I assume (this is my assumption so please don't believe it blindly) these are bad reads that didn't pass the quality criteria and were flagged as bad reads. I don't think there should be many of such reads if any present in the xsq file. I have used lifescope quiet a bit and in the end it gives you a statistics table that tells you total reads in the xsq file and how many of these got filtered before the alignment and how many of those that were not filtered got aligned and all other information. I have rarely seen any numbers in the row that mentions the numbers of reads filtered before alignment.

I think I kinda figured out what filtering here does. Without the -f option, I would get .csfasta files with few reads like this:

>1_44_92_F3    
T..................................................

When I use -f, these kind of sequences/reads are not present in the csfasta file. But does removing these reads make any difference in the further analysis like mapping?