I thought about this for 5 minutes and really got nowhere. Can anyone come up with a nice analytic way to think about an algorithm?

There are a couple of brute force approaches: a) always pick the direction that will result in the most blocks to combine, b) always pick the direction that will result in the largest score.