If sd.sites_alleles[:] are byte strings, as occurs e.g. when using sgkit.simulate_genotype_call_dataset(), then when performing inference we get an error

>>> print(sd.sites_alleles[:])
[[b'T' b'A']
 [b'C' b'T']
 [b'A' b'G']
 [b'C' b'G']
 [b'C' b'T']
 [b'G' b'C']
 [b'A' b'G']
 [b'T' b'A']]

tsinfer.infer(sd)

... 
 1886     derived_state[mutation_id] = site.reorder_alleles().index(allele)
   1887     mutation_id += 1
   1888 site_id += 1

ValueError: tuple.index(x): x not in tuple

Because allele here has been translated to a normal string, e.g. "T", whereas the site.reorder_alleles() method returns a tuple of byte strings ((b'C', b'T'))