How do you approach structuring an analysis when the initial data is incomplete, messy, or distributed across multiple incompatible systems at GitHub?
