1: \begin{proof}

2:  The formulas for the pullbacks of divisors $\mathcal{H}_{i,j}$ are immediate from the fact that, depending on whether $i \in A$ or $i \in B$ , the composition of the $i$th evaluation map on $Y_{d,n}$ with $\tau$ only depends on one of the factors $\overline M_A$, $\overline M_B$. Moreover, the horizontal position of the points in $B$ agrees with the horizontal position of the gluing point $p$ coming from $\overline M_A$. The pullback of $\mathcal{G}$ in case $d=0$ is also clear.

3:  

4:  It remains to consider the boundary divisors. Two different boundary divisors intersect transversally by \cite[Theorem 3]{fultonpandha} and above we have just distinguished all cases that are combinatorially possible. They correspond to decorated dual graphs of curves with three vertices that are specializations of the dual graphs coming from $D_{B,k}$ and $D_{B',k'}$.

5:  

6:  We are left to consider the self-intersection of the boundary divisor $D_{B,k}$. It is known that at a generic point $(f: C \coprod_{p,p'} C' \to \PP^1 \times \PP^1) \in D_{B,k}$, the normal space of $D_{B,k}$ in $Y_{d,n}$ is exactly given by $T_p C \otimes T_{p'} C'$, the fibre of the bundle $\psi_{p}^\vee \boxtimes \psi_{p'}^\vee$. If $B \neq \emptyset$ or $2k>d$, the map $\tau$ is an embedding and thus indeed the restriction of $D_{B,k}$ via $\tau$ is the normal bundle $\psi_{p}^\vee \boxtimes \psi_{p'}^\vee$.

7:  

8:  However, if $B= \emptyset$ and $2k \leq d$, then $\tau$ is still birational, but it is not injective exactly on the divisor $D_{\emptyset,k}\times_{\PP^1} \overline M_B$, where it is generically $2:1$. Thus we have to add this cycle with multiplicity $1$. \todo{Please proofread} \todoOld{Rigorous argument? ... use that two critical components of $C_W B \times B$ intersect in codimension $2$ set, which can be ignored. Still missing: deformation theoretic argument ...}

9: \end{proof}

10: