I think I missed why is this faster? What I’m reading here is it’s similar to constrained decoding but I’m not seeing the explanation of why it’s able to get those results.