[EM] 0% IC manipulability in (n-1)-out-of-n elections: bug or just that good?

[Kristofer Munsterhjelm]

I thought I'd give some results from my testing of multiwinner methods 
when asked to fill n-1 seats with n candidates running. I should still 
do so, but I got some strange results that seem almost too good to be true.

Schulze STV has manipulability zero under impartial culture, which means 
it's almost everywhere strategyproof under that model if you have enough 
voters.

This seems obviously wrong, like it should be a bug; but I haven't had 
much luck finding *what* that bug is. So if anybody could try to 
independently verify it or prove it wrong, that would be much appreciated!

Concretely, my Monte-Carlo solver found the following manipulability values:

99 voters, 3 cands, 2 seats, impartial culture: 0/15000 (0%)
99 voters, 3 cands, 2 s., Gaussian spatial 4D, sigma=1: 2558/15000 (17%)
99 voters, 4 cands, 3 s., impartial culture: 0/15000 (0%)
99 voters, 4 cands, 3 s., Gaussian spatial 4D, sigma=1: 4398/15000 (29%)

As a comparison, Meek STV has the following values:

99/3/2, impartial culture:          7042/15000 (47%)
99/3/2, Gaussian spatial, 4D, s=1:  6468/15000 (43%)

99/4/3, impartial culture:         11904/15000 (79%)
99/4/3, Gaussian spatial, 4D, s=1: 10969/15000 (73%)

It's hard to think of a bug that only hits IC unless the bug is in the 
IC generator itself. But 0% manipulability is very strange, even for a 
DSV-like method - which, as far as I understand, Schulze STV is, due to 
calculating the strength of vote managements.

-km