[P] Is there a way to defend using a subset of data for ablation studies?

I wanted to do an ablation study, but due to limited compute, I just compared a subset of the data of a very large