[EM] Introducing Pivot

[Carl Schroedl]

Hi Kristofer!

Sorry for the delayed reply. I agree that formal verification would be more
rigorous. I don't know how to do formal verification either, but I would be
open to collaborating with someone who had experience doing it.

I'm glad you are excited about the idea of a collection of
implementation-independent tests! A few different Pivot Libre projects have
been using one format for ranked ballots and ranked results. I posted a
draft specification of the format here:
https://pivot-libre.github.io/bff/

Folks who are handy with git and GitHub are welcome to discuss and propose
changes to the source file here:
https://github.com/pivot-libre/pivot-libre.github.io/blob/master/bff.md
I'm also open to discussing it over email; either on the list or privately.

Some of the many open questions:
 * Should the spec say something about how head-to-head matchup matrices
are encoded? If it's relevant to include, I'm hoping something standard
like a JSON array or map could be used so that we don't have to write a
parser, but I'm open to other ideas.

 * Should the ballots and the expected result (and possibly the
head-to-head matchup matrix) for a single test election all be in one file
together? All in separate files?

 * What lessons can be learned from previous attempts to standardize in
this space? Most notably Election Markup Language
https://en.wikipedia.org/wiki/Election_Markup_Language .

One of our group members had a somewhat similar idea to you about
generating random ballots. His work-in-progress tool generates random
graphs with a Condorcet winner and then calculates the BFF ballots needed
to generate such a graph.
https://github.com/pivot-libre/ballot-gen
He's not on this mailing list yet, so let me know if you want to be
introduced.

Thanks for pointing me to Steve's examples, for pointing me to your
program, and for drafting a few test ballots of your own! I admire the
drive to create a useful test scenario with a minimal number of ballots.
I'll be slower to reply to emails over the holidays, but I look forward to
further discussion and collaboration!

Wishing you all the best,

Carl

On Wed, Dec 19, 2018 at 4:07 PM Kristofer Munsterhjelm <km_elmet at t-online.de>
wrote:

> On 12/12/2018 21.20, Kristofer Munsterhjelm wrote:
>
> > I'd like to make a test set, but I've recently found myself rather short
> > on spare time, so I don't think I could contribute much, at least not at
> > the moment.
>
> Here's a test entry that I made while constructing a Smith,Minmax
> mono-add-top failure example with as few voters as possible:
>
> Election 1:
>
> 8: B>D>A>C
> 9: C>A>B>D
> 3: C>D>A>B
> 5: D>A>B>C
>
> The Smith set is {A, B, D}. C is the Plurality winner and also the
> Minmax winner. Schulze elects A.
>
> Election 2:
>
> 8: B>D>A>C
> 9: C>A>B>D
> 3: C>D>A>B
> 5: D>A>B>C
> 1: A>B>C>D
> 1: A>C>D>B
>
> The Smith set consists of all four candidates, and the outcomes should
> be the same.
>
> This is a Smith,Minmax failure example because C is automatically
> disqualified in the first election since C is not in the Smith set; but
> adding some A-first ballots gets C into the Smith set, after which it wins.
>
> Incidentally, I haven't been able to find a four-candidate Smith,Minmax
> failure example that also fails MAM. That might be a good challenge to
> either show impossible or to devise.
>
> Finally, there are example elections where every positional system
> except Plurality fails Majority. Could those be good test cases?
>


-- 

Carl Schroedl | carlschroedl at gmail.com | http://carlschroedl.com/blog
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20181220/b6466765/attachment.html>