I can do better than the Kentucky Legislature

This story at Kentucky.com says:

The Supreme Court held that the populations of all of the state House and Senate districts shouldn't deviate more than 5 percent, and that the plan should divide as few counties as possible. The proposed new legislative districts had population deviations of up to 10 percent, and the plan split up to 28 counties between legislative districts. Justices found both to be excessive.

10 percent? Really? My redistricting maps for Kentucky have deviation of 2.5% or better. I don't have any guarantees about County unity, but the population equality constraint seems like a basic minimum bar to clear in terms of making correct districts. I have that better than the Kentucky Legislature.

Next steps - towards version 2

I have the world's greatest compact district solver. Now what?
I think the biggest shortcoming that I run into the most often when talking to people about this is the realpolitik and practical administerability shortcomings of my maps. Rivers and county lines and city lines are being crossed willy nilly, and things would actually be better if we kept some people together in definable 'natural' groupings.

So, two likely next major features:

• Keep-Together Regions. If the entirety of some defined region isn't within an end district, people in it on both sides count as further away from their district center by some penalty amount.
• Do Not Cross Lines. If a district is split by one of these lines, people on the far side of it from the center count as further away from the center by some penalty amount.

The keep-together-regions can probably be automatically extracted from Census data that defines city lines and county lines and metropolitan areas. I think do-not-cross-lines will probably have to be set by hand, probably down the middle of rivers and mountain ranges and such. If adopted by law, I'd expect the initial law set by the legislature to define these kinds of features and the strengths of their penalties. Could this be used to game the system? Probably. But I hope these kinds of things will be more justifiable and requiring of justification, and hopefully still produce good results.
I think my goal is to still have a fully automatic solver that makes some sort of 'optimal' districting that is impartial to political gerrymandering. I still think it is key to have a codified measure of what a good district is. I think these kinds of modifications will tweak the naive definition of optimal towards something more practically implementable and palatable to the public and the politicians.