Evaluates text or model outputs against specified values or policies and returns a numeric alignment score with explanatory feedback.