I build models that people downstream actually have to trust: forecasting, causal inference, and the occasional argument for why the p-value isn't the point.