[EM] Introducing Pivot

[Kristofer Munsterhjelm]

On 18/02/2019 23.28, Carl Schroedl wrote:
> Hi Kristofer,
> 
> Thanks again for the example election scenarios. I created a repository
> for ranked ballot test scenarios on GitHub and added your examples.
> https://github.com/pivot-libre/ranked-ballot-scenarios/
> 
> I've been thinking about this project a bit more broadly. In many
> disciplines it is difficult for engineers to create practical beneficial
> solutions from science's discoveries. Engineers' implementations can be
> incorrect or lag behind the latest science. I'm hoping this ranked
> ballot scenarios repository will make it easy for software engineers to
> faithfully implement the best and latest election science. Viewed
> another way, I'm hoping the repository will make it easy for the work of
> election scientists to be adopted more widely, correctly, and quickly.
> 
> Does the structure I proposed support that goal? How could it be improved?

More to come when I have more time, most likely, but I would suggest
that property failure examples should have a specific directory or file
structure.

There are relative and absolute criteria. Absolute criteria go like "in
any election of this type, that should happen" (e.g. majority: in every
election where A is ranked top by more than a majority, A should win).

Relative criteria go like "if A wins in this election, and then you
modify it like this, then A should/shouldn't win in that election". E.g.
mono-add-top (if A wins, then after adding some ballots ranking A top, A
should still win) or cloning (crowding: if A wins and you clone B, then
A should still win; vote-splitting: if A wins and you clone A, then A
should still win; teaming: if A doesn't win and you clone A, then A
still shouldn't win).

For absolute criteria, the directory structure you've detailed in the
readme is probably good enough (e.g.
"scenarios/borda-majority-failure/"). But for relative criteria
failures, I would suggest that the file/directory structure makes it
clear which ballot set is before and which ballot set is after.

E.g.
ranked-ballot-scenarios/scenarios/smith-minmax-mono-add-top-failure/before/
(the ballot with the three-candidate Smith set) and
ranked-ballot-scenarios/scenarios/smith-minmax-mono-add-top-failure/after/
(after some A-top ballots push C into the Smith set).

The names don't necessarily have to be "before" and "after", but that's
what first came to mind.

You could possibly further categorize the examples, e.g.
criterion-failures/mono-add-top/smith-minmax/before/ or
criterion-failures/smith-minmax/mono-add-top/before/ depending. But
you'd have to judge how much that obscures the point of the repository.