Yes, it has. Folks hundreds of years ago could not articulate the rules of evidence in Bayesian terms, but intuitively understood that it is probabalistic. The question is, what are the odds that all these facts are true and the accused is not the killed. If that is below the threshold for reasonable doubt (1-10%) then a conviction is in order. Circumstantial evidence like “direct” evidence narrows that probability. Knowing the killer stayed at the same hotel as the victim on the night of the murder is circumstantial evidence, but dramatically narrows the probability that the accused is innocent compared to the prior odds ratio. Any given piece of circumstantial evidence may not be enough, but together it can narrow the odds sufficiently to achieve a conviction. (He was in the same hotel, had an insurance policy on the victim’s life, and had recently purchased a gun, which cannot be found.)

What is "circumstantial" evidence? Scientifically, really, isn't all evidence cicumstantial, and "circumstantial" just means evidence with some unspecified (low) level of Bayesian weight or something?

If it weren't, it would absolutely cripple the justice system. Any crime committed with no other people or recording devices present would be literally immune to conviction, no matter how strong the rest of the evidence.

Consider: a closed-room murder mystery. For example, there are five people on an island, and in the night one of them is murdered.

To prove who committed the murder (after proving that the person's death was a homicide at all), all you need is to exculpate two of the three remaining people (by e.g. having them supply air-tight alibis.)

Then, by process of elimination, the last one must be the murderer. Even though that's entirely "circumstantial evidence", it really is a proof, in the deductive sense.

Direct evidence is a witness testifying they directly observed the defendant commit the crime. Any other evidence is circumstantial.