I'll try and refactor a sample. The real data-set is very large.
I took a look at the data on a row level, creating a table that had usr_email as the main dimension and recp_email, send_email as plain expressions. Then I added these scripts as expressions and got a right-looking count on row-level.
I also tried checking for count integrity using IsNull() and noted that NullCount did not seem to produce correct results. I have not named or otherwise changed null values in the data, they should be true nulls.
I did note, however, that Qlikview marks values in tabular data as null and seemed to evaluate multiple matches of emails into nulls in tabular context. So, if there were several emails matching send_email (several senders to a single recipient), tabular view would show these as null instead of pointing to this "array". Distinct count of these was, however, returned correct on row level upon closer inspection.
Mostly, I end up having too many distincts, not too few. But I'll try and refactor the loader to get a small obfuscated sample :-).