The original problem boils down to randomly sampling a 7 dimensional search space with a small amount of information that suggests some corners don't need to be looked in. The space is presumed to be too big and possibly chaotic to be amenable to being searched using a grid or gradient type search for example, so the adopted technique used throws a bunch of darts and hopes some will hit interesting mountains.

For example it may be known that shell fish can not comprise more than 70% of a diet because that causes death in short order (value invented BTW, effect real). But there is no information to say what the probability function is so a flat distribution is used for the search.