To me CoT is nothing but lowering learning rate and increasing iterations in a typical ML model. It's basically to force the model to make a small step at a time and try more times to increase accuracy.