @@ def derive_n8n_status(db: Session) -> N8nIntegrationStatus:
     failed = counts.get("failed", 0)
     succeeded = counts.get("succeeded", 0)
 
+    # Prepared demo failures are props, not health signals. They stay visible and
+    # counted -- hiding them would be its own kind of lie -- but they are counted
+    # *separately*, and only genuinely unexpected failures are allowed to move n8n off
+    # "operational". Without this split the demo seed's single staged failure pins the
+    # integration to "degraded" forever, which tells a viewer something untrue about
+    # the automation.
+    demo_scenario_failed = (
+        db.scalar(
+            select(func.count())
+            .select_from(OutboxEvent)
+            .where(
+                OutboxEvent.delivery_status == "failed",
+                OutboxEvent.last_error_code == DEMO_SCENARIO_ERROR_CODE,
+            )
+        )
+        or 0
+    )
+    unexpected_failed = max(failed - demo_scenario_failed, 0)
+
     latest_success_at = db.scalar(
         select(func.max(OutboxEvent.updated_at)).where(OutboxEvent.delivery_status == "succeeded")
     )
+    # Health talks about real failures only, so the "latest failure" a health reader
+    # sees must exclude the staged one too.
     latest_failure_at = db.scalar(
-        select(func.max(OutboxEvent.updated_at)).where(OutboxEvent.delivery_status == "failed")
+        select(func.max(OutboxEvent.updated_at)).where(
+            OutboxEvent.delivery_status == "failed",
+            OutboxEvent.last_error_code != DEMO_SCENARIO_ERROR_CODE,
+        )
+    )
+    latest_demo_scenario_at = db.scalar(
+        select(func.max(OutboxEvent.updated_at)).where(
+            OutboxEvent.delivery_status == "failed",
+            OutboxEvent.last_error_code == DEMO_SCENARIO_ERROR_CODE,
+        )
     )
 
     state: Literal["disabled", "unavailable", "degraded", "operational", "no_evidence"]
     if not settings.n8n_dispatch_enabled:
         state = "disabled"
-    elif failed > 0 and succeeded == 0:
+    elif unexpected_failed > 0 and succeeded == 0:
         state = "unavailable"
-    elif failed > 0:
+    elif unexpected_failed > 0:
         state = "degraded"
     elif succeeded > 0 or pending > 0 or delivering > 0:
         state = "operational"
     else:
         state = "no_evidence"
 
