Every qcqs cdh generator is an arc covering family

For a Nisnevich family and a rank-at-most-one valuation ring $V$ mapping to $X$, choose a member with a point over the image of the closed point; its pullback to $\mathrm{Spec}V$ is étale and has a point over the closed point. Choose an affine neighborhood $\mathrm{Spec}B$ of that point. By the exact valuation input [F1], its local ring $W=B_{\mathfrak{m}}$ is an extension of valuation rings with the same value group as $V$. It therefore has the same rank, and $\mathrm{Spec}W$ supplies the required lift into that cover member. For a family on a qcqs object take a finite subfamily, which exists since étale images are open.

For an abstract blowup family, if the generic point of $\mathrm{Spec}V$ maps into $Z$, the pulled-back ideal is zero in $\mathrm{Frac}V$ and therefore zero in the domain $V$, so the entire map factors through $Z$. Otherwise the generic point has the unique lift to $Y$ over the open complement. Properness and the valuative criterion in [F1] extend it to $\mathrm{Spec}V\to Y$. Thus the abstract family even permits the identity extension $V\to V$.

These valuation tests are the defining tests for arc covers; the source topology is generated by the listed families. Both tests are stable under the source base changes, so every generated cdh covering sieve is arc covering. No finite-type or dimension bound on $X$ was used.