Lessons Learned

  1. Computers can solve redistricting, impartially, unassisted, with precision probably good enough to pass a court challenge. A home computer left to itself for a week can do everything your state needs.
  2. Those maps (so far) have some shortcomings and inconveniences in realpolitik. Sometimes the lines are annoyingly just a little bit off of city/county lines where actually following those lines would be just as good and make everything easier to administer.

Originally I thought that maybe we could enact just purely the automatic maps. Now I think that wouldn't quite work well.
One solution might be that the legislature gets to draw a map that cannot deviate by more than 5% from the ideal impartial compact map. Even better if they have to publish some sort of audit log about why they make each change. Shenanigans would be limited and detectable.
A persistent core philosophy might be: define a measure for what a good district is
Define a philosophy of what we want out of district mappings, then design a mathematical measurement of that.
(But be careful when designing metrics because the devil is in the unexpected emergent properties of optimizing towards those metrics.)

Lastly, automatic "gerrymandering for good" as mandated by the Voting Rights Act would be hard, but maybe we don't have to anymore after a SCOTUS decision last year.