I managed to get the script working. It worked flawlessly on sample data. Now works almost flawless on real life data with solid output that does not need to be checked for errors. The human invention can still make a change in it which can sometimes improve on the result. But thats to be expected when you dont use heuristics but a genetic algorithm.
In the last few days of the challenge i will try to improve a bit on the fitness function not to mark “choice” as “failure” when a doctor asks for a doubleshift or similar.
I will clearly miss time to do any work on the Optional goal of importing the data from a web source.