[EM] CIVS update

[Steve Eppley]

Hi,

Andrew M replied to Ernie P:
>>> The major difference between CIVS Ranked Pairs and MAM is the rule on 
>>> when to keep a preference. A preference is kept exactly when it does 
>>> not create any new cycles when considered in conjunction with strictly 
>>> stronger, kept preferences. Thus, preferences of equal strength may be 
>>> kept even though in conjunction they produce a new cycle, as long as 
>>> individually they do not.
>> 
>> This sounds to me that Same-Sized Majorities with no other 
>> differentiators will lead to cycles, and CIVS specifies no mechanism 
>> for breaking that, and is thus incomplete.  Is that correct?
> 
> That is correct, and intentional.  CIVS RP doesn't have either 
> of the two random tie-breaking mechanisms in MAM, which
> are necessary to ensure that you always obtain a total
> ordering of the alternatives.

I want to add that CIVS RP is "complete" in the sense that 
it does select one or more winners, as the other so-called 
"deterministic" variations of MAM do.  If CIVS RP affirms 
a cycle at the top, then it selects every alternative in 
that cycle. (Assuming my memory is correct when recalling 
what Andrew told me.)

Perhaps it should be modified so it selects only the 
alternatives in that cycle whose pairwise "defeats" 
were affirmed latest. (That is, had the smallest
affirmed opposing majorities.)  Probably just two 
alternatives instead of three.

--Steve