We aim to solve this problem for good in PLINK 2.0. The file format explicitly defines reference vs. alternate alleles, and this information will be preserved across runs by default. In addition, the file format will include a flag distinguishing provisional reference allele assignments from those derived from an actual reference genome. When PLINK 2.0 operates on.ped/.map or similar data lacking a reference vs. alternate distinction, it will treat a highest-frequency allele as the reference, while flagging it as a provisional assignment. When a file with flagged-as-provisional reference alleles is merged with another file with unflagged reference alleles, the unflagged reference allele assignments take precedence. (Merges involving conflicting unflagged reference alleles will fail unless the user specifies which source file takes precedence.) It will also be straightforward to import real reference allele assignments with an analogue of –a2-allele.
