Thanks! Certainly giving the model more context on the task it needs to perform would help it. This was actually a part of training that we improved going from Needle 2 to Needle 3