diff --git a/.github/workflows/e2e.yml b/.github/workflows/e2e.yml
index 98d60e77..c77482ca 100644
--- a/.github/workflows/e2e.yml
+++ b/.github/workflows/e2e.yml
@@ -111,9 +111,38 @@ jobs:
 
       - name: Start server
         run: |
-          npx serve out -l 3000 &
-          sleep 5
-          npx wait-on http://localhost:3000 --timeout 60000
+          set -euo pipefail
+          npx serve out -l 3000 > serve.log 2>&1 &
+          SERVE_PID=$!
+          echo "serve started, PID=$SERVE_PID"
+
+          # Brief grace period for the child to spawn; wait-on does the real polling.
+          sleep 2
+
+          # Verify serve didn't already exit (port in use, missing out/, etc.).
+          if ! kill -0 "$SERVE_PID" 2>/dev/null; then
+            echo "::error::serve exited immediately after launch"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            exit 1
+          fi
+
+          # Explicit exit-code propagation. Don't rely on -e interacting cleanly
+          # with backgrounded jobs.
+          if ! npx wait-on http://localhost:3000 --timeout 60000 --verbose; then
+            echo "::error::wait-on timed out — serve never bound to port 3000 within 60s"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            echo "--- listening sockets ---"
+            ss -tlnp 2>/dev/null || netstat -tlnp 2>/dev/null || true
+            echo "--- serve process state ---"
+            ps -fp "$SERVE_PID" 2>/dev/null || echo "serve process $SERVE_PID is gone"
+            echo "--- direct curl attempt ---"
+            curl -v --max-time 5 http://localhost:3000/ 2>&1 || true
+            exit 1
+          fi
+
+          echo "serve is responding on http://localhost:3000"
 
       - name: Run smoke tests (sign-up only - auth tests run after auth-setup)
         run: |
@@ -183,9 +212,38 @@ jobs:
 
       - name: Start server
         run: |
-          npx serve out -l 3000 &
-          sleep 5
-          npx wait-on http://localhost:3000 --timeout 60000
+          set -euo pipefail
+          npx serve out -l 3000 > serve.log 2>&1 &
+          SERVE_PID=$!
+          echo "serve started, PID=$SERVE_PID"
+
+          # Brief grace period for the child to spawn; wait-on does the real polling.
+          sleep 2
+
+          # Verify serve didn't already exit (port in use, missing out/, etc.).
+          if ! kill -0 "$SERVE_PID" 2>/dev/null; then
+            echo "::error::serve exited immediately after launch"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            exit 1
+          fi
+
+          # Explicit exit-code propagation. Don't rely on -e interacting cleanly
+          # with backgrounded jobs.
+          if ! npx wait-on http://localhost:3000 --timeout 60000 --verbose; then
+            echo "::error::wait-on timed out — serve never bound to port 3000 within 60s"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            echo "--- listening sockets ---"
+            ss -tlnp 2>/dev/null || netstat -tlnp 2>/dev/null || true
+            echo "--- serve process state ---"
+            ps -fp "$SERVE_PID" 2>/dev/null || echo "serve process $SERVE_PID is gone"
+            echo "--- direct curl attempt ---"
+            curl -v --max-time 5 http://localhost:3000/ 2>&1 || true
+            exit 1
+          fi
+
+          echo "serve is responding on http://localhost:3000"
 
       - name: Run rate-limiting tests (ordered)
         run: pnpm test:e2e --project=rate-limiting --project=brute-force --project=signup --reporter=list --trace=on-first-retry
@@ -245,9 +303,38 @@ jobs:
 
       - name: Start server
         run: |
-          npx serve out -l 3000 &
-          sleep 5
-          npx wait-on http://localhost:3000 --timeout 60000
+          set -euo pipefail
+          npx serve out -l 3000 > serve.log 2>&1 &
+          SERVE_PID=$!
+          echo "serve started, PID=$SERVE_PID"
+
+          # Brief grace period for the child to spawn; wait-on does the real polling.
+          sleep 2
+
+          # Verify serve didn't already exit (port in use, missing out/, etc.).
+          if ! kill -0 "$SERVE_PID" 2>/dev/null; then
+            echo "::error::serve exited immediately after launch"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            exit 1
+          fi
+
+          # Explicit exit-code propagation. Don't rely on -e interacting cleanly
+          # with backgrounded jobs.
+          if ! npx wait-on http://localhost:3000 --timeout 60000 --verbose; then
+            echo "::error::wait-on timed out — serve never bound to port 3000 within 60s"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            echo "--- listening sockets ---"
+            ss -tlnp 2>/dev/null || netstat -tlnp 2>/dev/null || true
+            echo "--- serve process state ---"
+            ps -fp "$SERVE_PID" 2>/dev/null || echo "serve process $SERVE_PID is gone"
+            echo "--- direct curl attempt ---"
+            curl -v --max-time 5 http://localhost:3000/ 2>&1 || true
+            exit 1
+          fi
+
+          echo "serve is responding on http://localhost:3000"
 
