For some studies this works really well - clinical trials, for example. But a lot of studies don't fit into a clean "Single experiment to measure a single outcome" paradigm, and pre-registration becomes somewhere between difficult and impossible.