The final group of experiments analyzes ICET's method for searching in bias space. Section 4.3.1 studies the roles of the mutation and crossover operators. It appears that crossover is mildly beneficial, compared to pure mutation. Section 4.3.2 considers what happens when ICET is constrained to search in a binary bias space, instead of a real bias space. This constraint actually improves the performance of ICET. We hypothesized that the improvement was due to a hidden advantage of searching in binary bias space: When searching in binary bias space, ICET has direct access to the true costs of the tests. However, this advantage can be available when searching in real bias space, if the initial population of biases is seeded with the true costs of the tests. Section 4.3.3 shows that this seeding improves the performance of ICET.