I have 4 pairs of samples for tumour and normal. I constructed a PoN with these 4 normal samples and ran each pair of sample sets individually, and each pair found a certain number of variants marked as 'PASS', about 100 SNV each. 
But I question the authenticity of these variants. 
So I deceptively swapped the identities of tumour and normal: i.e. build the PoN with 4 tumour samples, treat tumour as 'normal' samples, and normal as 'tumour' samples. Then run the entire pipeline again in exactly the same way with the fake pairs of 'tumour' and 'normal' samples. 
My deceptive test of swapping identities expects to get a variant count of 'pass' close to 0. Because my tumour samples should contain a small amount of normal tissue, that is, the tumour samples should contain almost all possible variants or alleles when constructing the PoN. When running Mutect2 with normal samples as 'tumour' samples, almost all variants should be marked as unPASS or PoN-labeled. However, I still got almost the same number of PASS variants as the ones mentioned at the beginning, even though the two batches of PASS variants themselves are not the same. 
This is contrary to my expectations. I like to know if this means that the PASS variant I initially found is not real, but a false positive due to things like the analysis or the sequencing depth of the sample itself. Or I just understand the pipeline in the wrong way.