The current matching algorithm gives some unpleasant result (higher shifts go to lower shifts and vice versa). I suggest to retry the old algorithm. I could reactivate that on the test server and some evaluation would be needed how it compares to the one on www.
