Re the past proxy data, and possibly the recent data, why is the assumption that causation follows gas->temp and not temp(unknown_var)->gas?