Errors in Mutect2 AD, F1R2 and F2R1 counts

Hello GATK team,

We are using GATK 4.0.10.1, and we noticed error in mutect2 ALT/REF count, that we cannot explain. Some variant will be reported with coverage from forward and reverse read were there is no reverse read with alternative.
We looked at original bam and the --bam-output results with samtools tview and none of them display the same numbers as Mutect2 vcf.

Here is an example with Mutect2 VCF line, and corresponding count in orignal bam and bam-out from mutect2. Mutect2 report 5 alternative in reverse strand where at most we see one in mutect2 bam.

We noticed this for ~2% of driver mutations that where "wrongly(?)" called in our project.

Hopefully it comes from our understanding of the data, if so could you explain to us how to correctly check mutect2 results.

Thanks,

Ismael

############
## Mutect2 VCF
X 76939699 . T C . . DP=27;ECNT=1;POP_AF=5.000e-08;TLOD=32.34 GT:AD:AF:DP:F1R2:F2R1:MBQ:MFRL:MMQ:MPOS:SA_MAP_AF:SA_POST_PROB 0/1:16,9:0.370:25:10,4:6,5:34,42:195,232:60:19:0.222,0.364,0.360:0.069,0.013,0.918

ref = 16
alt = 9

ALT_F1R2: 4
ALT_F2R1: 5
REF_F1R2: 10
REF_F2R1: 6

############
## Original bam

ref = 17
alt = 8

ALT_F1R2: 8
ALT_F2R1: 0
REF_F1R2: 12
REF_F2R1: 5


############
## Mutect2 --bam-output
ref = 18
alt = 9

ALT_F1R2: 8
ALT_F2R1: 1
REF_F1R2: 12
REF_F2R1: 6