According to VCF documentation, QUAL values are phred-scored and indicate the confidence on the assertion made on the ALT allele, whether it is variant or invariant site. That way, QUAL>60 always indicate prob(no-variant) < 10e-6 in variant sites, and prob(variant) < 10e-6 in non-variant sites.

The conflicting point for me is that in my VCF files, the distribution of QUAL scores for variant and invariant sites is dramatically different, with QUAL always being lower than 1, and close to zero, for non-variant sites, thus indicating that p(variant) is high in that sites.

Is that a normal behavior? If p(variant) is that high, e.g. PHRED = 3.06061e-17, why is not the site called as variant?