Same approach: write a version which tracks how much information you could possibly have inferred from each existing move and swaps colours round so as to be compatible with previous revelations but to maximize the necessary chain of tests to reach a solution?