+1 at some point, you need to expect to train a much better base model using everything you've learnt. At the least, you probably want to bring on line the next 10 clever RL environments and ideas your team has been cooking up (which will pipeline into v2.7 etc.)