In the field of statistics, the term “sample size” universally means “number of samples”. See here, or here, for example.

I find this convention unfortunate, and I believe that “number of samples” is a clearer term. But, if the term “sample size” is used, IMHO it should only be used in the standard sense. It’s really confusing when two benchmarking crates for Rust each have methods called sample_size, but these methods control reciprocal quantities! Here is the documentation of the criterion method.

I suggest replacing the term “sample size” in divan by something like “runs per sample”, or “iterations per sample”, or perhaps “batch size”. Obviously this change could occur in a gradual way, by initially keeping the old name, but deprecating it.

This could be partially related to #58.