Right, but you can push unknowns into tmp vars only so much before you have to introduce constraints, otherwise it's all downright undetermined. You have to inject a structure into the placeholder soup or you're just pushing ambiguity around with no real net gain.. which is also fun to play around, question is will you get a paper out of it or even paid if you play like that to no end.

Maybe, (I don't know), but it's easy to accidentally come up with a theory of "mysterious stuff" that appears to explain something, but neither constrains your expectation nor provides predictions.

Phlogiston is the classic example. https://www.lesswrong.com/posts/RgkqLqkg8vLhsYpfh/fake-causa...