A deterministic algorithm that accepts no inputs other than the census count and geodetic coordinates of the census tract, and outputs district boundaries, is i

A deterministic algorithm that accepts no inputs other than the census count and geodetic coordinates of the census tract, and outputs district boundaries, is inherently non-unfair. If you cannot inject unfairness during the process, the only unfairness you can output is a different expression of the unfairness that already existed when you started it. You can't intentionally make it worse. I have personally seen dozens of algorithmic redistricting proposals that naively optimize for "geographic compactness". Those maps look fair. But you don't need to use the interocular test--you can use math and produce an objective measure of compactness. The only question is which one to pick. Pick any of them, arbitrarily. It will set an upper limit on the impact of malicious gerrymandering, and be more fair than when completely unconstrained.