I'm familiar with a few independence proofs e.g. Continuum Hypothesis, but I was wondering whether there's a vague since in which we got "lucky" with these, and some statements which are undecidable in ZFC will also have their undecidability be undecidable.
My knee-jerk reaction was no, because by Godel's Completeness Theorem if we have some statement $S$ and $D$:="$S$ is independent of ZFC", then if $D$ is independent of ZFC, there exists a witness in some models to a proof or a disproof of $S$, and since proofs are syntactic a statement can't be (dis)provable in only some models. But this is very hand-wavy, and moreover seems like could equally well be used to show that $CON(PA)$ is not independent of $PA$.
I'm guessing the error is that, even though there are witnesses in some models which the model "thinks" encode a (dis)proof of $S$, they're an analog of nonstandard numbers in $PA$, and can't actually be decoded into a finite proof
So I guess my question is two-fold: 1. Is the above reasoning valid (if a bit hand-wavy), 2. Can the independence of a statement itself indeed be independent? Further, can this "recurse" infinitely, such that for some statements not only can we never decide it, we can't decide that we can't decide it, and we can't decide that we can't decide that...?