also we have a pre-shaded version of Spark that eliminates class dependency issues with our own code, and helps to run directly from the IDE more quickly. One we deploy, we do the reverse and shade some of our things out of the way but that is a slower build process.