+    # Scheduled scan evidence: only service-triggered runs count as n8n evidence, not
+    # runs an operator triggered manually from the Data Quality page.
+    latest_scan_at = db.scalar(
+        select(func.max(AuditEvent.occurred_at)).where(
+            AuditEvent.action == "data_quality_scan_run",
+            AuditEvent.actor_type == "service",
+        )
+    )
+
+    # RAGcore Procedure Sync evidence: result reports posted by the workflow itself once
+    # it finishes uploading procedures to RAGcore (app/api/routers/integrations.py::
+    # procedures_sync_result), the same "the workflow's own callback is the evidence"
+    # pattern the scheduled scan and error handler already use below.
+    latest_procedure_sync_at = db.scalar(
+        select(func.max(AuditEvent.occurred_at)).where(AuditEvent.action == "n8n_procedures_synced")
+    )
+
+    # Error handler evidence: registrations posted by the "Fleet Ops — Workflow Error
+    # Handler" n8n workflow itself, which also doubles as proof that workflow is wired
+    # up and firing correctly.
+    total_failures_registered = (
+        db.scalar(
+            select(func.count(AuditEvent.id)).where(
+                AuditEvent.action == "n8n_workflow_failure_registered"
+            )
+        )
+        or 0
+    )
+    latest_failure_row = db.execute(
+        select(AuditEvent.occurred_at, AuditEvent.after_json)
+        .where(AuditEvent.action == "n8n_workflow_failure_registered")
+        .order_by(AuditEvent.occurred_at.desc())
+        .limit(1)
+    ).first()
+    latest_handler_failure_at = latest_failure_row[0] if latest_failure_row else None
+    latest_handler_failure_workflow = (
+        (latest_failure_row[1] or {}).get("workflow_name") if latest_failure_row else None
+    )
+
+    legacy_evidence_by_workflow = {
+        "Fleet Ops — Vehicle Return Orchestration": latest_success_at,
+        "Fleet Ops — Scheduled Data Quality Scan": latest_scan_at,
+        "Fleet Ops — RAGcore Procedure Sync": latest_procedure_sync_at,
+        "Fleet Ops — Workflow Error Handler": latest_handler_failure_at,
+    }
+    heartbeat_by_workflow: dict[str, tuple[datetime, str, str | None]] = {}
+    heartbeat_rows = _latest_rows_per_workflow(db, "n8n_workflow_heartbeat")
+    for occurred_at, after, metadata in heartbeat_rows:
+        workflow_name = (after or {}).get("workflow_name")
+        if workflow_name in _CANONICAL_WORKFLOWS and workflow_name not in heartbeat_by_workflow:
+            heartbeat_by_workflow[workflow_name] = (
+                occurred_at,
+                (after or {}).get("status", "succeeded"),
+                (metadata or {}).get("execution_id"),
+            )
+
+    failure_by_workflow: dict[str, tuple[datetime, str | None]] = {}
+    failure_rows = _latest_rows_per_workflow(db, "n8n_workflow_failure_registered")
+    for occurred_at, after, metadata in failure_rows:
+        workflow_name = (after or {}).get("workflow_name")
+        if workflow_name in _CANONICAL_WORKFLOWS and workflow_name not in failure_by_workflow:
+            failure_by_workflow[workflow_name] = (
+                occurred_at,
+                (metadata or {}).get("execution_id"),
+            )
+
+    now = datetime.now(UTC)
+    workflows: list[N8nWorkflowEvidence] = []
+    for name in _CANONICAL_WORKFLOWS:
+        legacy_seen = legacy_evidence_by_workflow[name]
+        heartbeat = heartbeat_by_workflow.get(name)
+        failure_signal = failure_by_workflow.get(name)
+        seen_at = heartbeat[0] if heartbeat else legacy_seen
+        last_status: Literal["succeeded", "failed"] | None = (
+            "succeeded" if seen_at is not None else None
+        )
+        execution_id = heartbeat[2] if heartbeat else None
+        if heartbeat and heartbeat[1] == "failed":
+            last_status = "failed"
+        if failure_signal and (seen_at is None or failure_signal[0] > seen_at):
+            seen_at = failure_signal[0]
+            last_status = "failed"
+            execution_id = failure_signal[1]
+        workflow_state: Literal["no_evidence", "healthy", "stale", "failed"]
+        if seen_at is None:
+            workflow_state = "no_evidence"
+        elif last_status == "failed":
+            workflow_state = "failed"
+        elif name in _STALE_AFTER and now - seen_at > _STALE_AFTER[name]:
+            workflow_state = "stale"
+        else:
+            workflow_state = "healthy"
+        workflows.append(
+            N8nWorkflowEvidence(
+                name=name,
+                built=True,
+                last_seen_at=seen_at,
+                state=workflow_state,
+                last_status=last_status,
+                last_execution_id=execution_id,
+            )
+        )
+
+    if state in {"operational", "no_evidence"} and any(
+        workflow.state in {"failed", "stale"} for workflow in workflows
+    ):
+        state = "degraded"
+
     return N8nIntegrationStatus(
         configured=bool(settings.n8n_webhook_url),
         dispatch_enabled=settings.n8n_dispatch_enabled,
@@ -49,7 +230,77 @@ def derive_n8n_status(db: Session) -> N8nIntegrationStatus:
         pending=pending,
         delivering=delivering,
         failed=failed,
+        unexpected_failed=unexpected_failed,
+        demo_scenario_failed=demo_scenario_failed,
         succeeded=succeeded,
         latest_success_at=latest_success_at,
         latest_failure_at=latest_failure_at,
+        latest_demo_scenario_at=latest_demo_scenario_at,
+        expected_workflow_count=len(_CANONICAL_WORKFLOWS),
+        known_workflow_count=sum(1 for w in workflows if w.last_seen_at is not None),
+        workflows=workflows,
+        error_handler=N8nErrorHandlerStatus(
+            total_failures_registered=total_failures_registered,
+            latest_failure_at=latest_handler_failure_at,
+            latest_failure_workflow=latest_handler_failure_workflow,
+        ),
     )
+
+
+def derive_mcp_hub_status(db: Session) -> McpHubIntegrationStatus:
+    """Evidence-based MCP Hub status: real tool-call audit history, not just the
+    `MCP_HUB_REGISTRATION_ENABLED` flag flipped on. Every `mcp_tool_request` call
+    already writes an `AuditEvent` (see `app/api/routers/mcp_integrations.py`)."""
+    total_calls = (
+        db.scalar(select(func.count(AuditEvent.id)).where(AuditEvent.action == "mcp_tool_request"))
+        or 0
+    )
+    latest_call_row = db.execute(
+        select(AuditEvent.occurred_at, AuditEvent.actor_label, AuditEvent.metadata_json)
+        .where(AuditEvent.action == "mcp_tool_request")
+        .order_by(AuditEvent.occurred_at.desc())
+        .limit(1)
+    ).first()
+    last_called_at = latest_call_row[0] if latest_call_row else None
+    last_client = latest_call_row[1] if latest_call_row else None
+    last_tool = (latest_call_row[2] or {}).get("tool") if latest_call_row else None
+
+    state: Literal["not_configured", "no_evidence", "operational"]
+    if not settings.mcp_hub_registration_enabled:
+        state = "not_configured"
+    elif total_calls > 0:
+        state = "operational"
+    else:
+        state = "no_evidence"
+
+    hub_reachable = _check_hub_reachable()
+
+    return McpHubIntegrationStatus(
+        registration_enabled=settings.mcp_hub_registration_enabled,
+        state=state,
+        total_calls=total_calls,
+        last_tool=last_tool,
+        last_client=last_client,
+        last_called_at=last_called_at,
+        hub_reachable=hub_reachable,
+    )
+
+
+def _check_hub_reachable() -> bool | None:
+    """Real Hub-side health signal (MCP Hub's own registration is catalog-driven on
+    its side, so this is the only thing Fleet Ops itself can honestly check).
+    `None` means not configured / not checked, never a guess."""
+    if not settings.mcp_hub_base_url:
+        return None
+    global _hub_health_cached_at, _hub_health_cached_value
+    now = time.monotonic()
+    with _hub_health_lock:
+        if now - _hub_health_cached_at < settings.mcp_hub_health_cache_seconds:
+            return _hub_health_cached_value
+        try:
+            response = httpx.get(f"{settings.mcp_hub_base_url.rstrip('/')}/health", timeout=1.5)
+            _hub_health_cached_value = response.status_code == 200
+        except httpx.HTTPError:
+            _hub_health_cached_value = False
+        _hub_health_cached_at = time.monotonic()
+        return _hub_health_cached_value

backend/app/services/knowledge/__init__.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from datetime import datetime
 from functools import lru_cache
 from typing import Literal, Protocol
 
@@ -8,6 +9,7 @@ from pydantic import BaseModel
 from app.core.config import get_settings
 
 EvidenceState = Literal["grounded", "insufficient", "unavailable"]
+KnowledgeStatisticsState = Literal["verified", "sync_reported", "not_reported"]
 
 
 class SourceCard(BaseModel):
@@ -33,7 +35,14 @@ class KnowledgeHealth(BaseModel):
     tenant: str
     workspace: str
     collection: str
-    document_count: int
+    # A provider may be healthy without exposing a corpus-size endpoint. `None` means
+    # unknown, never "zero procedures".
+    document_count: int | None
+    source_document_count: int
+    reported_synced_document_count: int | None
+    reported_failed_document_count: int | None
+    last_sync_at: datetime | None
+    statistics_state: KnowledgeStatisticsState
 
 
 class KnowledgeProvider(Protocol):

backend/app/services/knowledge/demo.py

@@ -7,35 +7,161 @@ from pathlib import Path
 
 from app.core.config import get_settings
 from app.services.knowledge import GroundedAnswer, KnowledgeHealth, SourceCard
+from app.services.knowledge.procedures import parse_frontmatter
 
 SUPPORTED_LANGUAGES = ("nl-BE", "en-GB", "fr-BE")
 DEFAULT_LANGUAGE = "en-GB"
 
