Chris Larson method of automatic redistricting

A new automated redistricting algorithm has appeared with concrete examples for Michigan and Virginia. It sounds like it's doing something like a minimum cut through the blocks that have lowest population. It winds up doing this in mostly horizontal or vertical lines, based on how the Michigan map looks. It mostly keeps counties together, but when it decides it needs to cut it can make a long cut on a straight-ish line through many counties.

In the dichotomy of solutions being "fair process" vs "good outcome" this is another "fair process" entry. I like its output subjectively better than the "shortest splitline" algorithm, and the county-cohesion feature will make it more practical, but I still think there's nothing inherent in its structure that will ensure a "good outcome". I still prefer to start at the end, define what kind of outcome we want, and achieve that by whatever methods happen to work. If we have the right metric, then optimizing to it will be okay. (On the other hand, the last century of industrial best practices point out that picking the right metric is problematic and good process needs to be blended with metrics for actual best outcomes.)

https://medium.com/@chrismlarson/a-new-redistricting-algorithm-motivations-and-method-1515233fce04
https://github.com/chrismlarson/redistricting