Maybe I'm simply misunderstanding how library generation works. The inserted fragment should be size selected, my assumption was that the insert size was the expected size of the read? So if I have a 100bp read that means I'm sequencing the same 100bp fragment but from both directions.

Or is it more like one read is an anchor for the other which is further downstream or upstream of that insert? Eg., say the insert is 1kb, and I'm paired end sequencing 100bp, that would mean each paired read is either ~900bp upstream or downstream of the other read?