Conflicting variants in results.json file and collated variants

[KWollenbergPhylo]

I'm using TB Profiler v6.3 in a conda environment on our HPC to generate a collated variants matrix. One of the samples is being reported as having two different nonsynonymous substitutions at the same codon, due to variant calls at the second and third positions. Both variants are reported with 1.00 frequency in the results.json file. Looking into the vcf files I see one variant reported in the targets.csq.vcf but both reported in the targets.vcf. Looking in the vcf generated by pilon for this sample I see the 3rd position has a low frequency variant (2 calls out of 11, so a low coverage site as well). Somehow this low-frequency variant is being elevated to a "frequency = 1.00" variant in the results.json file. Is there perhaps an inconsistency in how variant calls are evaluated in the targets.vcf file (which then propagates to the results.json file)?