[guided]The atomic proof is a simultaneous induction on
\begin{align*}
(\operatorname{rank}(\sigma),\operatorname{rank}(\tau)).
\end{align*}
The order is
\begin{align*}
(\alpha',\beta')\prec(\alpha,\beta)
\end{align*}
when $\alpha'\leq\alpha$, $\beta'\leq\beta$, and at least one inequality is strict. Thus, if $(\rho,r)\in\tau$, then
\begin{align*}
(\operatorname{rank}(\sigma),\operatorname{rank}(\rho))
\prec
(\operatorname{rank}(\sigma),\operatorname{rank}(\tau)),
\end{align*}
and if $(\rho,r)\in\sigma$, then
\begin{align*}
(\operatorname{rank}(\rho),\operatorname{rank}(\tau))
\prec
(\operatorname{rank}(\sigma),\operatorname{rank}(\tau)).
\end{align*}
These are exactly the lower atomic cases used in the recursion.
For membership, assume first that $p\in G$ and
\begin{align*}
p\Vdash_M\sigma\in\tau.
\end{align*}
The recursive clause says that below $p$ it is dense to find $q$ and a tag $(\rho,r)\in\tau$ such that
\begin{align*}
q\leq r
\end{align*}
and
\begin{align*}
q\Vdash_M\sigma=\rho.
\end{align*}
Since $G$ is $M$-generic and $p\in G$, choose such a $q\in G$. From $q\leq r$ and $q\in G$, upward closure gives $r\in G$. The equality case for the lower pair $(\sigma,\rho)$ gives
\begin{align*}
\sigma^G=\rho^G.
\end{align*}
Because $(\rho,r)\in\tau$ and $r\in G$, we have $\rho^G\in\tau^G$, and hence
\begin{align*}
\sigma^G\in\tau^G.
\end{align*}
Conversely, suppose
\begin{align*}
\sigma^G\in\tau^G.
\end{align*}
Choose $(\rho,r)\in\tau$ such that $r\in G$ and
\begin{align*}
\sigma^G=\rho^G.
\end{align*}
By the lower equality case, choose $q\in G$ with
\begin{align*}
q\Vdash_M\sigma=\rho.
\end{align*}
Directedness gives $p\in G$ such that
\begin{align*}
p\leq q
\end{align*}
and
\begin{align*}
p\leq r.
\end{align*}
For every $u\leq p$, monotonicity gives
\begin{align*}
u\Vdash_M\sigma=\rho,
\end{align*}
and also $u\leq r$. Therefore every $u\leq p$ belongs to the dense set required by the recursive clause for $p\Vdash_M\sigma\in\tau$, using the same tag $(\rho,r)$. Thus
\begin{align*}
p\Vdash_M\sigma\in\tau.
\end{align*}
For equality, first assume $p\in G$ and
\begin{align*}
p\Vdash_M\sigma=\tau.
\end{align*}
Let $(\rho,r)\in\sigma$ with $r\in G$. Choose $p_0\in G$ such that
\begin{align*}
p_0\leq p
\end{align*}
and
\begin{align*}
p_0\leq r.
\end{align*}
The recursive equality clause gives a dense set below $p$ of conditions $q$ such that either
\begin{align*}
q\perp r
\end{align*}
or
\begin{align*}
q\Vdash_M\rho\in\tau.
\end{align*}
Genericity below $p_0$ gives $q\in G$ in this dense set. Since $q,r\in G$, the alternative $q\perp r$ is impossible. Therefore
\begin{align*}
q\Vdash_M\rho\in\tau.
\end{align*}
The lower membership case for $(\rho,\tau)$ gives
\begin{align*}
\rho^G\in\tau^G.
\end{align*}
So every element contributed to $\sigma^G$ lies in $\tau^G$. The symmetric half of the equality clause gives
\begin{align*}
\tau^G\subseteq\sigma^G,
\end{align*}
and therefore
\begin{align*}
\sigma^G=\tau^G.
\end{align*}
For the converse equality direction, assume
\begin{align*}
\sigma^G=\tau^G.
\end{align*}
Define $D$ to consist of all conditions $s$ such that either
\begin{align*}
s\Vdash_M\sigma=\tau,
\end{align*}
or there is $(\rho,r)\in\sigma$ with
\begin{align*}
s\leq r
\end{align*}
and no extension of $s$ forces
\begin{align*}
\rho\in\tau,
\end{align*}
or there is $(\rho,r)\in\tau$ with
\begin{align*}
s\leq r
\end{align*}
and no extension of $s$ forces
\begin{align*}
\rho\in\sigma.
\end{align*}
This definition uses only $\sigma,\tau$ and lower-rank forcing relations, so $D\in M$.
To see that $D$ is dense, fix $p_0\in\mathbb{P}$. If some $s\leq p_0$ satisfies
\begin{align*}
s\Vdash_M\sigma=\tau,
\end{align*}
then $s\in D$. Otherwise no extension of $p_0$ forces $\sigma=\tau$. Hence $p_0$ fails the recursive equality clause. For example, suppose the failed requirement is witnessed by $(\rho,r)\in\sigma$. Then
\begin{align*}
A_{\rho,r}=\{q\leq p_0:q\perp r\text{ or }q\Vdash_M\rho\in\tau\}
\end{align*}
is not dense below $p_0$. Choose $s_0\leq p_0$ such that no $q\leq s_0$ lies in $A_{\rho,r}$. Since no extension of $s_0$ is incompatible with $r$, the conditions $s_0$ and $r$ are compatible. Choose
\begin{align*}
s\leq s_0,r.
\end{align*}
Then every $u\leq s$ is also below $s_0$, so no such $u$ forces
\begin{align*}
\rho\in\tau.
\end{align*}
Thus $s\in D$ by the second alternative. The symmetric failed requirement for a tag in $\tau$ gives the third alternative. Therefore $D$ is dense.
Choose $p\in G\cap D$. If $p$ is in the second alternative, then for some $(\rho,r)\in\sigma$ we have
\begin{align*}
p\leq r
\end{align*}
and no extension of $p$ forces $\rho\in\tau$. Since $p\in G$ and $p\leq r$, also $r\in G$, so
\begin{align*}
\rho^G\in\sigma^G.
\end{align*}
Using $\sigma^G=\tau^G$, this gives
\begin{align*}
\rho^G\in\tau^G.
\end{align*}
The lower membership case gives $q\in G$ such that
\begin{align*}
q\Vdash_M\rho\in\tau.
\end{align*}
Choose $u\in G$ with
\begin{align*}
u\leq p
\end{align*}
and
\begin{align*}
u\leq q.
\end{align*}
By monotonicity,
\begin{align*}
u\Vdash_M\rho\in\tau,
\end{align*}
contradicting the assertion that no extension of $p$ forces $\rho\in\tau$. The third alternative is symmetric. Thus the only possible alternative for $p\in D$ is
\begin{align*}
p\Vdash_M\sigma=\tau.
\end{align*}[/guided]