Mathematically, one can either count the number of cells with hits, or count the number of hits. Either one allows valid statistics.

It just so happens that counting the number of cells with hits can more reliably reveal whether a method works or not, because we all know earthquakes are clustered in time because of aftershocks and triggered earthquakes, although just why would take a while to explain to people not versed in numbers of independent variables and variance reduction.

So sure, one could count multiple hits in a different scoring scheme, and then know LESS well whether one's predictions have any value, perhaps in line with your perennially making predictions and never applying statistical analysis to see whether they have validity.

This is why Roger (and most other quantitatively qualified students of earthquakes) decluster their catalogs. It's because they want to perform the most telling test.