Quis custodiet ipsos custodes? Common ground between testing and data science

How do data scientists formulate the hypothesis that they want to test? What does overfitting has in common with the hard quality control rules? What is critical rationalism and why we should remember about it when working with real-world data?