We tried Strelka2 (actually joint with Manta) for somatic variant calling. As the side products, we also got germline calls. However, when I compare germline calls generated from different tumor-normal pairs (same normal, different tumor), I got very different results: even after FILTER=PASS, only 2/3 of the calls are common.

Is this expected, or do I understand the germline call wrong that it actually used the tumor reads for detection?