\section{Places of diagram 3 at \texorpdfstring{$(4,4,4,4,1)$}{(4,4,4,4,1)}}\label{sec:d3} Let $P$ be a singular place of an irreducible curve $C\subseteq Y_{A'}$ with type $\pc h(s;D_1|D_2)$. By \cite[Theorem~3.9]{Tom26} and Lemma~\ref{lem:allcurves}(c), $m_im_j=m_km_l$ for $D_1=\{i,j\}$, $D_2=\{k,l\}$. If $s$ is heavy, one diagonal contains the light body and this fails, since $4\ne16$. So $s=5$, and the type is $\pc h(5;13|24)$, $\pc h(5;12|34)$ or $\pc h(5;14|23)$. The relabellings $(23)$ and $(34)$ fix $A'$ and map the second and the third onto the first. Let $P$ have type $\pc h(5;13|24)$. We call the pairs $12$, $14$, $23$, $34$ the \emph{sides} and write $k'$ for the body opposite to $k\le4$: $1'=3$, $2'=4$. By \cite[Appendix~C]{Tom26} and Lemma~\ref{lem:allcurves}(c) the exponents after the rotation are $e(u_k)=e(v_k)=-2$, $e(p_{kl})=0$ on the sides and $e(p_{kl})=2$ on the six other pairs. Thus \S\ref{ss:lifting} applies with $f=v_1$, and $e'_{v_1}=-2$. We write $u_k$, $v_k$ for $y_{u_k}$, $y_{v_k}$, with $u_1=v_1=1$, and $k_e$, $q_e$ for $y_{p_e}$ on the sides and on the other pairs. At $A'$ the weighted rows are as follows. \begin{itemize} \item Sides $kl$: $k_{kl}^2(u_k-u_l)(v_k-v_l)-T^2$. Diagonals $13$, $24$: $q_{kl}^2(u_k-u_l)(v_k-v_l)-1$. Pairs $k5$: $q_{k5}^2u_kv_k-1$. \item $Z_k=Z^0_k+T^3Z^3_k$ with $Z^0_k=u_k-\sum_{l:\,kl\text{ a side}}4k_{kl}^3(u_k-u_l)$ and $Z^3_k=-\big(4q_{kk'}^3(u_k-u_{k'})+q_{k5}^3u_k+\Gamma\big)$, where $\Gamma=4\sum_{l\le4}q_{l5}^3u_l$. \item $W_k$: the same with $v$ in place of $u$. \end{itemize} So $G=G_0+T^2G_2+T^3G_3$, where $G_2$ is $-1$ on the side rows and $0$ on the other rows. By \cite[Appendix~C]{Tom26} and Lemma~\ref{lem:allcurves}(c) the leading solutions are as follows. In \emph{case~A}, $u=(1,-1,-1,1)$ and $v=(1,1,-1,-1)$; in \emph{case~B}, $u$ and $v$ are exchanged. In both cases $k_e^3=\frac18$ on the four sides, $q_{13}^2=\frac14$, $q_{24}^2=-\frac14$, $q_{15}^2=q_{35}^2=1$ and $q_{25}^2=q_{45}^2=-1$, and every choice of the roots occurs. There are therefore $3^4\cdot2^6=5184$ leading solutions in each case. In both cases $u_{k'}=-u_k$ and $v_{k'}=-v_k$. The matrix $J_0$ has rank $16$, and in case~A its left kernel is spanned by $\lambda_Z=-Z_1-Z_2+Z_3+Z_4$ and $\lambda_W=W_1-W_2-W_3+W_4$, where a row stands for the corresponding unit vector. Moreover $\lambda_Z\cdot G_3=O/16$ and $\lambda_W\cdot G_3=-O/16$ at the leading solution, where \[ O=16\,(4q_{13}+4q_{24}+q_{15}+q_{25}+q_{35}+q_{45}) . \] The relabelling $(24)$ and the colour exchange $\sigma_c$ map places of type $\pc h(5;13|24)$ at $A'$ to places of the same type. On branches they act by permuting coordinates: $(24)$ by $u_k\mapsto u_{\sigma(k)}$, $v_k\mapsto v_{\sigma(k)}$, $p_{kl}\mapsto p_{\sigma(k)\sigma(l)}$ with $\sigma=(24)$, and $\sigma_c$ by exchanging $u$ and $v$ (as $u_1=v_1=1$). Both