[guided]Suppose $B \cap (U \cup V) = V$, so $B$ contains all of $V$ and none of $U$. The compression does not fix $B$: $C_{UV}(B) = (B \setminus V) \cup U \neq B$. To place $B$ in $C_{UV}(\partial \mathcal{A})$, we show both $B \in \partial \mathcal{A}$ and $C_{UV}(B) \in \partial \mathcal{A}$. (The definition of $C_{UV}(\partial \mathcal{A})$ retains any set $D \in \partial \mathcal{A}$ satisfying $C_{UV}(D) \in \partial \mathcal{A}$.)
**Showing $B \in \partial \mathcal{A}$ by proving $A' \in \mathcal{A}$.** We have $V \subseteq B \subset A'$, so $V \subseteq A'$. We claim $A' \in \mathcal{A}$. Suppose for contradiction that $A' \notin \mathcal{A}$. Since $A' \in C_{UV}(\mathcal{A})$, it must appear as $C_{UV}(\tilde{A})$ for some $\tilde{A} \in \mathcal{A}$ with $\tilde{A} \cap (U \cup V) = V$. Then $A' = (\tilde{A} \setminus V) \cup U$, which satisfies $A' \cap V = \varnothing$ (the elements of $V$ were removed and the disjoint set $U$ was added). But $V \subseteq A'$ and $A' \cap V = \varnothing$ force $V = \varnothing$, contradicting $s = |V| \geq 2$. Therefore $A' \in \mathcal{A}$.
Since $B \subset A' \in \mathcal{A}$ with $|A' \setminus B| = 1$, we get $B \in \partial \mathcal{A}$. This handles half of what we need.
**Showing $C_{UV}(B) \in \partial \mathcal{A}$: finding a parent.** Write $A' = B \cup \{x\}$. Since $V \subseteq B$ and $x \in A' \setminus B$, we have $x \notin V$. We split on whether $x \in U$ or $x \notin U$, as this determines whether $C_{UV}$ acts on $A'$.
**If $x \notin U$:** Then $A' \cap U = \varnothing$ (since $B \cap U = \varnothing$ and $x \notin U$), so $A' \cap (U \cup V) = V$ and $C_{UV}(A') = (A' \setminus V) \cup U \neq A'$.
We verify $C_{UV}(A') \in \mathcal{A}$. The set $A'$ belongs to $C_{UV}(\mathcal{A})$ by hypothesis. How does $A'$ enter $C_{UV}(\mathcal{A})$? Recall $C_{UV}(\mathcal{A}) = \{C_{UV}(A) : A \in \mathcal{A}\} \cup \{A \in \mathcal{A} : C_{UV}(A) \in \mathcal{A}\}$. Could $A'$ appear in the first component, i.e., $A' = C_{UV}(\tilde{A})$ for some $\tilde{A} \in \mathcal{A}$? That would require $A' = (\tilde{A} \setminus V) \cup U$, giving $A' \cap V = \varnothing$. But $V \subseteq A'$, so $V = \varnothing$, contradicting $s \geq 2$. Therefore $A'$ enters $C_{UV}(\mathcal{A})$ only via the second component: $A' \in \mathcal{A}$ and $C_{UV}(A') \in \mathcal{A}$.
Now we relate $C_{UV}(B)$ to $C_{UV}(A')$:
\begin{align*}
C_{UV}(B) &= (B \setminus V) \cup U, \\
C_{UV}(A') &= (A' \setminus V) \cup U = ((B \cup \{x\}) \setminus V) \cup U = (B \setminus V) \cup \{x\} \cup U,
\end{align*}
where the last equality uses $x \notin V$. Since $x \notin U$ either, the element $x$ belongs to $C_{UV}(A')$ but not to $U$, so $x \in C_{UV}(A') \setminus C_{UV}(B)$. Therefore $C_{UV}(B) = C_{UV}(A') \setminus \{x\}$, giving $C_{UV}(B) \subset C_{UV}(A') \in \mathcal{A}$ with $|C_{UV}(A') \setminus C_{UV}(B)| = 1$. Hence $C_{UV}(B) \in \partial \mathcal{A}$.
**If $x \in U$:** Set $u_0 = x$. Then $A' \cap U = \{u_0\}$ (since $B \cap U = \varnothing$ and $u_0 \in A' \setminus B$), so $A' \cap (U \cup V) = V \cup \{u_0\} \neq V$ and $C_{UV}(A') = A' \in \mathcal{A}$. The compression $C_{UV}$ fixes $A'$, so $C_{UV}(A')$ does not give us a new set to use as a parent for $C_{UV}(B)$.
