A more practical way to do context might be: always include ghost of previous model run?