 STOPWORDS_BY_LANGUAGE: dict[str, set[str]] = {
     "en-GB": {
-        "a", "an", "the", "is", "are", "was", "were", "be", "been", "being",
+        "a",
-        "to", "of", "in", "on", "at", "for", "and", "or", "but", "if", "then",
+        "an",
-        "do", "does", "did", "must", "may", "can", "could", "should", "would",
+        "the",
-        "i", "you", "it", "we", "they", "my", "your", "what", "when", "how",
+        "is",
-        "with", "without", "this", "that", "these", "those", "not", "no",
+        "are",
+        "was",
+        "were",
+        "be",
+        "been",
+        "being",
+        "to",
+        "of",
+        "in",
+        "on",
+        "at",
+        "for",
+        "and",
+        "or",
+        "but",
+        "if",
+        "then",
+        "do",
+        "does",
+        "did",
+        "must",
+        "may",
+        "can",
+        "could",
+        "should",
+        "would",
+        "i",
+        "you",
+        "it",
+        "we",
+        "they",
+        "my",
+        "your",
+        "what",
+        "when",
+        "how",
+        "with",
+        "without",
+        "this",
+        "that",
+        "these",
+        "those",
+        "not",
+        "no",
     },
     "nl-BE": {
-        "een", "de", "het", "is", "zijn", "was", "waren", "worden", "wordt",
+        "een",
-        "van", "in", "op", "voor", "en", "of", "maar", "als", "dan",
+        "de",
-        "moet", "mag", "kan", "kunnen", "zou", "zouden",
+        "het",
-        "ik", "jij", "u", "we", "wij", "zij", "mijn", "jouw", "wat", "wanneer", "hoe",
+        "is",
-        "met", "zonder", "dit", "dat", "deze", "die", "niet", "geen",
+        "zijn",
+        "was",
+        "waren",
+        "worden",
+        "wordt",
+        "van",
+        "in",
+        "op",
+        "voor",
+        "en",
+        "of",
+        "maar",
+        "als",
+        "dan",
+        "moet",
+        "mag",
+        "kan",
+        "kunnen",
+        "zou",
+        "zouden",
+        "ik",
+        "jij",
+        "u",
+        "we",
+        "wij",
+        "zij",
+        "mijn",
+        "jouw",
+        "wat",
+        "wanneer",
+        "hoe",
+        "met",
+        "zonder",
+        "dit",
+        "dat",
+        "deze",
+        "die",
+        "niet",
+        "geen",
     },
     "fr-BE": {
-        "un", "une", "le", "la", "les", "des", "est", "sont", "était", "être",
+        "un",
-        "de", "du", "en", "sur", "pour", "et", "ou", "mais", "si", "alors",
+        "une",
-        "doit", "peut", "peuvent", "pourrait", "devrait",
+        "le",
-        "je", "tu", "vous", "il", "elle", "nous", "ils", "mon", "votre", "quoi", "quand", "comment",
+        "la",
-        "avec", "sans", "ce", "cette", "ces", "cela", "pas", "non",
+        "les",
+        "des",
+        "est",
+        "sont",
+        "était",
+        "être",
+        "de",
+        "du",
+        "en",
+        "sur",
+        "pour",
+        "et",
+        "ou",
+        "mais",
+        "si",
+        "alors",
+        "doit",
+        "peut",
+        "peuvent",
+        "pourrait",
+        "devrait",
+        "je",
+        "tu",
+        "vous",
+        "il",
+        "elle",
+        "nous",
+        "ils",
+        "mon",
+        "votre",
+        "quoi",
+        "quand",
+        "comment",
+        "avec",
+        "sans",
+        "ce",
+        "cette",
+        "ces",
+        "cela",
+        "pas",
+        "non",
     },
 }
 
-_WORD_RE = re.compile(r"[a-z0-9]+")
+# Includes the Latin-1 accented-letter range (à-ö, ø-ÿ) so French/Dutch words with
+# diacritics (véhicule, réservation, geëscaleerd) tokenize as one word instead of
+# splitting apart at the accented character -- a plain [a-z0-9]+ pattern silently
+# drops every accent and fragments the word either side of it.
+_WORD_RE = re.compile(r"[a-zà-öø-ÿ0-9]+")
 
 
 def _stem(word: str) -> str:
@@ -70,23 +196,6 @@ class ScoredSection:
     body_tokens: set[str]
 
 
-def _parse_frontmatter(raw: str) -> tuple[dict[str, str], str]:
-    if not raw.startswith("---"):
-        return {}, raw
-    end = raw.find("\n---", 3)
-    if end == -1:
-        return {}, raw
-    block = raw[3:end].strip()
-    body = raw[end + 4 :].lstrip("\n")
-    meta: dict[str, str] = {}
-    for line in block.splitlines():
-        if ":" not in line:
-            continue
-        key, _, value = line.partition(":")
-        meta[key.strip()] = value.strip().strip('"')
-    return meta, body
-
-
 def _split_sections(body: str) -> list[tuple[str, str]]:
     sections: list[tuple[str, str]] = []
     current_heading = "Overview"
@@ -110,7 +219,7 @@ def _load_sections(procedures_dir: Path, language: str) -> list[ScoredSection]:
     sections: list[ScoredSection] = []
     for path in sorted(procedures_dir.glob("*.md")):
         raw = path.read_text(encoding="utf-8")
-        meta, body = _parse_frontmatter(raw)
+        meta, body = parse_frontmatter(raw)
         title = meta.get("title", path.stem)
         doc = Document(
             document_id=meta.get("document_id", path.stem),
@@ -153,7 +262,7 @@ _LOW_CONFIDENCE_TEXT = {
 _LEAD_ANSWER_TEMPLATE = {
     "en-GB": 'Per "{title}" (v{version}), section "{heading}": {excerpt}',
     "nl-BE": 'Volgens "{title}" (v{version}), sectie "{heading}": {excerpt}',
-    "fr-BE": 'Selon « {title} » (v{version}), section « {heading} » : {excerpt}',
+    "fr-BE": "Selon « {title} » (v{version}), section « {heading} » : {excerpt}",
 }
 
 
@@ -213,22 +322,38 @@ class DemoKnowledgeProvider:
             workspace=self._settings.ragcore_workspace,
             collection=self._settings.ragcore_collection,
             document_count=self._document_count_by_language[language],
+            source_document_count=self._document_count_by_language[language],
+            reported_synced_document_count=None,
+            reported_failed_document_count=None,
+            last_sync_at=None,
+            statistics_state="verified",
         )
 
     def _score(
         self, query_tokens: set[str], section: ScoredSection, idf: dict[str, float]
     ) -> float:
+        # The section body is the strongest relevance signal -- it's the actual
+        # substance a heading or title can only hint at -- so a body match is weighted
+        # *above* heading/title matches, not below them. The previous 3x/2x/1x
+        # (heading/title/body) ordering let a single generic word in a heading (e.g.
+        # "vehicle", present in nearly every section) or a document's own title
+        # outrank a section whose body genuinely covers multiple, more distinctive
+        # query terms -- confirmed to misrank the brief's exact validation question in
+        # every one of the three languages (see docs/fleet-ops-correction/
+        # current-gap-audit.md and i18n-inventory.md): nl-BE picked a checkout section
+        # over the damage procedure, en-GB and fr-BE picked the return procedure over
+        # the damage procedure, purely from heading/title overlap on common words.
         score = 0.0
         for token in query_tokens:
             token_idf = idf.get(token, 0.0)
             if token_idf == 0.0:
                 continue
-            if token in section.heading_tokens:
+            if token in section.body_tokens:
                 score += 3 * token_idf
-            elif token in section.document.title_tokens:
+            elif token in section.heading_tokens:
                 score += 2 * token_idf
-            elif token in section.body_tokens:
+            elif token in section.document.title_tokens:
-                score += token_idf
+                score += 1.5 * token_idf
         return score
 
