I've read in other closed issues that you suggest not to pre-filter reads because that will distort the signal and that's reasonable but I noticed a strange situation.
Here's the raw counts of a variant obtained from IGV (reference is T)

chr10:123,241,496
Total count: 137
A : 0
C : 137 (100%, 59+, 78- )
G : 0
T : 0
N : 0

So, I would expect even with strict filters to see an AF of 100%.
Instead using -q 30 -Q 30 -m 30 as quality filters this is what I get
chr10 123241496 . T C 2874 PASS DP=123;AF=0.788618;SB=0;DP4=0,0,52,71
Can you help understand why that is happening?