We recently began considering another benchmark for the shootout, namely a Magic Square via best-first search. This is fairly inefficient, and we may need to shift to another approach due to the extremely large times required to find a solution for larger squares. I thought the Haskell community might be interested in the performance we have measured so far (see "http:// shootout.alioth.debian.org/sandbox/fulldata.php? test=magicsquares&p1=java-0&p2=javaclient-0&p3=ghc-0&p4=psyco-0" Interestingly, Java actually beats the tar out of GHC and Python for N=5x5 (and I assume higher, though this already takes on the order of 2 hours to solve on the benchmark machine). Memory use in GHC stays nice and low, but the time to find the result rapidly grows. I was hoping for an order of magnitude increase with each increase in N, but discovered that it is more like an exponential... Thanks, -Brent