     def ask(

backend/app/services/knowledge/procedures.py

@@ -0,0 +1,73 @@
+from __future__ import annotations
+
+import hashlib
+import uuid
+from dataclasses import dataclass
+from pathlib import Path
+
+SUPPORTED_LANGUAGES = ("nl-BE", "en-GB", "fr-BE")
+
+# Stable across runs (and across which language ships first) so a document's RAGcore
+# source_id never changes just because the sync ran on a different day or in a
+# different order -- required for RAGcore's upload idempotency to work per document.
+_SOURCE_ID_NAMESPACE = uuid.uuid5(
+    uuid.NAMESPACE_URL, "https://mobilityops.internal/knowledge/procedures"
+)
+
+
+def parse_frontmatter(raw: str) -> tuple[dict[str, str], str]:
+    if not raw.startswith("---"):
+        return {}, raw
+    end = raw.find("\n---", 3)
+    if end == -1:
+        return {}, raw
+    block = raw[3:end].strip()
+    body = raw[end + 4 :].lstrip("\n")
+    meta: dict[str, str] = {}
+    for line in block.splitlines():
+        if ":" not in line:
+            continue
+        key, _, value = line.partition(":")
+        meta[key.strip()] = value.strip().strip('"')
+    return meta, body
+
+
+@dataclass(frozen=True)
+class ProcedureDocument:
+    source_id: str
+    language: str
+    document_id: str
+    title: str
+    version: str
+    content: str
+    content_hash: str
+
+
+def iter_procedure_documents(knowledge_dir: Path) -> list[ProcedureDocument]:
+    """Read every procedure Markdown file Fleet Ops ships, across every supported
+    language, as a flat list ready for external sync (e.g. into RAGcore). Frontmatter
+    fields (title, version) come from the same files the demo knowledge provider
+    already reads -- see parse_frontmatter -- so the two never drift apart."""
+
+    documents: list[ProcedureDocument] = []
+    for language in SUPPORTED_LANGUAGES:
+        language_dir = knowledge_dir / language
+        if not language_dir.is_dir():
+            continue
+        for path in sorted(language_dir.glob("*.md")):
+            raw = path.read_text(encoding="utf-8")
+            meta, body = parse_frontmatter(raw)
+            document_id = meta.get("document_id", path.stem)
+            content = body.strip()
+            documents.append(
+                ProcedureDocument(
+                    source_id=str(uuid.uuid5(_SOURCE_ID_NAMESPACE, f"{language}:{document_id}")),
+                    language=language,
+                    document_id=document_id,
+                    title=meta.get("title", path.stem),
+                    version=meta.get("version", "1.0"),
+                    content=content,
+                    content_hash=hashlib.sha256(content.encode("utf-8")).hexdigest(),
+                )
+            )
+    return documents

backend/app/services/knowledge/ragcore.py

@@ -1,26 +1,159 @@
 from __future__ import annotations
 
+from concurrent.futures import ThreadPoolExecutor
+from pathlib import Path
+from threading import Lock
+from time import monotonic
+
 import httpx
 
 from app.core.config import get_settings
-from app.services.knowledge import GroundedAnswer, KnowledgeHealth, SourceCard
+from app.core.observability import KNOWLEDGE_PROVIDER_REQUESTS, KNOWLEDGE_RETRIEVAL_SCORE
+from app.services.knowledge import EvidenceState, GroundedAnswer, KnowledgeHealth, SourceCard
+from app.services.knowledge.procedures import ProcedureDocument, iter_procedure_documents
+
+_GROUNDED_ANSWERABILITY = {"answerable", "partially_answerable"}
+
+# Mirrors DemoKnowledgeProvider's own extractive template in spirit: a real cited
+# excerpt wrapped in a fixed sentence, never a generated summary. Used only as a
+# fallback when RAGcore's own /v1/answers (generation + citation validation) is
+# unavailable but its retrieval (/v1/search) still returns real, relevant, cited
+# results -- see ask() below. Unlike the demo corpus's own markdown frontmatter, RAGcore's
+# `document_version_id` is an opaque UUID, not a human-meaningful version string, so it
+# is deliberately left out of this sentence (it still appears on the source card itself).
+_LEAD_ANSWER_TEMPLATE = {
+    "en-GB": 'Per "{title}": {excerpt}',
+    "nl-BE": 'Volgens "{title}": {excerpt}',
+    "fr-BE": "Selon « {title} » : {excerpt}",
+}
+_DEFAULT_LANGUAGE = "en-GB"
+_MAX_SOURCE_CARDS = 3
+_INDEX_LOOKUP_TIMEOUT_SECONDS = 2.0
+_INDEX_VERIFICATION_TTL_SECONDS = 300.0
+_INDEX_VERIFICATION_WORKERS = 6
+
+_DOMAIN_CONCEPTS: dict[str, tuple[str, ...]] = {
+    "damage": ("damage", "damaged", "schade", "beschadigd", "dommage", "endommagé"),
+    "vehicle": ("vehicle", "car", "voertuig", "wagen", "véhicule", "voiture"),
+    "return": ("return", "returned", "retour", "terugbrengen", "restitution"),
+    "fuel": ("fuel", "brandstof", "carburant"),
+    "booking": ("booking", "reservation", "boeking", "réservation"),
+    "odometer": ("odometer", "mileage", "kilometer", "kilométrage", "compteur"),
+    "customer": ("customer", "client", "klant"),
+    "cleaning": ("cleaning", "reiniging", "poetsen", "nettoyage"),
+    "maintenance": ("maintenance", "service", "onderhoud", "entretien"),
+    "conflict": ("conflict", "overlap", "overlapping", "conflit", "chevauchement"),
+    "checkout": ("checkout", "departure", "vertrek", "départ"),
+    "availability": ("available", "availability", "beschikbaar", "disponible", "disponibilité"),
+    "technical": ("technical", "warning", "technisch", "waarschuwing", "technique", "alerte"),
+}
+
+
+def _question_concepts(question: str) -> set[str]:
+    normalized = question.casefold()
+    return {
+        concept
+        for concept, terms in _DOMAIN_CONCEPTS.items()
+        if any(term in normalized for term in terms)
+    }
+
+
+def _deduplicate_sources(sources: list[SourceCard]) -> list[SourceCard]:
+    """Collapse duplicate chunks and re-uploaded document versions.
+
+    RAGcore document/version UUIDs change across uploads, so they are not useful
+    deduplication keys. Human-visible citation identity is the normalized title,
+    section. Chunks from the same unsectioned document collapse into one card; distinct
+    named sections remain independently citable.
+    """
+    seen: set[tuple[str, str]] = set()
+    unique: list[SourceCard] = []
+    for source in sources:
+        key = (
+            source.title.strip().casefold(),
+            source.section.strip().casefold(),
+        )
+        if key in seen:
+            continue
+        seen.add(key)
+        unique.append(source)
+        if len(unique) == _MAX_SOURCE_CARDS:
+            break
+    return unique
+
+
+def _rank_sources_for_concepts(sources: list[SourceCard], concepts: set[str]) -> list[SourceCard]:
+    if "damage" not in concepts:
+        return sources
+    return sorted(
+        sources,
+        key=lambda source: (
+            0 if "damage" in f"{source.document_id} {source.title}".casefold() else 1
+        ),
+    )
+
+
+def _retrieval_score(result: dict) -> float | None:
+    scores = result.get("scores")
+    if not isinstance(scores, dict):
+        return None
+    for name in ("rerank", "fused"):
+        value = scores.get(name)
+        if isinstance(value, int | float) and not isinstance(value, bool):
+            return float(value)
+    return None
 
