diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index d37b71c..fe3f2e3 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -351,8 +351,20 @@ jobs:
         run: pnpm --filter @capakraken/web exec playwright install-deps chromium
 
       - name: Push DB schema & seed
+        env:
+          PGPASSWORD: capakraken_test
         run: |
-          pnpm --filter @capakraken/db exec prisma db push --schema ./prisma/schema.prisma --force-reset --accept-data-loss --skip-generate
+          # Nuke any leftover schema state from a previous job that shared the
+          # postgres service container (act_runner reuses service volumes).
+          # --force-reset alone proved unreliable here: push reported "in sync"
+          # but audit_logs ended up missing, suggesting a split-brain between
+          # the reset and the subsequent DDL.
+          psql -h postgres -U capakraken -d capakraken_test -v ON_ERROR_STOP=1 \
+            -c "DROP SCHEMA IF EXISTS public CASCADE; CREATE SCHEMA public; GRANT ALL ON SCHEMA public TO capakraken; GRANT ALL ON SCHEMA public TO public;"
+          pnpm --filter @capakraken/db exec prisma db push --schema ./prisma/schema.prisma --accept-data-loss --skip-generate
+          # Sanity check: confirm the schema actually landed before seeding.
+          psql -h postgres -U capakraken -d capakraken_test -v ON_ERROR_STOP=1 \
+            -c "SELECT to_regclass('public.audit_logs') AS audit_logs, to_regclass('public.users') AS users;"
           pnpm db:seed
 
       - name: Run E2E tests
@@ -387,6 +399,13 @@ jobs:
           PGADMIN_PASSWORD=ci-pgadmin
           EOF
 
+      - name: Tear down any stale stack & volumes
+        # act_runner on self-hosted QNAP keeps named compose volumes between
+        # runs. A previous run's failed migration entry in _prisma_migrations
+        # causes P3009 on the next migrate deploy; wipe volumes for a truly
+        # fresh deploy test every time.
+        run: docker compose -f docker-compose.yml -f docker-compose.ci.yml down -v --remove-orphans || true
+
       - name: Start infrastructure (postgres + redis)
         run: docker compose -f docker-compose.yml -f docker-compose.ci.yml up -d postgres redis