Reply To: Help with hillclimb
There is an article by Jakobsen in Cryptologia in 1995 about doing just this. He uses digrams, which is a bad choice with modern computers.
As for my recommendations, you can try my book and work up through Unit 36. There is a link in there for a corpus that has only ASCII characters, and there are formulae for fitness and stuff. There is also a dark yellow book on substitution ciphers (link in the BOSS library) with pseudocode and everything.
As for your post, a fitness of -33 sounds outrageous if you are using logs of frequencies. I suggest looking for coding bugs.
Can you post the ciphertext you are attacking? And the two keys that give -33 and -9?