so the one-time pad is introduced in the disappointing way it always is (emphasis mine):

In the one-time pad, keys are random binary strings. A message, assumed to be a binary string, is encrypted by exclusive-oring the key to it, one bit at a time. It is not hard to see that this system is unconditionally secure in the sense that the eavesdropper can gain no knowledge about the plaintext (except its length) even with infinite computer resources. The one-time pad however suffers from the obvious defect that its key is as long as the message.

so basically literally ZERO textbooks actually approach this topic in a rigorous way but there is actually a very specific rationale for "unconditional security" here that can be codified without too much effort. and if you do that, you'll note that while there is a 1:1 mapping from secret key x plaintext => secret message, it also ends up leaking info about the secret key--so if your RNG isn't a CSPRNG and/or if the attacker can influence its values, then you lose security entirely (in fact, you lose message bits as a linear function of your secret state bits).

so there is clearly more than one axis here. and we haven't gotten into public-key cryptography yet, but this is super fucking important because THERE IS NO ATTEMPT TO CREATE AN ACTUALLY UNDECIDABLE SYMMETRIC CIPHER ALGORITHM