I agree. Araver's algorithm was pretty clever and certainly did a pretty good job at obfuscating, but it did seem like it would be a pain to calculate each character (especially the whole last prime position thing). I've developed an algorithm that's more in keeping with octopuppy's original method. I think that it's fairly easy to do in your head, though I admit that I wrote up a program to crunch the numbers (er, letters) faster. I'm actually glad I did because I caught an error in one of the ones I did by hand when I was developing it...
Hooray for octopuppy! Showing you are the true Evil Mastermind.
Here's a question resulting from my attempted code development: What's the requirement for having to remember all the steps in the code? You've established that since you cant see the password characters after you've entered them, the remaining letters shouldn't depend too much on what was already entered (for those of us with short memories). But in this hash, there may be an implementation problem in remembering which rules to apply to which characters of the password (eg, does the munber of vowels become the third character or the fourth?).
Octopuppy's original code had some elegance that later ones have lacked: That the same basic rules are applied throughout the construction of the password. Code #2 mixed that up a bit, but at least repeated the same requirements for the first 4 characters and the last 4. Code #3 has essentially 8 independent requirements. The one I've been trying to construct does also, but I've found when trying to apply it I have to look back at my cheat sheet to know what to do next, which is not in keeping with the spirit of the game.
In any case, I didn't need a cheat sheet for my method and I only figured programming it would be faster because I have always been really slow at calculating anything in my head. I'm much faster when I can see it.