Does a "multiobjective" return multiple solutions then? To be evaluated by humans?
Yes, the human has to decide which tradeoffs are the best.
Some algorithms have an "archive" which collects multiple non-dominated solutions.
A non-dominated solution means there is no other solution that is better for EVERY objective (i.e., some objectives might be better, and others worse.)
There are techniques for keeping the archive to a reasonable size (so that the solutions aren’t clustered and there is a good variety.)
It would be really interesting to create an n-order homophonic cipher that has all of the edge cases.
Plaintext 1: Very high ngram score, but bad IOC.
Plaintext 2: Very good IOC but bad ngram score.
Plaintext 3: Many dictionary words but all other scores low.
etc
Then combine them into a single n-order homophonic. Make it long enough so the multiplicity is low. Then run it through a multiobjective solver that can still pick out all of the messages.