diff --git a/Dockerfile b/Dockerfile
index 9ce1a488..9ffd5eef 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -67,7 +67,8 @@ EOF
 EXPOSE 8000
 
 # Health check
-HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \
+# start-period must outlast the startup script's 60s database wait.
+HEALTHCHECK --interval=30s --timeout=3s --start-period=75s --retries=3 \
     CMD curl -f http://localhost:8000/up || exit 1
 
 # Create startup script for production
@@ -77,32 +78,44 @@ set -e
 
 # Wait for database to be ready
 echo "Waiting for database..."
+db_ready=0
 for i in $(seq 1 30); do
     if mysqladmin ping -h "$DB_HOST" -u "$DB_USERNAME" -p"$DB_PASSWORD" --silent 2>/dev/null; then
         echo "Database is ready!"
+        db_ready=1
         break
     fi
     echo "Waiting for database... ($i/30)"
     sleep 2
 done
 
+if [ "$db_ready" -ne 1 ]; then
+    echo "Database unreachable at ${DB_HOST} after 60s; exiting so the restart policy retries." >&2
+    exit 1
+fi
+
 # Cache config with runtime env vars
 echo "Caching configuration..."
 php artisan config:cache
 
-# Run migrations
+# Run migrations. A partial schema is worse than a container that refuses to start.
 echo "Running migrations..."
-php artisan migrate --force || echo "Migrations failed or already up-to-date"
+php artisan migrate --force
 
 # Run seeders (idempotent - uses updateOrInsert)
 echo "Running seeders..."
-php artisan db:seed --force || echo "Seeders failed or already run"
+php artisan db:seed --force
 
-# Start Horizon in the background
-php artisan horizon &
+# Restart a worker that dies instead of leaving the queue unconsumed.
+supervise() {
+    while true; do
+        "$@" || echo "[supervisor] '$*' exited with $?; restarting in 5s." >&2
+        sleep 5
+    done
+}
 
-# Start the scheduler in the background
-php artisan schedule:work &
+supervise php artisan horizon &
+supervise php artisan schedule:work &
 
 # Start FrankenPHP
 exec frankenphp run --config /etc/caddy/Caddyfile