Let $U \subset \mathbb{N} \times \mathbb{N}$, and let $U_n = \{ x \mid (n, x) \in U \}$. Let's call $U$ universal for some class $\mathcal{S}$ of subsets of $\mathbb{N}$ if $S \in \mathcal{S} \iff \exists n : S = U_n$. In other words, the set of "cuts" of $U$ contains all the sets in $\mathcal{S}$ and nothing more.
Now, let $U$ be universal for the class of all enumerable subsets of $\mathbb{N}$. Prove that $K = \{ x \mid (x, x) \in U \}$ is enumerable but not decidable.
So, my proof sketch is as follows:
Enumerability is obvious, so let's focus on decidability. I know that there exists an enumerable undecidable $F \subset \mathbb{N}$. By definition of $U$, it must be contained in a cut with some index $k$. Now, deciding whether $(k, k) \in U$ is analogous to deciding whether $k \in F$. Any function $f$ for the latter is not computable, hence, the whole function $u$ that decides whether $(x, x) \in U$ for arbitrary $x$ is not computable, so $K$ is undecidable.
There are a couple of things I don't like about this.
Firstly, the proposition that $f$ being undecidable implies $u$ is undecidable seems to follow from the definitions in my book (where a set is decidable if its characteristic function is computable), but, intuitively, it just feels unjustified.
Secondly, have I proved too much? A similar argument could be applied to show that any $K' = \{ x \mid (x, f(x)) \in U \}$ is undecidable for arbitrary $f$ — it's sufficient for $x$ to range over the whole $U$ to capture the undecidable set that's contained in its cut. Is that correct?
I guess I was overly obsessed with reducing this to something having to do with an enumerable undecidable set.
Indeed, the usual diagonalization approach just works: assume $K$ is decidable, then $\overline{K} = \{ x \mid (x, x) \not \in U \}$ is also decidable, then $\overline{K}$ is enumerable, then it's contained, let's say, in $U_x$ (by definition of $U$). Then the contradiction easily follows.