It seems right now idea splits are just assigning people to ideas fairly randomly, which leads to high variation of the n-size for each. If we could have stricter least fill when assigning respondents, that'd be great (at least as an option)!