Nice project! How does it compare to DuckDB on messy world data and complex joins?