quick reminder: this is a one shot pydantic token guess. if you try to use a model that was trained on reasoning? you're not getting any of the reasoning.