Hill-Climbing Algorithm

We have some ideas for 10B but the problem is that our hill-climbing algorithm depends too much on randomness (ie it takes some random variables and sees how close the fitness of the deciphered text is to english), which means is would take a long time to find the things that we need (as it would just ‘stumble upon’ it). Looking through some online resources, the methods are very similar to the one(s) we’re using. We were wondering if anyone knew how to better optimise the hill-climbing algorithm so as not to rely so much on randomness? Some other ones we’ve coded (such as vigenere and monoalphabetic) use finding the best variables for some critera that we give it, but we are struggling to figure out how to best approach this one (as the cipher is different and works differently so finding the first bit may not mean the best bit). Sorry for the vagueness, didn’t want to specify too much and spoil the challenge! Any help would be much appreciated 🙂

We might delay replies to this depending on how generic they are, but it is an excellent question so we thought we would put it out there to see what responses it gets. Please don’t post code, as the Elves are now busy packing the sleigh and won’t have time to check it! Harry