Mismatch between Hail and GATK/Picard counting?

Hi everyone, I have a whole-exome sequencing dataset that is processed by GATK, before importing into Hail. After restricting to exome intervals and sample_qc, the average n_snp for each sample is roughly 35,000.

However, from the metrics obtained from the VCF using this tool (https://gatk.broadinstitute.org/hc/en-us/articles/360037057132-CollectVariantCallingMetrics-Picard-), the expected coding variants are in the range of 20,000 - 25,000.

I have already ruled out the contribution of splitting multi-allelic sites. Would the devs have any comments on the likely reason for the discrepancy? Thank you!

Could you point me to picard documentation where they define the semantic meaning of this field?

In Hail, the n_snp produced by sample_qc is defined as “the number of SNP alternate alleles” per sample. This means a 1/1 genotype for a SNP alternate allele counts as 2.