Will min_non_overlap influence sensitivity of lumpy?

[xyw1]

Hi Ryan, I'm recently using lumpy on some data sets with known structural variants （3 known SVs totally） and I'm trying to optimize my parameters. I found that when I change the value of min_non_overlap （range from 0 to read length 100） , the sensitivity of lumpy varies (the blue line in the figure below). At some values of min_non_overlap(like 25,30,35,40), lumpy omitted some variants in the data sets. How does this happen? Or is that I just didn't run it correctly?

I read in the documentation that min_non_overlap is described as following

Number of base pair positions that must be unique to each end of a read pair. Some library preps are created with large reads and small library sizes such that read overlap, in all over cases overlapping reads tends to be a sign of an error. We typically set this to read length (pairs cannot overlap).

What does 'overlapping reads tends to be a sign of an error' mean？

Thanks a lot.