It seems like we are getting the grip on both dynamical downscaling and running things to the supercomputer. At the moment we are running the model for one week (2005 data). Tomorrow morning we expect the computations to finish. Then, once again, it is time to break everything and feed the model with even more data. I guess that is how it works; once you have something that is working, the fun is gone, and you have to make improvements which likely will ruin the setup.