I should say all tests were done using the LARGE unit size. Huge might have a massive impact on results seen here. Normal or small would probably be much easier for the AI.