The method of Ali et al (2016, Genetics) deemed Rapture has two parts to it - first the new RADseq library prep and then the capture. Here, I am talking about sequencing RADseq libraries without the capture part.
As documented elsewhere on this blog, you have to use paired-end sequencing to get as many reads as possible from the new RADseq method because half the reads will have the barcoded adapter ligated to one end and the other half will have it ligated to the other during the sequencing.
Once the raw data is generated, you have to trim and organize the output (with the trim_flip perl script shown elsewhere on this blog) such that all the restriction enzyme sites align on the left and are put into one file and all the other reads are put into a separate file ("reverse") - which will still be the paired end in order with the "forward" file.
The program Stacks can only use the "forward" reads with all the restriction sites aligned to the left and all the same length. But, if you are mapping to a reference first you can use both the forward and reverse reads to map and then use that for input into the ref_map pipeline of Stacks. but the trim/organize step is "necessary for any de-multiplexing program and probably makes the aligned data easier to deal with, so all the read pairs run in the same direction." (Hohenlohe, pers comm).