You can make a prediction about
what not
how. Set things up and E Coli will develop antibiotic immunity, repeat the experiment 100 billion times and you might see 1+ billion different solutions. Insert fragments of a solution and repeat the experiment and you increase the odds to get the same solution, but the original solution space was still vast.
Consider, we say cancer mutates to disable specific genes, but not how those genes are disabled because it's outcomes not methods that are so common.