"System one" decision models are models that infer and respond with calibrated probabilities or every allowed answer. Consider your everyday language model, to get typed output from it (JSON), you may use Structured Output to constrain the output to guaranteed valid JSON. While model prefills the input in one pass,