 
 class RAGcoreKnowledgeProvider:
-    """Adapter for the central RAGcore service.
+    """Adapter for the central RAGcore service, against its real `/v1/*` contract
+    (see `docs/contracts/openapi.yaml` in the RAGcore checkout -- RAGcore is built and
+    owned separately, MobilityOps only ever talks to its documented HTTP API).
 
-    RAGcore is built and owned separately (see contracts/ragcore-contract-assumptions.md).
+    Authenticates as a service account via `Authorization: Bearer <token>` (RAGcore's
-    No live RAGcore instance was reachable during this build, so the exact request/response
+    session-cookie auth is for its own browser admin UI only). Any connection error,
-    shape below is a best-effort guess at a REST contract; any failure (connection, timeout,
+    timeout, non-2xx response, or malformed body degrades to `evidence_state:
-    malformed response) degrades to `unavailable` rather than raising, per the architecture's
+    "unavailable"` rather than raising -- this is the adapter that actually exercises the
-    reliability boundary: RAGcore failure disables knowledge answers only, never the rest of
+    architecture's reliability boundary: RAGcore failure disables knowledge answers only,
-    the app, and never fabricates an answer.
+    never fabricates an answer, never affects the rest of the app.
+
+    `/v1/answers` (RAGcore's own generation + citation-validation step) is tried first;
+    if it is itself unavailable (non-2xx or unreachable -- as opposed to a real 200
+    classifying the question as insufficiently answerable), `ask()` falls back to
+    RAGcore's `/v1/search` retrieval, which is a materially different, simpler pipeline
+    stage with no generation step to fail. The fallback answer is always an extractive
+    excerpt RAGcore's own search actually found, wrapped in the same fixed citation
+    template `DemoKnowledgeProvider` uses -- never a fabricated summary.
+
+    Known gap, not fixable from this side: RAGcore's ingest pipeline currently tags every
+    chunk's `language` payload field as `"en"` regardless of actual document language (the
+    `/v1/uploads` contract has no per-file language field for a caller to set correctly).
+    Filtering search/answer requests by requested UI language would therefore silently
+    exclude genuinely-relevant nl-BE/fr-BE content, so this adapter deliberately does not
+    filter by language -- retrieval relies on the embedding model's cross-lingual matching.
     """
 
     name = "ragcore"
 
     def __init__(self) -> None:
         self._settings = get_settings()
+        self._verification_cache: dict[str, tuple[float, int]] = {}
+        self._verification_lock = Lock()
+        self._answers_circuit_lock = Lock()
+        self._answers_circuit_open_until = 0.0
+
+    def _answers_circuit_is_open(self) -> bool:
+        with self._answers_circuit_lock:
+            return monotonic() < self._answers_circuit_open_until
+
+    def _open_answers_circuit(self) -> None:
+        with self._answers_circuit_lock:
+            self._answers_circuit_open_until = monotonic() + max(
+                0.0, self._settings.ragcore_answers_circuit_breaker_seconds
+            )
+
+    def _close_answers_circuit(self) -> None:
+        with self._answers_circuit_lock:
+            self._answers_circuit_open_until = 0.0
 
     def _client(self) -> httpx.Client:
         headers = {}
@@ -33,13 +166,36 @@ class RAGcoreKnowledgeProvider:
         )
 
     def health(self, language: str = "en-GB") -> KnowledgeHealth:
+        documents = [
+            document
+            for document in iter_procedure_documents(Path(self._settings.knowledge_dir))
+            if document.language == language
+        ]
+        verified_document_count: int | None = None
         try:
             with self._client() as client:
-                response = client.get("/health")
+                response = client.get("/health/ready")
-                response.raise_for_status()
+                body = response.json()
-                available = True
+                if not isinstance(body, dict):
-                detail = "RAGcore reachable."
+                    raise ValueError("health response is not a JSON object")
-        except httpx.HTTPError as exc:
+                available = response.status_code == 200 and body.get("status") == "ok"
+                detail = (
+                    "RAGcore reachable and ready."
+                    if available
+                    else f"RAGcore degraded: {body.get('status', 'unknown')}"
+                )
+                if available:
+                    verified_document_count = self._verify_indexed_documents(
+                        client, language, documents
+                    )
+                    if verified_document_count is None:
+                        detail += " Index verification is temporarily unavailable."
+                    else:
+                        detail += (
+                            f" {verified_document_count}/{len(documents)} managed sources have "
+                            "an exact active published document in the configured space."
+                        )
+        except (httpx.HTTPError, ValueError) as exc:
             available = False
             detail = f"RAGcore unavailable: {type(exc).__name__}: {exc}"
         return KnowledgeHealth(
@@ -49,51 +205,251 @@ class RAGcoreKnowledgeProvider:
             tenant=self._settings.ragcore_tenant,
             workspace=self._settings.ragcore_workspace,
             collection=self._settings.ragcore_collection,
-            document_count=0,
+            # RAGcore deliberately has no browse/count endpoint. Fleet Ops instead
+            # verifies each managed source through its exact identity lookup and only
+            # counts an active document with a published active version. RAGcore's
+            # content_sha256 describes its canonical parsed artifact, not the uploaded
+            # source bytes, so comparing it with Fleet Ops's source hash would be false.
+            document_count=verified_document_count,
+            source_document_count=len(documents),
+            reported_synced_document_count=None,
+            reported_failed_document_count=None,
+            last_sync_at=None,
+            statistics_state=(
+                "verified" if verified_document_count is not None else "not_reported"
+            ),
         )
 
