Explore and implement AltTrain, a structural fine-tuning approach to reasoning models. This changes how models process logic to bake-in safety without sacrificing performance.