Yes, let me clarify. The implementation of BIFT in RAW has positivity strictly enforced, because of how its programmed. That’s not a general constraint of the method, just our current implementation, so the version I linked I believe supports negative values. And you are correct in that negative values in the P(r) function are fine in some cases.
I’ve added the ability to turn off the BIFT positivity constraint to my list of requested features for RAW, but I can’t promise how soon I’ll be able to get to it.