I feel like in practice, in data science, you can rarely justify things theoretically and mostly rely on best practices for certainty of conclusions.