Phase 46: intake→completion e2e test (KI-016 O, demo spine) + CI Postgres job

.claude/memory.md

@@ -3,6 +3,16 @@
 Append discoveries as you find them. Read this file at session start. Keep
 entries short: date, what, why.
 
+## 2026-06-16 — Phase 46 KI-016 O: intake→completion e2e ("demo spine") (branch `feature/46-intake-completion-e2e`)
+
+- TEST/CI/DOCS only — zero product/runtime code changed; the spine was already correctly wired and now `__tests__/e2e/intake-to-completion.test.ts` proves it against a real DB. Gate green (lint/tc/prisma/2835 unit/build) + e2e 2 tests pass, idempotent across 3 reruns.
+- **Real status-transition map (correct one — prompt had two wrong):** intake creates VR `UNTRIAGED` → inline `autoTriageService.triageEnquiry` promotes a *complete non-urgent* request to **PLANNING_POOL** (not just TRIAGED). Route proposal (`routeProposalService.generateProposals(targetDate?, originOverride?)` — it's a service method, NOT a bare `generateProposals()`) moves the VR to **CLUSTERED** (the prompt said "PROPOSED" — wrong; there is no PROPOSED planningStatus on the VR). Booking then needs the RouteRun at **APPROVED** first — there's a vet gate `routeRunRepository.update(id,{status:'APPROVED'})` (DRAFT→APPROVED) BETWEEN proposal and `bookingService.bookRouteRun`. Booking: Appointment PROPOSED + VR→BOOKED + RouteRun→BOOKED. Confirmation flips Appointment PROPOSED→CONFIRMED + writes ConfirmationDispatch. Completion: VisitOutcome + Appointment→COMPLETED + VR→COMPLETED + (followUpRequired) a new FOLLOW_UP VR in PLANNING_POOL.
+- **To get a single fixture all the way to a PROPOSAL (the fiddly bit):** (a) customer must have exactly ONE yard → `yardMatcherService` returns high-confidence `sole-yard` so intake auto-fills `VisitRequest.yardId` (drives `needsMoreInfo:false` → PLANNING_POOL) regardless of message text; (b) the message must contain an explicit horse count (else `ASK_HORSE_COUNT` missing-field → stuck UNTRIAGED); (c) **horseCount must be ≥3** because `clustering.service` only keeps a single-stop cluster when `horseCount >= MIN_HORSES_FOR_SINGLE_STOP (3)` AND densityScore ≥ `minDensityScoreThreshold (20)` — a 1- or 2-horse lone yard yields NO proposal; (d) the yard needs lat/lng pre-set so the planner skips geocoding (offline-deterministic); (e) keep the e2e DB with **0 active VET staff** so `planPairing` collapses to the solo path (parallel path needs `activeVets.length>=2`).
+- **Urgent path:** an urgent keyword → `urgencyLevel=URGENT`, `planningStatus=READY_FOR_REVIEW`, and `emergency.service.raiseUrgentAlert` sets `automationHold=true`; the proposal query excludes `automationHold:true` AND `urgencyLevel:'URGENT'`, so it's held out. With no `VET_ALERT_*` env + no active vet, the alert just warns + dead-letters — never throws (intake completes).
+- **Real-DB test plumbing:** the prisma singleton has a soft-delete *read* extension (customer/yard/horse/enquiry) but it does NOT touch `deleteMany`/`create`, so cleanup hard-deletes fine. Pass `deletedAt: undefined` in cleanup `findMany` to also see tombstoned residue. Created rows (enquiry/VR/appt/routerun) get uuid ids — cleanup must walk relations from the known `e2e-` customer ids, not by id-prefix. Delete order: VisitOutcome (no cascade, FK line 724) before Appointment (which cascades dispatch/response/statusHistory/assignment). DEMO_MODE WhatsApp confirmation routes through the simulator → `success:true` so the CONFIRMED transition + dispatch row actually happen.
+- **Vitest split:** default `vitest.config.ts` needed an explicit `exclude` (it had none → relied on vitest defaults) listing the vitest defaults + `__tests__/e2e/**`; new `vitest.e2e.config.ts` includes ONLY e2e, no react plugin, own setup that does NOT stub DATABASE_URL (fails fast if RUN_DB_E2E set but URL is the unit stub). Suite gated `describe.skipIf(!RUN_DB_E2E)`. CI: new additive `e2e` job (needs: check) with a `postgres:16` service container + `prisma migrate deploy` then `npm run test:e2e`; `check`/`docker`/`security` untouched (`docker` still `needs: check` only).
+- New KI-031 (confirmation `shouldUseTemplate()` 24h-window is DEMO-only / `TODO(prod)`, Meta-gated, deferred). KI-016 O marked delivered.
+
 Format:
 
 ```

.github/workflows/ci.yml

@@ -14,6 +14,9 @@ on:
       - 'public/**'
       - 'i18n/**'
       - 'scripts/**'
+      - '__tests__/**'
+      - 'vitest.config.ts'
+      - 'vitest.e2e.config.ts'
       - 'auth.ts'
       - 'middleware.ts'
       - 'instrumentation.ts'
@@ -55,6 +58,54 @@ jobs:
       - run: npm run test -- --passWithNoTests
       - run: npm run build
 
+  # Phase 46 — KI-016 O: real-database end-to-end "demo spine" test.
+  # Isolated, additive job — it does NOT touch the `check` gate above.
+  # Runs the intake → triage → proposal → booking → confirmation →
+  # completion pipeline against a fresh Postgres 16 service container with
+  # the real (non-stub) DATABASE_URL, after applying migrations.
+  e2e:
+    if: github.event_name == 'pull_request' || github.event_name == 'push' || github.event_name == 'merge_group'
+    runs-on: ubuntu-latest
+    needs: check
+    services:
+      postgres:
+        image: postgres:16
+        # Throwaway credentials for an ephemeral CI service container (the
+        # same test:test convention the `check` job's stub DATABASE_URL uses)
+        # — not a secret; the container exists only for this job's lifetime.
+        env:
+          POSTGRES_USER: test
+          POSTGRES_PASSWORD: test
+          POSTGRES_DB: equismile_e2e
+        ports:
+          - 5432:5432
+        options: >-
+          --health-cmd "pg_isready -U test -d equismile_e2e"
+          --health-interval 5s
+          --health-timeout 5s
+          --health-retries 10
+    env:
+      # Real (connecting) service-container DB — unlike the `check` job's
+      # non-connecting stub. Throwaway test:test creds, ephemeral container.
+      DATABASE_URL: "postgresql://test:test@localhost:5432/equismile_e2e"
+      RUN_DB_E2E: "1"
+      DEMO_MODE: "true"
+      SKIP_ENV_VALIDATION: "true"
+      NEXT_PUBLIC_BUILD_SHA: ${{ github.sha }}
+      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+    steps:
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+        with:
+          node-version: 20
+          cache: npm
+      - run: npm ci
+      - run: npx prisma generate
+      - name: Apply migrations to e2e database
+        run: npx prisma migrate deploy
+      - name: Run intake → completion e2e
+        run: npm run test:e2e
+
   docker:
     if: github.event_name == 'pull_request' || github.event_name == 'push' || github.event_name == 'merge_group'
     runs-on: ubuntu-latest