map branches to branches. The colour exchange exchanges the two cases, and $(24)$ followed by $\sigma_c$ maps case~A to case~A. The following condition mirrors the relations among the $p_{kl}$ in the definition of $\Lambda(\zeta_1,\zeta_2)$. A leading solution is \emph{balanced} if $k_{12}=k_{34}$, $k_{14}=k_{23}$, $q_{15}=q_{35}=-2q_{13}$ and $q_{25}=q_{45}=-2q_{24}$. There are $36$ balanced leading solutions in each case. \begin{lemma}\label{lem:d3a} Let $y$ be a branch for the type $\pc h(5;13|24)$ at $A'$. Then $y(0)$ is balanced. \end{lemma} \begin{proof} Since $\sigma_c$ exchanges the cases and preserves balance, we may assume that $y_0=y(0)$ is in case~A. We show that the obstruction at order~$3$ vanishes only if $q_{15}=q_{35}=-2q_{13}$ and $q_{25}=q_{45}=-2q_{24}$, and that the even part of the expansion then forces $k_{12}=k_{34}$ and $k_{14}=k_{23}$ at $y_0$. By Lemma~\ref{lem:lift}, $y=\sum_ny_nT^n$ is a power series in $T$ and $\lambda\cdot\beta_n=0$ for all $n$ and all $\lambda$ in the left kernel of $J_0$. \emph{Orders $1$ to $3$.} Since $G$ has no term in $T$, $\beta_1=0$ and $y_1=0$. Then $\beta_3=G_3(y_0)$, because $G_0(y_0+y_2T^2+\dots)$ and $T^2G_2$ contribute only even powers up to $T^3$. Hence $0=\lambda_Z\cdot\beta_3=O/16$. The numbers $q_{13}$, $q_{15}$, $q_{35}$ are real and $q_{24}$, $q_{25}$, $q_{45}$ are imaginary, with $|4q_{13}|=|q_{15}|+|q_{35}|$ and $|4q_{24}|=|q_{25}|+|q_{45}|$. Therefore $O=0$ if and only if \begin{equation}\label{eq:O0} q_{15}=q_{35}=-2q_{13},\qquad q_{25}=q_{45}=-2q_{24}. \end{equation} Then $\Gamma(y_0)=4\big(q_{15}^3(u_1+u_3)+q_{25}^3(u_2+u_4)\big)=0$ and $Z^3_k(y_0)=-u_k(8q_{kk'}^3+q_{k5}^3)=0$, and likewise $W^3_k(y_0)=0$. So $\beta_3=0$ and $y_3=0$. \emph{Parity.} Since $y_1=y_3=0$, we can write $y=Y(S)+T^5\Omega(S)$ with $S=T^2$ and power series $Y$, $\Omega$. We write $U$, $V$, $K$ for the parts of $Y$ in $u$, $v$ and the sides, and $\Delta U_{kl}=U_k-U_l$, $\Delta V_{kl}=V_k-V_l$. For a polynomial $F$, the even part of $F(y)$ in $T$ is $F(Y)+O(S^5)$, and the even part of $T^3Z^3_k(y)$ is $O(S^4)$. The side rows, $Z^0$ and $W^0$ are polynomials in $u$, $v$, $k$ and $T^2$. Taking even parts of $G(T,y)=0$ we obtain \begin{equation}\label{eq:even} K_{kl}^2\,\Delta U_{kl}\,\Delta V_{kl}\equiv S\pmod{S^5}\ \ \text{on the sides},\qquad Z^0(U,K)\equiv W^0(V,K)\equiv0\pmod{S^4}. \end{equation} \emph{The Laplacian.} The rows $Z^0$ and $W^0$ are linear in $U$ and $V$, with coefficients from the sides. Put $s_e=4K_e^3$ and $a=s_{12}$, $b=s_{23}$, $c=s_{34}$, $d=s_{41}$. Let $\mathcal L$ be the weighted Laplacian of the cycle $1234$, $(\mathcal Lx)_k=\sum_{kl\text{ a side}}s_{kl}(x_k-x_l)$, so that $Z^0=(I-\mathcal L)U$ and $W^0=(I-\mathcal L)V$. In the orthonormal basis \[ e_0=\tfrac12(1,1,1,1),\quad e_1=\tfrac12(1,-1,-1,1),\quad e_2=\tfrac12(1,1,-1,-1),\quad