Begin with the question.
How can a data system work better under a concrete workload and resource budget?
We approach this field by making assumptions explicit, connecting methods to measurable behavior, and allowing experiments and systems to refine the question itself.