-    def ask(self, question: str, correlation_id: str, language: str = "en-GB") -> GroundedAnswer:
+    def _verify_indexed_documents(
-        try:
+        self, client: httpx.Client, language: str, documents: list[ProcedureDocument]
-            with self._client() as client:
+    ) -> int | None:
-                response = client.post(
+        if not self._settings.ragcore_space_id or not documents:
-                    "/api/v1/ask",
+            return None
-                    json={
+
-                        "tenant": self._settings.ragcore_tenant,
+        now = monotonic()
-                        "workspace": self._settings.ragcore_workspace,
+        with self._verification_lock:
-                        "collection": self._settings.ragcore_collection,
+            cached = self._verification_cache.get(language)
-                        "question": question,
+            if cached is not None and now - cached[0] < _INDEX_VERIFICATION_TTL_SECONDS:
-                        "correlation_id": correlation_id,
+                return cached[1]
-                        "language": language,
+
+        def is_verified(document: ProcedureDocument) -> bool:
+            response = client.get(
+                "/v1/documents",
+                params={
+                    "source_id": document.source_id,
+                    "external_id": f"{document.document_id}.md",
                 },
+                timeout=_INDEX_LOOKUP_TIMEOUT_SECONDS,
             )
-                response.raise_for_status()
+            if response.status_code != 200:
+                raise RuntimeError("RAGcore document verification failed")
             body = response.json()
-        except (httpx.HTTPError, ValueError):
+            items = body.get("items") if isinstance(body, dict) else None
-            return GroundedAnswer(
+            if not isinstance(items, list) or len(items) != 1 or not isinstance(items[0], dict):
+                return False
+            item = items[0]
+            active_version = item.get("active_version")
+            return bool(
+                item.get("space_id") == self._settings.ragcore_space_id
+                and item.get("source_id") == document.source_id
+                and item.get("external_id") == f"{document.document_id}.md"
+                and item.get("status") == "active"
+                and isinstance(active_version, dict)
+                and active_version.get("status") == "published"
+            )
+
+        try:
+            with ThreadPoolExecutor(
+                max_workers=min(_INDEX_VERIFICATION_WORKERS, len(documents))
+            ) as executor:
+                verified_count = sum(executor.map(is_verified, documents))
+        except (httpx.HTTPError, RuntimeError, TypeError, ValueError):
+            return None
+
+        with self._verification_lock:
+            self._verification_cache[language] = (monotonic(), verified_count)
+        return verified_count
+
+    def ask(self, question: str, correlation_id: str, language: str = "en-GB") -> GroundedAnswer:
+        unavailable = GroundedAnswer(
             answer="",
             evidence_state="unavailable",
             sources=[],
             provider=self.name,
             correlation_id=correlation_id,
         )
+        if not self._settings.ragcore_space_id:
+            return unavailable
+
+        if self._answers_circuit_is_open():
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("answers", "circuit_open").inc()
+        else:
+            answered = self._ask_via_answers(question, correlation_id)
+            if answered is not None:
+                return answered
+        # /v1/answers itself is unavailable (non-2xx or unreachable) -- fall back to
+        # real retrieval rather than degrading straight to "unavailable". This never
+        # fabricates an answer to the question: it only ever shows an actually-cited
+        # excerpt RAGcore's own search already found, using the same extractive
+        # citation-wrapper template DemoKnowledgeProvider uses, never RAGcore's
+        # generation step.
+        return self._ask_via_search_fallback(question, correlation_id, language)
+
+    def _ask_via_answers(self, question: str, correlation_id: str) -> GroundedAnswer | None:
+        """Returns None (not a GroundedAnswer) when /v1/answers itself is unavailable,
+        so the caller can fall back to search -- as opposed to a real 200 response
+        classifying the question as insufficiently answerable, which is a genuine,
+        final result, not a reason to fall back."""
+        try:
+            with self._client() as client:
+                response = client.post(
+                    "/v1/answers",
+                    json={
+                        "query": question,
+                        "requested_space_ids": [self._settings.ragcore_space_id],
+                    },
+                )
+                if response.status_code != 200:
+                    KNOWLEDGE_PROVIDER_REQUESTS.labels("answers", "non_2xx").inc()
+                    self._open_answers_circuit()
+                    return None
+                body = response.json()
+        except (httpx.HTTPError, ValueError):
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("answers", "error").inc()
+            self._open_answers_circuit()
+            return None
 
