It would be interesting to take the same model but use known inputs for different eras, i.e. 1600s, 1700s, 1800s, and see when it would then predict collapse.