You can make predictions and verify them. For example, I remember reading that in the context of trying to figure out the extreme ancient history of live billio

You can make predictions and verify them. For example, I remember reading that in the context of trying to figure out the extreme ancient history of live billions of years ago someone made the prediction of some element that I forgot to be present in stone layers of a certain age period at to a very exact amount that was different from what would be expected without that specific theory. When they found such stone that element was present with exactly the predicted concentration. Darn if I remember the details, but this serves to illustrate a point: You don't need to repeat the creation of the earth. You can develop theories and instead of repeating the whole process predict something obscure that isn't obvious at all to a very high precision, and see if at some later point what you predict might be found. Same here: They could make predictions about what they expect to find at what locations, and when something is eventually found we can check if it's true. If you are specific enough with your prediction and not just "we'll find ancient bones" your prediction is as good a validation of the theory as is possible, and the more of it the better. After all, your one prediction isn't the only one, there will be others and everything builds on top of other stuff equally verified. Either a whole section of such a house of knowledge crashes if a prediction doesn't come true (something is found and it's different from predicted), or, on the other hand, more and more predictions are added on top and more and more come true, with each new one you can be a little surer of the basis than before. You won't get much better than that in any field. Maybe math, but that's a "helper", not a field of science (insert the famous Einstein quote here...). Repeatable experiments don't give you that much more - if a lot of such predictions are made and verified it may even be better. A network of many things that all fit together (each one verified by such predictions) is really good evidence, also compared to just a single repeatable experiment. Sure, having both is even better. There are no absolutes.