Two sounds differ phonetically in a predictable environment but never distinguish words there. Which evidence most strongly supports treating them as allophones rather than separate phonemes in that environment?