This is extremely contrived, but it seems quite close to what you're looking for.

Suppose $M$ is transitive model of ZFC and $P\in M$. We want to describe a process for adding a subset $G$ of $P$ to $M$, and in the end we want this process to yield a model $N_G$ extending $M$ with $\text{Ord}\cap M = \text{Ord}\cap N$. 

Obviously not just any set $G$ can be adjoined to $M$ while preserving ZFC, so we our process will only apply to certain "good" sets $G$. We have to figure out what these good sets are.

If one can somehow approximately describe the extension $N_G$ within $M$, one might hope it is possible to propagate ZFC from $M$ to $N_G$. So we want our good sets to be the ones that permit such a description. What would it mean to describe $N_G$?

Well, to start, we should have a definable subclass $M^P$ of $M$ that we can view as terms for the elements of the model $N_G$ we are trying to describe (although of course, we cannot hope to have complete information about membership relation on the terms $M^P$ within $M$). Each good $G$ should yield a way to evaluate these terms and obtain an element of $N_G$. Let $i_G : M^P\to N_G$ be this interpretation function. Of course, $i_G$ should be a surjection.

The structure of $N_G$ should only depend on which elements of $P$ belong to $G$, so we might hope that for any terms $\vec \tau\subseteq M^P$, any elements $\vec a\subseteq M$, any $\Sigma_0$-formula $\varphi$, and any good $G$, if $N_G\vDash \varphi(i_G(\vec \tau),\vec a)$ then in fact there is some $p\in G$ such that for any good $H$ with $p\in H$, $N_H \vDash \varphi(i_H(\vec \tau),\vec a)$. Call this the *truth hypothesis.*

Also if $M$ is to approximate the structure of $N_G$, we might hope that $M$ can see which facts occur no matter which good $G$ we are trying to adjoin. So we demand that for any terms $\vec \tau\subseteq M^P$, any elements $\vec a\subseteq M$ and any $\Sigma_0$-formula $\psi$, the class
$\{(\sigma,b)\in M^P\times M : \text{for all good $G$, }N_G\vDash \psi(i_G(\vec \tau),\vec a,G,\sigma,b)\}$ is definable over $M$. Call this the *definability hypothesis.*

These hypotheses alone imply that for every good $G$, the model $N_G$ is a set-generic extension of $M$. To see this, it suffices by a recently famous theorem of Bukovsky to show that $(M,N_G)$ has the uniform covering property: for any ordinal $\alpha\in M$ and any function  $f : \alpha\to \text{Ord}$ in $N_G$, there is some $F : \alpha\to P(\beta)$ in $M$ such that for all $\xi < \alpha$, $|F(\xi)| \leq |P|$ in $M$ and $f(\xi)\in F(\xi)$. 

Fix $f : \alpha\to \text{Ord}$ in $N_G$, and we will define a uniform cover in $M$. Take a term $\tau\in M^P$ with $i_G(\tau) = f$. For each $\xi < \alpha$, let $F(\xi) = \{\beta : \text{for all good $H$ with $p\in H$}, i_H(\tau)(\xi) = \beta\}$. 

Fix $\xi < \alpha$. We know $F(\xi)$ is a definable class of $M$ by the definability hypothesis. Moreover, the definability hypothesis implies the $M$-definability of the partial function $f_\xi : P\to F(\xi)$ given by setting $f_\xi(p) = \beta$ if for all good $H$ with $p\in H$, $i_H(\tau)(\xi) = \beta$. By the truth hypothesis, for each $\beta\in F(\xi)$, there is some $p\in G$ such that for all good $H$ with $p\in H$, $i_H(\tau)(\xi) = \beta$, or in other words, $f_\xi(p) = \beta$. Thus $f_\xi$ is a partial surjection from $P$ to $F(\xi)$, so $|F(\xi)|\leq |P|$ in $M$. This establishes the Bukovsky criterion.