Evaluation runs

Agent turns and scenario trials with configuration, environments and outcome.