[EM] Write-ins break condorcet summability

[bql at bolson.org]

On Sun, 10 Jul 2005, Dave Ketchum wrote:

> I believe I have the answer for Condorcet, and that you should be able to 
> build on this for ratings.  I also think of voting at multiple precincts, 
> doing an array at each precinct, and summing the arrays for total district.
>
> May have to scan each ballot for new names at start of its processing, so 
> that new names are added only between ballots.  Certainly do not have to scan 
> all the ballots for new names before starting to count.
>
> Besides the array for all current candidates, keep a dummy column for a dummy 
> candidate, who always loses to every voted candidate.
>
> When a new candidate turns up, start its new array column and row with a copy 
> of the dummy column and a row of zeros, for this candidate has won nothing to 
> date.

Yep. This definitely works to make a good one-pass solution with 
write-ins. Keeping a dummy un-voted-on choice works to sub in those values 
to initialize a new choice.

And I think it will also work for summing across precincts. If a write-in 
choice exists in one precinct and not another then the dummy values should 
be used in the precinct where it doesn't exist.

Brian Olson
http://bolson.org/