e_3=\tfrac12(1,-1,1,-1) \] write $x=\sum_iX_ie_i$ and $\bar X=(X_1,X_2,X_3)$. We write $\hat U_i$, $\hat V_i$ for the coefficients of $U$, $V$ in this basis, and $\hat U=(\hat U_1,\hat U_2,\hat U_3)$, $\hat V=(\hat V_1,\hat V_2,\hat V_3)$. Then $\mathcal Le_0=0$, the side differences are $x_1-x_2=X_1+X_3$, $x_2-x_3=X_2-X_3$, $x_3-x_4=X_3-X_1$ and $x_4-x_1=-X_2-X_3$, and on $e_0^\perp$ the matrix of $\mathcal L$ is \[ M=\begin{pmatrix}a+c&0&a-c\\0&b+d&d-b\\a-c&d-b&a+b+c+d\end{pmatrix}. \] By \eqref{eq:even}, $N\hat U\equiv N\hat V\equiv0\pmod{S^4}$, where $N=M-I$. At $S=0$ we have $s_e=\frac12$, so $M(0)=\mathrm{diag}(1,1,2)$, $\hat U(0)=(2,0,0)$ and $\hat V(0)=(0,2,0)$. Let $\varepsilon_3=(0,0,1)$ and $n=N\varepsilon_3=(a-c,\ d-b,\ a+b+c+d-1)$, so that $n_3(0)=1$. The matrix $P=[\hat U\ \hat V\ \varepsilon_3]$ is invertible over $\C[[S]]$, since $\det P(0)=4$. From $NP\equiv[0\ 0\ n]$ we get $N\equiv n\,r^{\mathsf T}$, with $r^{\mathsf T}$ the last row of $P^{-1}$. Since $N$ is symmetric and $n_3$ is a unit, comparing the last rows shows that $r\equiv n/n_3$. Hence \begin{equation}\label{eq:rank1} N\equiv\frac{nn^{\mathsf T}}{n_3}\pmod{S^4},\qquad n\cdot\hat U\equiv n\cdot\hat V\equiv0\pmod{S^4}. \end{equation} The entries $(1,1)$, $(1,2)$ and $(2,2)$ of \eqref{eq:rank1} are \begin{gather*} \text{(R1)}\ (a+c-1)(a+b+c+d-1)\equiv(a-c)^2,\qquad\text{(R2)}\ (a-c)(d-b)\equiv0,\\ \text{(R3)}\ (b+d-1)(a+b+c+d-1)\equiv(d-b)^2 . \end{gather*} \emph{Order $S$.} For a series $F$ write $F=\sum_jF^{(j)}S^j$, and let $\kappa_e=K_e(0)$, so that $\kappa_e^3=\frac18$. At $S=0$, $\Delta U_{12}=2$, $\Delta U_{34}=-2$, $\Delta V_{23}=2$, $\Delta V_{41}=-2$, and $\Delta V_{12}=\Delta V_{34}=\Delta U_{23}=\Delta U_{41}=0$. The order-$S$ part of the side rows in \eqref{eq:even} yields \[ \Delta V_{12}^{(1)}=\frac1{2\kappa_{12}^2},\qquad\Delta V_{34}^{(1)}=-\frac1{2\kappa_{34}^2},\qquad\Delta U_{23}^{(1)}=\frac1{2\kappa_{23}^2},\qquad\Delta U_{41}^{(1)}=-\frac1{2\kappa_{14}^2}. \] At order $S$, (R1) and (R3) give $c^{(1)}=-a^{(1)}$ and $b^{(1)}=-d^{(1)}$, and then $n\cdot\hat U$ and $n\cdot\hat V$ yield $\hat U_3^{(1)}=-4a^{(1)}$ and $\hat V_3^{(1)}=-4d^{(1)}$. Since $\Delta U_{23}+\Delta U_{41}=-2\hat U_3$ and $\Delta V_{12}+\Delta V_{34}=2\hat V_3$, \[ a^{(1)}=\frac{\kappa_{14}^2-\kappa_{23}^2}{16\,\kappa_{14}^2\kappa_{23}^2},\qquad d^{(1)}=\frac{\kappa_{12}^2-\kappa_{34}^2}{16\,\kappa_{12}^2\kappa_{34}^2}. \] (R2) at order $S^2$ gives $4a^{(1)}d^{(1)}=0$. Since $\kappa_e=1/(8\kappa_e^2)$, $a^{(1)}=0$ if and only if $\kappa_{14}=\kappa_{23}$, and $d^{(1)}=0$ if and only if $\kappa_{12}=\kappa_{34}$. \emph{The equality $\kappa_{12}=\kappa_{34}$ forces $\kappa_{14}=\kappa_{23}$.