Missing positions DeepVariant output VCF

Hi, I am working with a VCF which was generated using DeepVariant and find that a lot of positions are missing which I would expect to find.

I can see that the data in the VCF file is not continous. So it doesn’t contain positions which I can see in the CRAM files should be covered by plenty reads.
The missing positions also include ones, where I know there should be variation (so they are not missing because all samples are the same, or the same as reference)

Do you have any ideas as to why these positions are not ending up in the VCF file or can you point me in a direction for some relevant reading material?