I created a new issue relative to multi-master coherence processing of multi satellite (S1A and S1B) stacks:
The issue is coherence pairs are computed pairwise through S1B dates, the last S1B and first S1A date, and then pairwise through S1A dates. So the only multi satellite pair is not useful since the longest time baseline. When computing multi-master the dates should be sorted and run pairwise. Currently coherence processing appears to run pairwise through a stack that is reverse sorted by satellite first, and then by date.