Let’s lay this one to rest, shall we?
All science is observational – observations are what we call “data”.
All science is predictive, whether it concerns events that happened in the past, and are unlikely to occur again, or events that are reproducible.
The scientific method is simple: you construct a model that fits a current set of observations, and then use that model to predict new observations.
There IS a difference between correlational and experimental inference, and some of what the likes of Ken Ham call “historical science” is correlational rather than experimental inference. The difference is that in experimental methodology the experimenter manipulates a randomly allocated variable. That way, if the observations correlate with the manipulation, you know the observed phenomena were the result of the manipulation – you know that the observation did not cause the manipulation. Although that may be moot in QM, I don’t know.
But it has no bearing on Ham’s faux distinction. We can make models about the past that we can test by making predictions about what we will find e.g. Tiktaalik, as Bill Nye pointed out. The fact that Tiktaaliks are long dead is as irrelevant to the methodology as the fact that a murder victim is dead is irrelevant to forensic methodology. In fact there is a sense in which all observations are in the past by the time we’ve observed them.
And it is possible to make predictions about what observations we will make if the world is 13 billion years old, and if it is 6000 years old. And the predictions arising from the former model are confirmed by multiple independent observations, and those from the latter by zilch.