       - name: Run auth setup
         run: pnpm exec playwright test --project=setup --reporter=list --timeout=180000
@@ -352,9 +439,38 @@ jobs:
 
       - name: Start server
         run: |
-          npx serve out -l 3000 &
-          sleep 5
-          npx wait-on http://localhost:3000 --timeout 60000
+          set -euo pipefail
+          npx serve out -l 3000 > serve.log 2>&1 &
+          SERVE_PID=$!
+          echo "serve started, PID=$SERVE_PID"
+
+          # Brief grace period for the child to spawn; wait-on does the real polling.
+          sleep 2
+
+          # Verify serve didn't already exit (port in use, missing out/, etc.).
+          if ! kill -0 "$SERVE_PID" 2>/dev/null; then
+            echo "::error::serve exited immediately after launch"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            exit 1
+          fi
+
+          # Explicit exit-code propagation. Don't rely on -e interacting cleanly
+          # with backgrounded jobs.
+          if ! npx wait-on http://localhost:3000 --timeout 60000 --verbose; then
+            echo "::error::wait-on timed out — serve never bound to port 3000 within 60s"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            echo "--- listening sockets ---"
+            ss -tlnp 2>/dev/null || netstat -tlnp 2>/dev/null || true
+            echo "--- serve process state ---"
+            ps -fp "$SERVE_PID" 2>/dev/null || echo "serve process $SERVE_PID is gone"
+            echo "--- direct curl attempt ---"
+            curl -v --max-time 5 http://localhost:3000/ 2>&1 || true
+            exit 1
+          fi
+
+          echo "serve is responding on http://localhost:3000"
         env:
           CI: true
 
@@ -482,9 +598,38 @@ jobs:
           path: tests/e2e/fixtures/
       - name: Start server
         run: |
-          npx serve out -l 3000 &
-          sleep 5
-          npx wait-on http://localhost:3000 --timeout 60000
+          set -euo pipefail
+          npx serve out -l 3000 > serve.log 2>&1 &
+          SERVE_PID=$!
+          echo "serve started, PID=$SERVE_PID"
+
+          # Brief grace period for the child to spawn; wait-on does the real polling.
+          sleep 2
+
+          # Verify serve didn't already exit (port in use, missing out/, etc.).
+          if ! kill -0 "$SERVE_PID" 2>/dev/null; then
+            echo "::error::serve exited immediately after launch"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            exit 1
+          fi
+
+          # Explicit exit-code propagation. Don't rely on -e interacting cleanly
+          # with backgrounded jobs.
+          if ! npx wait-on http://localhost:3000 --timeout 60000 --verbose; then
+            echo "::error::wait-on timed out — serve never bound to port 3000 within 60s"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            echo "--- listening sockets ---"
+            ss -tlnp 2>/dev/null || netstat -tlnp 2>/dev/null || true
+            echo "--- serve process state ---"
+            ps -fp "$SERVE_PID" 2>/dev/null || echo "serve process $SERVE_PID is gone"
+            echo "--- direct curl attempt ---"
+            curl -v --max-time 5 http://localhost:3000/ 2>&1 || true
+            exit 1
+          fi
+
+          echo "serve is responding on http://localhost:3000"
         env:
           CI: true
       - name: Prime Supabase connection pool
@@ -596,9 +741,38 @@ jobs:
           path: tests/e2e/fixtures/
       - name: Start server
         run: |
-          npx serve out -l 3000 &
-          sleep 5
-          npx wait-on http://localhost:3000 --timeout 60000
+          set -euo pipefail
+          npx serve out -l 3000 > serve.log 2>&1 &
+          SERVE_PID=$!
+          echo "serve started, PID=$SERVE_PID"
+
+          # Brief grace period for the child to spawn; wait-on does the real polling.
+          sleep 2
+
+          # Verify serve didn't already exit (port in use, missing out/, etc.).
+          if ! kill -0 "$SERVE_PID" 2>/dev/null; then
+            echo "::error::serve exited immediately after launch"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            exit 1
+          fi
+
+          # Explicit exit-code propagation. Don't rely on -e interacting cleanly
+          # with backgrounded jobs.
+          if ! npx wait-on http://localhost:3000 --timeout 60000 --verbose; then
+            echo "::error::wait-on timed out — serve never bound to port 3000 within 60s"
+            echo "--- serve.log ---"
+            cat serve.log || true
+            echo "--- listening sockets ---"
+            ss -tlnp 2>/dev/null || netstat -tlnp 2>/dev/null || true
+            echo "--- serve process state ---"
+            ps -fp "$SERVE_PID" 2>/dev/null || echo "serve process $SERVE_PID is gone"
+            echo "--- direct curl attempt ---"
+            curl -v --max-time 5 http://localhost:3000/ 2>&1 || true
+            exit 1
+          fi
+
+          echo "serve is responding on http://localhost:3000"
         env:
           CI: true
       - name: Prime Supabase connection pool