another attempt on the same training data with lower learning rate, simply works better for realism
Taylor blends cool restraint with subtle tension.
