Kneaddata removed rRNA reads accounting

I’m trying to figure out what the exact number of reads removed by kneaddata that was rRNA.
However, there’s a catch.
I need the actual reads that were removed by kneaddata.

I am using the bog-standard libraries that kneaddata comes with (Silva DB)
I see a few files, but the logs and labelling make it a bit confusing to understand the accounting.

Of the files that I have, I see there’s a <my_data>_<SILVA_plus a big tail>_unmatched_1_contam.fastq
Is this fastq a collection of reads that have had rRNA removed?
Or is this a fastq of rRNA reads?
Or is this a fastq of my reads before rRNA filtration has been applied?

I ask because I see reads from the contamination file inside the final result file: <my_data>_kneaddata_unmatched_1.fastq

It is unclear what my data should be.

I wanted to ask: Is there any supporting documentation on how kneaddata cleans the data?
I see in paired-read data, there are many rRNA sequences found inside the final output file.
(This was ascertained by comparing sequences in the contamination file and the final output)

I can verify that kneaddata works as-advertised (no rRNA seqs found) in single-end data.
What is kneaddata doing with paired-end data?
Should I be only using forward-end reads in a paired-data scenario?

does anyone know?? I don’t know what is wrong here.