We have a few top-level output columns which are empty in the final CSV because the data is inconvenient to represent in a spreadsheet. For example, "potential_ref_transclusions" and "ref_error_counts_by_type" are maps of string and count, and very long.

It would be nice to represent this data in a way that's easy to explore. One possibility is a separate CSV with a row for each wiki and the columns are each one item from these lists? But that's not ideal. This task is done when we find a reasonable output format.