The multiple feature forecaster currently assumes completion of one feature before starting another. Although this is the optimal way, it seems rarer in the real world. Is there a way to allocate partial throughput to two or more features concurrently?