         try:
-            sources = [SourceCard(**s) for s in body.get("sources", [])]
+            citations = {c["id"]: c for c in body.get("citations", [])}
-            evidence_state = body.get("evidence_state", "insufficient")
+            sources = [
-            if evidence_state not in ("grounded", "insufficient", "unavailable"):
+                SourceCard(
-                evidence_state = "insufficient"
+                    document_id=str(citation["document_id"]),
+                    title=citation["title"],
+                    version=str(citation["document_version_id"]),
+                    section=citation.get("section") or "",
+                    excerpt=citation["excerpt"],
+                )
+                for citation in citations.values()
+            ]
+            sources = _deduplicate_sources(sources)
+            answerability = body.get("answerability", "not_answerable")
+            is_grounded = answerability in _GROUNDED_ANSWERABILITY and sources
+            evidence_state: EvidenceState = "grounded" if is_grounded else "insufficient"
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("answers", evidence_state).inc()
+            self._close_answers_circuit()
             return GroundedAnswer(
-                answer=body.get("answer", ""),
+                answer=body.get("answer", "") if evidence_state == "grounded" else "",
                 evidence_state=evidence_state,
+                sources=sources if evidence_state == "grounded" else [],
+                provider=self.name,
+                correlation_id=correlation_id,
+            )
+        except (TypeError, KeyError, ValueError):
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("answers", "malformed").inc()
+            self._open_answers_circuit()
+            return None
+
+    def _ask_via_search_fallback(
+        self, question: str, correlation_id: str, language: str
+    ) -> GroundedAnswer:
+        unavailable = GroundedAnswer(
+            answer="",
+            evidence_state="unavailable",
+            sources=[],
+            provider=self.name,
+            correlation_id=correlation_id,
+        )
+        try:
+            with self._client() as client:
+                response = client.post(
+                    "/v1/search",
+                    json={
+                        "query": question,
+                        "requested_space_ids": [self._settings.ragcore_space_id],
+                        "max_results": 5,
+                    },
+                )
+                if response.status_code != 200:
+                    KNOWLEDGE_PROVIDER_REQUESTS.labels("search", "non_2xx").inc()
+                    return unavailable
+                body = response.json()
+        except (httpx.HTTPError, ValueError):
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("search", "error").inc()
+            return unavailable
+
+        try:
+            results = body.get("results", [])
+            if not isinstance(results, list):
+                raise TypeError("results must be a list")
+            sources = [
+                SourceCard(
+                    document_id=str(result["citation"]["document_id"]),
+                    title=result["citation"]["title"],
+                    version=str(result["citation"]["document_version_id"]),
+                    section=result["citation"].get("section") or "",
+                    excerpt=result["citation"]["excerpt"],
+                )
+                for result in results
+            ]
+        except (TypeError, KeyError, ValueError):
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("search", "malformed").inc()
+            return unavailable
+
+        for result in results:
+            score = _retrieval_score(result)
+            if score is not None:
+                KNOWLEDGE_RETRIEVAL_SCORE.observe(score)
+
+        all_sources = _deduplicate_sources(sources)
+        concepts = _question_concepts(question)
+        qualified_sources = [
+            SourceCard(
+                document_id=str(result["citation"]["document_id"]),
+                title=result["citation"]["title"],
+                version=str(result["citation"]["document_version_id"]),
+                section=result["citation"].get("section") or "",
+                excerpt=result["citation"]["excerpt"],
+            )
+            for result in results
+            if (_retrieval_score(result) or 0.0) >= self._settings.ragcore_min_search_score
+        ]
+        sources = _rank_sources_for_concepts(_deduplicate_sources(qualified_sources), concepts)
+        if not all_sources:
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("search", "insufficient").inc()
+            return GroundedAnswer(
+                answer="",
+                evidence_state="insufficient",
+                sources=[],
+                provider=self.name,
+                correlation_id=correlation_id,
+            )
+
+        damage_evidence = any(
+            term
+            in (
+                f"{source.document_id} {source.title} {source.section} {source.excerpt}"
+            ).casefold()
+            for source in sources
+            for term in _DOMAIN_CONCEPTS["damage"]
+        )
+        if not concepts or not sources or ("damage" in concepts and not damage_evidence):
+            KNOWLEDGE_PROVIDER_REQUESTS.labels("search", "insufficient").inc()
+            return GroundedAnswer(
+                answer="",
+                evidence_state="insufficient",
+                sources=all_sources,
+                provider=self.name,
+                correlation_id=correlation_id,
+            )
+
+        template = _LEAD_ANSWER_TEMPLATE.get(language, _LEAD_ANSWER_TEMPLATE[_DEFAULT_LANGUAGE])
+        lead = sources[0]
+        answer = template.format(title=lead.title, excerpt=lead.excerpt)
+        KNOWLEDGE_PROVIDER_REQUESTS.labels("search", "grounded").inc()
+        return GroundedAnswer(
+            answer=answer,
+            evidence_state="grounded",
             sources=sources,
             provider=self.name,
             correlation_id=correlation_id,
         )
-        except (TypeError, ValueError):
-            return GroundedAnswer(
-                answer="",
-                evidence_state="unavailable",
-                sources=[],
-                provider=self.name,
-                correlation_id=correlation_id,
-            )

backend/app/services/returns.py

@@ -1,8 +1,10 @@
 from __future__ import annotations
 
+import hashlib
+import json
 import uuid
 from dataclasses import dataclass
-from datetime import UTC, datetime
+from datetime import UTC, datetime, timedelta
 
 from sqlalchemy import select
 from sqlalchemy.exc import IntegrityError
@@ -18,29 +20,53 @@ from app.models.vehicle import Vehicle
 from app.schemas import CurrentUser, RegisterReturnRequest
 from app.services.audit import record_audit_event
 
-REF_PREFIX = "INSP"
 
+def _new_inspection_ref() -> str:
+    """Generate a collision-resistant public reference without reading mutable counts.
 
-def _next_public_ref(db: Session) -> str:
+    Return commands for different bookings can commit concurrently. A count-based
-    existing = db.execute(select(Inspection.public_ref)).scalars().all()
+    reference made those independent transactions race for the same unique value.
-    return f"{REF_PREFIX}-{len(existing) + 1:04d}"
+    """
+    return f"INSP-{uuid.uuid4().hex[:10].upper()}"
 
 
 def _derive_vehicle_status_with_reason(
     body: RegisterReturnRequest, vehicle: Vehicle, new_odometer: int
-) -> tuple[str, str]:
+) -> tuple[str, str, dict[str, str | int]]:
+    # Stable, localizable codes + params -- the backend never emits prose here. The
+    # frontend renders review.reasonCodes.<code> in the selected locale; the mirrored
+    # raw-English fallback strings live only in status_reason (shown under "Technical
+    # details") for backward compatibility. See docs/fleet-ops-correction/i18n-inventory.md.
     if body.damage_reported and body.technical_warning:
-        return "blocked", "Damage and a technical warning were both reported on return."
+        return (
+            "blocked",
+            "returnBlockedDamageAndTechnical",
+            {},
+        )
     if body.damage_reported:
-        return "blocked", "Damage was reported on return."
+        return "blocked", "returnBlockedDamage", {}
     if body.technical_warning:
-        return "blocked", "A technical warning was reported on return."
+        return "blocked", "returnBlockedTechnicalWarning", {}
     if new_odometer >= vehicle.next_service_km:
         return (
             "maintenance",
-            f"Odometer reached the {vehicle.next_service_km:,} km service threshold.",
+            "returnServiceThresholdReached",
+            {"threshold_km": vehicle.next_service_km},
         )
-    return "cleaning", "No damage, technical warning or service threshold; routed to cleaning."
+    return "cleaning", "returnRoutedToCleaning", {}
+
+
+_STATUS_REASON_FALLBACK_TEXT: dict[str, str] = {
+    "returnBlockedDamageAndTechnical": (
+        "Damage and a technical warning were both reported on return."
+    ),
+    "returnBlockedDamage": "Damage was reported on return.",
+    "returnBlockedTechnicalWarning": "A technical warning was reported on return.",
+    "returnServiceThresholdReached": "Odometer reached the service threshold.",
+    "returnRoutedToCleaning": (
+        "No damage, technical warning or service threshold; routed to cleaning."
+    ),
+}
 
