Regarding this line:

Use -K 100000000 to achieve deterministic alignment results (Note: this is a hidden option)

Can you explain which randomness this hidden option eliminates? As far as I know, the alignments are deterministic given identical inputs, but batching or reordering of the reads would cause changes in output. Here is a description of the randomness: https://www.biostars.org/p/238628/#238817

What are the requirements for BWA-MEM version in order to comply with the functional equivalence specification?

In several documents it's referred to in several different ways - on github it's "Version: We will use 0.7.15" while in the paper 0.7.15 is referred to as "minimum version of the BWA-MEM aligner".

Judging by the manuscript this would mean that higher versions of BWA-MEM are allowed?

Could someone confirm that version above 0.7.15. comply with functional equivalence specification.

Which fasta file is the GRCh38DH reference that should be used in the pipeline standard?

GRCh38_full_analysis_set_plus_decoy_hla-extra.fa
GRCh38_full_analysis_set_plus_decoy_hla.fa