For me, at least, this has been a problem -- there is a bug in haplotypecaller when setting ploidy to 1. I think this must affect the data in that paper, too, since they say:
"We called variants for each parent using HaplotypeCaller and then merged using GenotypeGVCFs with default parameters except for sample ploidy 1"
https://github.com/broadinstitute/gatk/issues/8943
Basically, in Haplotypecaller, when ploidy is set to 1, using intervals of different length causes DP to be reported inconsistently at sites where there is no alt call. It was only accurate, for me, when I set the interval to single base pair resolution.
If the goal is confident genotypes, then the GATK folk claim this isn't really a problem. But, when you do care about the coverage over a given coordinate, it is of course.
The advice of GATK, and what makes sense to me, is that the variantcaller/genotyper isn't really the right tool for this. Better to use that to generate the high confident variant sites and then use a pileup.
Doing this is going to require that I re-do my current workflow, which is what has me wondering about the utility more generally. I'm curious if you have any thoughts