To date, we have seen the Higgs boson (I believe) twice, and gravitons exactly zero times, and yet both of these are considered properly scientific concepts. So something in your presentation of 'empirical reliability' is decidedly off.
Science is not really as data-driven as most people seem to believe. In science one makes a theory (often by looking at a small handful of curious observations) and then gives a few cogent examples to demonstrate that the theory works in principle. Usually the idea of replicability is sufficient, and so long as a theory can be used functionally within some practice, no one gives much thought to formal replication. Empirical data becomes important for distinguishing between different forms or nuances of the theory, or for elaborating further details — e.g., once we had the basic theory of the electrical properties of semiconductors, people began experimenting with different levels of doping to see how the material properties could tweak the electrical ones — but empirical research guides theories more than it establishes them.
History as a science is not quite as limited as you've suggested. I mean, sure, if we have a historical question like "What were the causes of World War II" it is obviously the case that there was only one World War II. But when seeking out the causes of that war there is a huge amount of empirical data that can be addressed: political speeches, demographic data, diplomatic communities, personal letters of world leaders, newspaper and magazine articles, economic data, patterns in previous historical conflicts, etc, etc. A historian will create a particular theory about the the origins of WWII, and then access as much of this trove of data as needed to make his case. Some other historian who wanted to replicate (or falsify) that theory, would dig further into the data. This is no different in principle from a what physicist or chemist does, except that a historian has the more difficult job of assessing motivation (something that material scientists can ostensibly ignore).