Instead, we use the sub-compression hypothesis. By hypothesis, there exists $v_0 \in V$ such that $\mathcal{A}$ is $(U', V')$-compressed, where $U' = U \setminus \{u_0\}$ and $V' = V \setminus \{v_0\}$.
We check that $C_{U'V'}$ acts on $A'$: $A' \cap U' = \varnothing$ (since $B \cap U = \varnothing$ and $u_0 \notin U'$) and $A' \cap V' = V'$ (since $V \subseteq B \subset A'$ and $V' \subset V$). So $A' \cap (U' \cup V') = V'$, and $C_{U'V'}(A') = (A' \setminus V') \cup U'$. Since $\mathcal{A}$ is $(U', V')$-compressed and $A' \in \mathcal{A}$, we get $C_{U'V'}(A') \in \mathcal{A}$.
Now we compute $C_{U'V'}(A')$. Since $V' = V \setminus \{v_0\}$ and $V \subseteq A'$:
\begin{align*}
A' \setminus V' = (A' \setminus V) \cup \{v_0\}.
\end{align*}
(We remove $V \setminus \{v_0\}$ from $A'$, keeping $v_0$.) So:
\begin{align*}
C_{U'V'}(A') = (A' \setminus V) \cup \{v_0\} \cup U'.
\end{align*}
Next we compute $C_{UV}(B)$. We have $B = A' \setminus \{u_0\}$, so:
\begin{align*}
C_{UV}(B) = (B \setminus V) \cup U = ((A' \setminus \{u_0\}) \setminus V) \cup U.
\end{align*}
Since $u_0 \in U$ and $U \cap V = \varnothing$, we have $u_0 \notin V$. Also $u_0 \in A' \setminus V$ (since $u_0 \in A'$ and $u_0 \notin V$). So $(A' \setminus \{u_0\}) \setminus V = (A' \setminus V) \setminus \{u_0\}$, and:
\begin{align*}
C_{UV}(B) = ((A' \setminus V) \setminus \{u_0\}) \cup U = (A' \setminus V) \cup U,
\end{align*}
where the last equality holds because $u_0 \in U$ (removing $u_0$ from $A' \setminus V$ then taking the union with $U$ restores it).
Comparing with $C_{U'V'}(A') = (A' \setminus V) \cup \{v_0\} \cup U'$: since $U = U' \cup \{u_0\}$ and $u_0 \in A' \setminus V$ (because $u_0 = x \in A'$ and $u_0 \notin V$), adding $\{u_0\}$ to $(A' \setminus V) \cup U'$ contributes nothing new. So $(A' \setminus V) \cup U = (A' \setminus V) \cup U'$, and therefore:
\begin{align*}
C_{UV}(B) = (A' \setminus V) \cup U' = C_{U'V'}(A') \setminus \{v_0\}.
\end{align*}
Since $v_0 \in C_{U'V'}(A')$ and $C_{U'V'}(A') \in \mathcal{A}$, we conclude $C_{UV}(B) \in \partial \mathcal{A}$.
**Conclusion for this case:** We have shown $B \in \partial \mathcal{A}$ (from the fact that $A' \in \mathcal{A}$) and $C_{UV}(B) \in \partial \mathcal{A}$ (by finding a parent $C_{UV}(A')$ or $C_{U'V'}(A')$ in $\mathcal{A}$ depending on whether $x \in U$). So $B$ is retained in $C_{UV}(\partial \mathcal{A})$.
Why is this case harder than the previous one? The difficulty is that showing $B \in \partial \mathcal{A}$ alone is not enough: since $C_{UV}(B) \neq B$, the definition of $C_{UV}(\partial \mathcal{A})$ retains $B$ only if *both* $B$ and $C_{UV}(B)$ belong to $\partial \mathcal{A}$. Finding a parent for $C_{UV}(B)$ in $\mathcal{A}$ requires either the structural fact that $C_{UV}(A') \in \mathcal{A}$ (which follows from $A' \in C_{UV}(\mathcal{A})$ when $x \notin U$) or the sub-compression hypothesis (when $x \in U$ and $C_{UV}$ fixes $A'$). This is where the inductive hypothesis is consumed.[/guided]