This 2014 paper by Barak and Goldreich defines the notion of a universal set $S_{\mathcal{U}}$ as the set of all tuples $(M,x,t)$ such that the non-deterministic machine $M$ accepts the input $x$ in $t$ steps or less (page 2). They further go on to describe a relation $R_{\mathcal{U}}$ for $S_{\mathcal{U}}$ which they call the 'natural witness-relation,' defined as
$$R_\mathcal{U}\equiv\{((M,x,t),w)\mid M\text{ accepts $(x,w)$ in $t$ steps or less}\}$$
Here $M$ is changed from a non-deterministic machine to a two-input deterministic machine. Since the purpose of this relation seems to be to list all instance-witness pairs, if I am understanding it correctly then $w$ is a witness for instance $x$ of the problem decided by $M$. That an NDTM can be expressed as a DTM is trivial.
However, just exactly how is the arbitrary machine $M$ in $S_\mathcal{U}$ transformed into the one that accepts $(t,w)$ in the definition of $R_\mathcal{U}$? Is the relation well-defined for arbitrary machines?
Sorry if the question is trivial, I come from a cryptography background and this is my first time dealing with these notions.