Can you, in the algo loop, re-register the evaluation function to something that returns a new fitness?
I explain here how to perform an action at generation N for the case of adding something to the pop, you can imagine how in your case you would instead register a new eval function as one approach:
https://groups.google.com/g/deap-users/c/moRTYQJMHek/m/uREkELBmAwAJAnother method may be to run your algo like eaSimple for n gens and let it finish, then re register a new eval function with the slight changes you need and then run the algo again which will use the existing pop to 'continue' the evolution with the desired changes.
Third option may be to modify the algo and how eval is called, passing along the generation or other info to trigger a difference in logic in the eval function. (need to use source and work with algo locally and modify the map call a little).