We are not working in Paratext, but with FMP (FilemakerPro) but I think the problems are the same and I think there is a bug in the interpretation of RTL chapter and verse numbers by SAB (I am using version SAB 6.2 MacOS) e.g. Gen. 11:12.
I enclose a screenshot where you can see the same string of cross-references at v2, v3 and v4. You have to read the string starting at the right side. They display differently. I will focus on the string in vs. 2.
In vs. 2 it reads as the first cross-reference Job 38:4 (so visually from RTL: 4:38 boJ).
So coming from the right (the reading direction) and going to the left - if you think consistently (starting to think in LTR then flipping it from there to RTL) you would say from what you see in vs. 2 (so visually) is chapter 38 and then verse 4 more to the left. But NO!
SAB takes this as chapter 4 and verse 38. If I change the order, again it switches verse and chapter and makes 38 the chapter and 4 the verse. I know what it takes for chapter when I click the cross-references and see it leads me to the chapter number which actually is the verse number at the left side (visually) of the colon,
There is a contrast between the display in SAB (source) and the function each element left and right from the colon should have. The number most to the right in the display should interpreted by SAB as the chapter number and the number left from the colon should be interpreted as the verse.
This happens also in the \xo part but I was able to force that the other way around starting from FMP (though of course this is the same sort of bug - see above CraigN), but I cannot do that for the cross-references themselves.
I have not heard of anybody coping with this same problem exporting to SFM from Paratext, but since nobody confirmed my question about this, I assume that when working with RTL cross-references and exporting from Paratext the same problem occurs.
The RTL cross-references are a long pending issue and I hope we can solve this together no matter whether input is from Paratext or FMP, because I guess the problem is the same.