As far as I can tell from the specification posts here, it's entirely to make implementation more feasible. Rather than some complicated calculation of preceden

As far as I can tell from the specification posts here, it's entirely to make implementation more feasible. Rather than some complicated calculation of precedence, you can just assign each rule a single score. As long as the weird exception essentially never comes up (which is the point of specifying "a number system with a large base"), it's an efficient and sensible way to implement the desired rule.