I do think an important part that we leave out when we analyze datapoints like "people who test well in one subject tend to test well in other subjects" is that "test taking" is itself a skill that people can be good or bad at.

I was always an expert test taker from a very young age. I loved standardized tests, and excelled every time.

I remember thinking many times how I was pretty sure I could do pretty well on a test on any subject, even if I didn't know the subject itself.

I had a chance to test (hah) that theory when I took a summer job at a college testing center when I was a college student myself. I would give out these tests in a number of different subjects, but I had time to kill while the students were taking the tests. I decided to try taking some for subjects I had never studied.

I scored well above the average scores on these subjects I knew nothing about, I just knew how tests were written. I could pick up context clues in the wording of questions, use my experience on how test questions usually go, and use clues from other parts of the test to figure out how to answer a different question.

That doesn't mean I was any good at those subjects, I just was good at that form of test.

There may be such a thing as pure test-taking skill, but I think it boils down to two things: General knowledge/literacy, and knowing the specific format of the test. A well-written test should not have question context for one thing that informs the answer to another question, or have logical inconsistencies/redundancies/ambiguity that can be exploited. Questions should also not build off of each other in a way that causes cascading failure if you can't answer one. These things may happen in low-stakes classroom exams but they certainly shouldn't in a standardized test.

Practically everyone goes to school in our society, so they have very similar opportunity to learn how to take a test. Some of them never get good at it after many years. Maybe that is a sign that they also wouldn't be good at any academic subjects. You wouldn't have a reliable way to measure that, because test-taking is out. I guess they could try to practice test-taking by taking a ton of random tests and grading themselves, but then they might accidentally become experts in some subjects!