Citing @tpoterba from another topic above.
So some normalisation does actually magically happen during split_multi_hts? Does anyone know: is there some room for ambiguity, or should split_multi_hts produce the same allels as bcftools -norm m for any input collection of alleles?