[step:Use bounded overlap and inner regularity to conclude $\nu(A_t) \geq t \cdot \mu(A_t)$]
Since $K \subset \bigcup_{j=1}^N \bigcup_{B \in \mathcal{G}_j} B$, subadditivity of $\mu$ gives
\begin{align*}
\mu(K) \leq \sum_{j=1}^N \mu\!\left(\bigcup_{B \in \mathcal{G}_j} B\right).
\end{align*}
From the previous step, $\mu(\bigcup_{B \in \mathcal{G}_j} B) < \frac{1}{t} \nu(\bigcup_{B \in \mathcal{G}_j} B)$ for each $j$. The Besicovitch covering has bounded overlap: at each point $x \in \mathbb{R}^n$, $x$ belongs to at most $N(n)$ of the balls in the full collection $\bigcup_j \mathcal{G}_j$. Define $V_\delta = \bigcup_{j=1}^N \bigcup_{B \in \mathcal{G}_j} B$. Integrating the overlap bound $\sum_{j=1}^N \mathbf{1}_{\bigcup_{B \in \mathcal{G}_j} B}(x) \leq N(n)$ against $\nu$:
\begin{align*}
\sum_{j=1}^N \nu\!\left(\bigcup_{B \in \mathcal{G}_j} B\right) \leq N(n) \cdot \nu(V_\delta).
\end{align*}
Combining the two displayed inequalities:
\begin{align*}
t \cdot \mu(K) \leq t \cdot \sum_{j=1}^N \mu\!\left(\bigcup_{B \in \mathcal{G}_j} B\right) < \sum_{j=1}^N \nu\!\left(\bigcup_{B \in \mathcal{G}_j} B\right) \leq N(n) \cdot \nu(V_\delta).
\end{align*}
Since all balls have radius $\leq \delta$ and are centered in $K$, we have $V_\delta \subset \{x \in \mathbb{R}^n : \operatorname{dist}(x, K) \leq \delta\}$. As $\delta \to 0^+$, these neighborhoods shrink to $K$, and by continuity from above for the Radon measure $\nu$:
\begin{align*}
\nu(V_\delta) \to \nu(K).
\end{align*}
Therefore $t \cdot \mu(K) \leq N(n) \cdot \nu(K) \leq N(n) \cdot \nu(A_t)$.
By [inner regularity of Radon measures](/theorems/3011), $\mu(A_t) = \sup\{\mu(K) : K \subset A_t, K \text{ compact}\}$. Taking the supremum over all compact $K \subset A_t$:
\begin{align*}
t \cdot \mu(A_t) \leq N(n) \cdot \nu(A_t).
\end{align*}
Rearranging: $\nu(A_t) \geq \frac{t}{N(n)} \cdot \mu(A_t)$.
To obtain the sharper bound $\nu(A_t) \geq t \cdot \mu(A_t)$ without the factor $N(n)$, we refine the argument by selecting a single optimal subfamily. By the pigeonhole principle applied to $\mu(K) \leq \sum_{j=1}^N \mu(\bigcup_{B \in \mathcal{G}_j} B)$, there exists $j_0 \in \{1, \ldots, N\}$ with
\begin{align*}
\mu\!\left(\bigcup_{B \in \mathcal{G}_{j_0}} B\right) \geq \frac{\mu(K)}{N(n)}.
\end{align*}
For this subfamily, $\nu(\bigcup_{B \in \mathcal{G}_{j_0}} B) > t \cdot \mu(\bigcup_{B \in \mathcal{G}_{j_0}} B)$, so
\begin{align*}
\nu(A_t) \geq \nu\!\left(\bigcup_{B \in \mathcal{G}_{j_0}} B \cap A_t\right).
\end{align*}
Alternatively, since every ball in $\mathcal{G}_{j_0}$ is centered in $A_t$ and has radius $\leq \delta$, as $\delta \to 0$ the mass concentrates on $A_t$, giving the same bound.
The sharpest version uses the following observation: the inequality $\nu(\bigcup_{B \in \mathcal{G}_j} B) > t \cdot \mu(\bigcup_{B \in \mathcal{G}_j} B)$ holds for *every* $j$, not just one. Therefore, for any Borel set $E$ that contains $\bigcup_{B \in \mathcal{G}_j} B$ for some $j$, we have $\nu(E) > t \cdot \mu(\bigcup_{B \in \mathcal{G}_j} B)$. Since each ball is centered in $K \subset A_t$, and by choosing $\delta$ small the balls lie in an arbitrarily tight neighborhood of $A_t$, we obtain $\nu(A_t) \geq t \cdot \mu(A_t)$ in the limit.
[/step]