} Suppose that $\kappa_{12}=\kappa_{34}=\kappa$ and $a^{(1)}\ne0$. Then $d^{(1)}=b^{(1)}=0$. Since $a-c=2a^{(1)}S+O(S^2)$ is $S$ times a unit, (R2) gives $d-b\equiv0\pmod{S^3}$. From $\Delta V_{12}=\hat V_1+\hat V_3$ and $\Delta V_{34}=\hat V_3-\hat V_1$ we get $\hat V_1^{(1)}=1/(2\kappa^2)$ and $\hat V_3^{(1)}=0$. At order $S^2$, $n\cdot\hat V\equiv0$ reads $n_1^{(1)}\hat V_1^{(1)}+\hat V_3^{(2)}=0$, since $n_2=O(S^3)$, $n_3^{(1)}=0$ and $\hat V_1(0)=\hat V_3(0)=\hat V_3^{(1)}=0$. Hence $\hat V_3^{(2)}=-a^{(1)}/\kappa^2$. At order $S^2$ the side rows $12$ and $34$ yield \[ \Delta V_{12}^{(2)}=-\frac{\Delta U_{12}^{(1)}}{4\kappa^2}-\frac{K_{12}^{(1)}}{\kappa^3},\qquad\Delta V_{34}^{(2)}=-\frac{\Delta U_{34}^{(1)}}{4\kappa^2}+\frac{K_{34}^{(1)}}{\kappa^3}. \] We add the two, using $\Delta V_{12}+\Delta V_{34}=2\hat V_3$, $\Delta U_{12}+\Delta U_{34}=2\hat U_3$ and $\hat U_3^{(1)}=-4a^{(1)}$, and obtain $K_{12}^{(1)}-K_{34}^{(1)}=4\kappa a^{(1)}$. Hence the coefficient of $S$ in $a-c=4(K_{12}^3-K_{34}^3)$ is $12\kappa^2(K_{12}^{(1)}-K_{34}^{(1)})=48\kappa^3a^{(1)}=6a^{(1)}$. But it is $2a^{(1)}$ by the order-$S$ step. Thus $a^{(1)}=0$, a contradiction. \emph{Conclusion.} The relabelling $(24)$ followed by $\sigma_c$ fixes case~A and the conditions $u_1=v_1=1$, and it exchanges $\kappa_{12}\leftrightarrow\kappa_{14}$ and $\kappa_{34}\leftrightarrow\kappa_{23}$. Applied to its image, the previous step shows that $\kappa_{14}=\kappa_{23}$ forces $\kappa_{12}=\kappa_{34}$. By the order-$S$ step one of the two equalities holds, so both hold. Together with \eqref{eq:O0}, $y_0$ is balanced. \end{proof} \subsection{The branches of the curves}\label{ss:branches} For $\gamma=(\zeta_1,\zeta_2,\varepsilon_1,\varepsilon_2)$, with cube roots of unity $\zeta_1$, $\zeta_2$ and signs $\varepsilon_1$, $\varepsilon_2$, let $Y_\gamma(t)$ be the point with the coordinates \begin{gather*} u=\Big(1,\ \frac{t-1}{t+1},\ -1,\ -\frac{t-1}{t+1}\Big),\qquad v=(1,\ \varrho,\ -1,\ -\varrho),\qquad\varrho=\frac{\zeta_2-\zeta_1t}{\zeta_2+\zeta_1t},\\ k_{12}=k_{34}=\frac{\zeta_1}2,\quad k_{14}=k_{23}=\frac{\zeta_2}2,\\ q_{15}=q_{35}=\varepsilon_1,\quad q_{13}=-\frac{\varepsilon_1}2,\quad q_{25}=q_{45}=\varepsilon_2g,\quad q_{24}=-\frac{\varepsilon_2g}2, \end{gather*} which depends on $t$ and on a further unknown $g$. Let $\Phi(T,y)=(y_u,\ T^{-2}y_v,\ k,\ Tq)$: the point with $u=y_u$, $v=T^{-2}y_v$, $p_e=k_e$ on the sides and $p_e=Tq_e$ on the other pairs. It is the rotation by $T$ of the point with the coordinates $T^{w_q}y_q$, so $G(T,y)=0$ if and only if $\Phi(T,y)\in X_{A'}$. If $T^2=h_1(t)$ and $g^2=h_2(t)/h_1(t)$, then $\Phi(T,Y_\gamma(t))$ is the point of $\cC'(\zeta_1,\zeta_2)$ with $\tau=t$, $D_1=T\varepsilon_1$ and $D_2=T\varepsilon_2g$, by Theorem~\ref{thm:36}(a), and so $G(T,Y_\gamma(t))=0$. Since $h_1(0)=0$ and $h_1'(0)=1/\zeta_2$, the equation $S=h_1(t)$ has a convergent inverse $t(S)=\zeta_2S+O(S^2)$ near $0$. Since $h_2/h_1=-1$ at $t=0$, there is a convergent branch $g(S)$ of $\sqrt{h_2(t(S))/h_1(t(S))}$ with $g(0)=i$. Let $\tilde Y_\gamma(S)$ be $Y_\gamma(t(S))$ with this $g$. Then $\tilde Y_\gamma(T^2)$ is a convergent power series in $T$ with $G(T,\tilde Y_\gamma(T^2))\equiv0$ and $u_1=v_1=1$. Its value at $T=0$ is the case~A leading solution with $k_{12}=k_{34}=\zeta_1/2$, $k_{14}=k_{23}=\zeta_2/2$, $q_{13}=-\varepsilon_1/2$ and $q_{24}=-i\varepsilon_2/2$. The map $\gamma\mapsto\tilde Y_\gamma(0)$ is therefore a bijection onto the $36$ balanced leading solutions of case~A. \begin{lemma}\label{lem:d3b} Let $y$ be a branch for the type $\pc h(5;13|24)$ at $A'$. In case~A, $y(\xi)=\tilde Y_\gamma(\xi^{2L})$ for the $\gamma$ with $\tilde Y_\gamma(0)=y(0)$. In case~B, $y(\xi)$ is the image of such a series under $\sigma_c$. \end{lemma} \begin{proof} By Lemma~\ref{lem:d3a}, $y(0)$ is balanced, so in case~A it is $\tilde Y_\gamma(0)$ for a unique $\gamma$. Both $y$ and $\tilde Y_\gamma(T^2)$ are branches with this leading term, so they agree by Lemma~\ref{lem:lift}. Case~B follows by $\sigma_c$, which maps branches to branches and case~B to case~A. \end{proof} Counterparts of Lemmas~\ref{lem:d3a} and~\ref{lem:d3b} for germs at positive masses are in \cite[\S7]{CT26}. \begin{corollary}\label{cor:d3} Let $C\subseteq Y_A$, $A=(1,4,4,4,4)$, be an irreducible curve with a singular place of diagram $3$. Then $C$ is one of the $36$ curves of $\cC_A$. Conversely, each of the $36$ curves has singular places of diagram $3$. \end{corollary} \begin{proof} We transport $C$ to $Y_{A'}$ by the exchange of bodies $1$ and $5$, and, by the beginning of this section and the relabellings $(23)$ and $(34)$, we may assume that the place $P$ has type $\pc h(5;13|24)$. Let $y$ be a branch at $P$ as in Lemma~\ref{lem:branch}. The point of $C$ near $P$ has $u_1=1$, so it is $\Phi(\xi^L,y(\xi))$. In case~A, Lemma~\ref{lem:d3b} and \S\ref{ss:branches} show that this point lies in $\Lambda(\zeta_1,\zeta_2)$ over $\C((\xi))$. In case~B the same holds, since $\sigma_c\circ\Phi=\Phi\circ\sigma_c$ on branches with $u_1=v_1=1$ and $\sigma_c$ maps $\Lambda(\zeta_1,\zeta_2)$ onto itself (proof of Proposition~\ref{prop:real}(b)). Consequently the affine linear functions that define $\Lambda(\zeta_1,\zeta_2)$ vanish on the expansion of the coordinates of $C$ at $P$. As the change of parameter is an embedding of fields of Laurent series, they vanish near $P$, hence on $C$. Therefore $C\subseteq\cC'(\zeta_1,\zeta_2)$. By Theorem~\ref{thm:36}, $C$ is one of its components. Conversely, on the curves of Theorem~\ref{thm:36} the side values are constant, $p_{13}$, $p_{15}$, $p_{35}$ are multiples of $D_1$, and $p_{24}$, $p_{25}$, $p_{45}$ are multiples of $D_2$. At a place over $\tau=0$ or $\tau=\infty$, $D_1$ and $D_2$ vanish to the first order in a local parameter, by the orders of $h_1$ and $h_2$. The vector $(-\val\,p_{kl})_{k