...get money for predicting history...
There's nothing wrong with that.
Let's say you have a computer model that purports to model the world climate. You could feed into it all the historical data we have, up to today, and generate predictions for the next 100 years. Then you just need to wait 100 years to test your model.
Alternatively, you could feed all the data into the model except that for the last 100 years, run it, and then compare its 1911 "predictions" to the actual instrument data.
Interesting idea, isn't it? Maybe someone should tell Phil Jones, PhD, about it.