 
 @dataclass
@@ -51,6 +77,8 @@ class ReturnEvaluation:
     resulting_odometer_km: int
     resulting_vehicle_status: str
     status_reason: str
+    status_reason_code: str
+    status_reason_params: dict[str, str | int]
     would_create_quality_issue: bool
     attention_reasons: list[str]
     next_booking_risk: dict | None
@@ -64,9 +92,10 @@ def evaluate_return(
     preview and commit can never drift apart."""
     odometer_regression = body.end_odometer_km < vehicle.odometer_km
     resulting_odometer_km = vehicle.odometer_km if odometer_regression else body.end_odometer_km
-    resulting_status, status_reason = _derive_vehicle_status_with_reason(
+    resulting_status, status_reason_code, status_reason_params = _derive_vehicle_status_with_reason(
         body, vehicle, resulting_odometer_km
     )
+    status_reason = _STATUS_REASON_FALLBACK_TEXT[status_reason_code]
 
     attention_reasons = []
     if body.damage_reported:
@@ -101,6 +130,8 @@ def evaluate_return(
         resulting_odometer_km=resulting_odometer_km,
         resulting_vehicle_status=resulting_status,
         status_reason=status_reason,
+        status_reason_code=status_reason_code,
+        status_reason_params=status_reason_params,
         would_create_quality_issue=odometer_regression,
         attention_reasons=attention_reasons,
         next_booking_risk=next_booking_risk,
@@ -142,17 +173,17 @@ def preview_vehicle_return(
     return booking, vehicle, evaluation
 
 
-def register_vehicle_return(
+def request_fingerprint(body: RegisterReturnRequest) -> str:
+    canonical = json.dumps(body.model_dump(mode="json"), sort_keys=True, separators=(",", ":"))
+    return hashlib.sha256(canonical.encode("utf-8")).hexdigest()
+
+
+def _replay_or_reject(
     db: Session,
+    existing: IdempotencyRecord,
     booking_ref: str,
-    body: RegisterReturnRequest,
+    fingerprint: str,
-    idempotency_key: str,
-    actor: CurrentUser,
 ) -> tuple[int, dict]:
-    existing = db.scalar(
-        select(IdempotencyRecord).where(IdempotencyRecord.idempotency_key == idempotency_key)
-    )
-    if existing is not None:
     booking = db.get(Booking, existing.booking_id)
     if booking is None or booking.public_ref != booking_ref:
         raise AppError(
@@ -160,8 +191,29 @@ def register_vehicle_return(
             "This idempotency key was already used for a different booking.",
             status_code=409,
         )
+    if existing.request_fingerprint is not None and existing.request_fingerprint != fingerprint:
+        raise AppError(
+            "IDEMPOTENCY_KEY_REUSED",
+            "This idempotency key was already used with a different request body.",
+            status_code=409,
+        )
     return existing.response_status, existing.response_body
 
+
+def register_vehicle_return(
+    db: Session,
+    booking_ref: str,
+    body: RegisterReturnRequest,
+    idempotency_key: str,
+    actor: CurrentUser,
+) -> tuple[int, dict]:
+    fingerprint = request_fingerprint(body)
+    existing = db.scalar(
+        select(IdempotencyRecord).where(IdempotencyRecord.idempotency_key == idempotency_key)
+    )
+    if existing is not None:
+        return _replay_or_reject(db, existing, booking_ref, fingerprint)
+
     booking, vehicle = _load_active_booking_and_vehicle(db, booking_ref, lock=True)
 
     # Re-check after acquiring the row lock: a concurrent identical-key request may have
@@ -170,7 +222,7 @@ def register_vehicle_return(
         select(IdempotencyRecord).where(IdempotencyRecord.idempotency_key == idempotency_key)
     )
     if existing is not None:
-        return existing.response_status, existing.response_body
+        return _replay_or_reject(db, existing, booking_ref, fingerprint)
 
     if booking.status != "active":
         raise AppError(
@@ -184,7 +236,7 @@ def register_vehicle_return(
     evaluation = evaluate_return(db, booking, vehicle, body, now=now)
 
     inspection = Inspection(
-        public_ref=_next_public_ref(db),
+        public_ref=_new_inspection_ref(),
         booking_id=booking.id,
         vehicle_id=vehicle.id,
         type="return",
@@ -226,6 +278,7 @@ def register_vehicle_return(
             },
             proposed_action_json={},
             detected_at=now,
+            due_at=now + timedelta(days=1),
         )
         db.add(issue)
         db.flush()
@@ -298,6 +351,7 @@ def register_vehicle_return(
         "odometer_regression": evaluation.odometer_regression,
         "quality_issue_ref": quality_issue_ref,
         "workflow_event_id": str(event.event_id),
+        "correlation_id": str(correlation_id),
         "next_booking_risk": evaluation.next_booking_risk,
     }
 
@@ -305,6 +359,7 @@ def register_vehicle_return(
         IdempotencyRecord(
             idempotency_key=idempotency_key,
             booking_id=booking.id,
+            request_fingerprint=fingerprint,
             response_status=201,
             response_body=response_body,
         )
@@ -318,7 +373,7 @@ def register_vehicle_return(
             select(IdempotencyRecord).where(IdempotencyRecord.idempotency_key == idempotency_key)
         )
         if existing is not None:
-            return existing.response_status, existing.response_body
+            return _replay_or_reject(db, existing, booking_ref, fingerprint)
         raise
 
     return 201, response_body

backend/app/services/sessions.py

@@ -0,0 +1,34 @@
+from __future__ import annotations
+
+from datetime import UTC, datetime
+
+from sqlalchemy import delete, select
+from sqlalchemy.orm import Session
+
+from app.core.config import get_settings
+from app.core.security import SessionPayload, session_token_hash
+from app.models.revoked_session import RevokedSession
+
+settings = get_settings()
+
+
+def is_session_revoked(db: Session, token: str) -> bool:
+    token_hash = session_token_hash(token)
+    revoked_id = db.scalar(select(RevokedSession.id).where(RevokedSession.token_hash == token_hash))
+    return revoked_id is not None
+
+
+def revoke_session(db: Session, token: str, payload: SessionPayload) -> None:
+    db.execute(delete(RevokedSession).where(RevokedSession.expires_at < datetime.now(UTC)))
+    token_hash = session_token_hash(token)
+    existing = db.scalar(select(RevokedSession.id).where(RevokedSession.token_hash == token_hash))
+    if existing is not None:
+        return
+    db.add(
+        RevokedSession(
+            token_hash=token_hash,
+            expires_at=datetime.fromtimestamp(
+                payload.issued_at + settings.session_ttl_seconds, UTC
+            ),
+        )
+    )