I have found the answer to my own question in Paul Newman's post of 21 Nov 2016:
"For peer-only grading, non-submitting students are assumed to have given everyone 5 marks - the actual number doesn’t matter, so long as it’s the same for everyone and greater than zero, it all normalises out to them giving everyone the same fractional score. The algorithm then runs as normal with these faked-compensatory scores."
With that rule I can replicate the WebPA scores generated by the system. However, that approach dampens extreme scores (pulls marks towards average) when there are missing values -- rightly or wrongly.
Apologies for not finding Paul's info last night before posting.