To avoid race conditions and overwriting of the data directory from subsequent model runs, each model fit should have its own data directory, within the current hierarchy. They should not use a single share data directory in the model_run_dir, as they currently do.