\documentclass[11pt]{article} \usepackage[margin=1in]{geometry} \usepackage{amsmath,amssymb,amsthm,booktabs} \usepackage{hyperref} \newtheorem{lemma}{Lemma} \newtheorem{proposition}{Proposition} \theoremstyle{remark}\newtheorem{remark}{Remark} \newcommand{\dd}{\mathrm{d}} \title{Can moving the mouths of a self-supporting wormhole make a shortcut?\\ A no-shortcut theorem in a two-dimensional loop model} \author{MengHan Tsai}\date{18 November 2026} \begin{document}\maketitle \begin{abstract} The Morris--Thorne--Yurtsever route to a time machine moves one wormhole mouth so that a twin-paradox lag becomes a time shift between the mouths. For wormholes supported by their own Casimir-like energy (Maldacena--Milekhin--Popov) the static throat is longer than the exterior path; we know of no explicit dynamical calculation for moving mouths. We study a toy model in which the question can be posed exactly: a JT-gravity throat whose two boundaries are transparent to conformal matter that also propagates through a rigid flat exterior. We prove that in this model no timelike motion of the mouths produces a shortcut through the throat in either direction, as long as both mouth clocks keep running. The proof combines (i) a lap identity, by which there is no shortcut iff a Dirichlet operator built from one mouth's Schwarzian energy is non-negative on each lap of a matter stream; (ii) a sum rule equating the averaged null energy along each complete null geodesic of the throat to $-\tfrac\beta2$ times the sum of the clock rates at its ends, which is the dynamical form of a relation of Maldacena and Qi and is not claimed as new; and (iii) a variational step. With the quantum inequality for the streams the same steps exclude a shortcut for any agent acting locally on the streams in the exterior. Two partial lower bounds on the size of the margin are given; no positive floor is proven. Simulations from one code base illustrate the results. The model has a non-gravitating exterior and a classical Schwarzian mode, and nothing is claimed about four dimensions. The derivations, code and text were produced with AI assistance under the author's direction; Section~6 states what was checked and by whom. No physicist has checked this work. \end{abstract} \section{Model} The throat is the global AdS$_2$ strip $\dd s^2=(-\dd t^2+\dd\sigma^2)/\sin^2\sigma$, $0<\sigma<\pi$, in JT gravity~\cite{MQ,MSY}; every null ray crosses it in bulk time $\pi$. The mouths $L,R$ are the two boundaries, with clocks $t_L(T)$, $t_R(T)$ as functions of exterior time $T$. Conformal matter of central charge $c$ passes freely through the boundaries and through a flat exterior segment of light time $d$, forming two chiral streams around the loop, as in the two-dimensional reduction of~\cite{MMP}. We use units $c/24\pi=1$. Each mouth carries the Schwarzian energy \begin{equation} E_i=-\beta\,S_i,\qquad S_i\equiv\{\tan(t_i/2),T\},\qquad \frac{\dd E_i}{\dd T}=T_{\rm in}-T_{\rm out}, \end{equation} where $\beta$ is the renormalized dilaton in these units and the fluxes are measured in exterior time. This is the balance for a mouth at rest; for a moving mouth with rapidity $\eta$ the balance in its proper time $u$ is $\dd E/\dd u=T_{\rm in}-T_{\rm out}+2\eta''$. Each stream is the vacuum of a uniformizing phase $w$ with period $P=2\pi/\kappa$ per lap, so its flux in a coordinate $y$ is $-\{\tan(\kappa w/2),y\}$. The static solution has $t_i=T$, $\beta=\tfrac\pi2(\kappa^2-1)$ and $P=\pi+d$; it exists for $\kappa>1$, which is the statement that the throat is longer than the exterior~\cite{MMP}. The static margin is $m_0=\pi-d=2\pi-P$. A mouth is moved by prescribing its exterior worldline, which reparametrizes the streams by the Doppler map; a ``kick'' does work on the system. A throat ray entering $L$ at $U$ exits $R$ at bulk time $t_L(U)+\pi$. It is a \emph{shortcut} if it exits before the exterior ray that left $L$ at $U$ arrives at $R$. \section{Exact criterion} Let $b(T)$ be the phase of the right-moving stream element emitted from $L$ at $T$, and $F(U)$ the time at which the element emitted at $U$ returns to $L$ after one lap (exterior $L\to R$, then throat $R\to L$), so $b(F(U))=b(U)+P$. \begin{lemma}[Lap identity]\label{lem:lap} The bulk-time margin of the $L\to R$ throat ray entering at $U$ is $\Phi_{LR}(U)=2\pi-[t_L(F(U))-t_L(U)]$. There is no shortcut iff the clock of mouth $L$ advances by at most $2\pi$ during every lap of each stream past $L$. \end{lemma} \begin{proof} The exterior ray from $L$ at $U$ reaches $R$ at bulk time $\tau_R(U)$; the stream element travelling with it crosses the throat and emerges at $L$ at $F(U)$ with $t_L(F(U))=\tau_R(U)+\pi$. Hence $\Phi_{LR}=t_L(U)+\pi-\tau_R(U)=2\pi-[t_L(F(U))-t_L(U)]$. The other stream gives the $R\to L$ rays in the same way. \end{proof} \begin{proposition}[Eigenvalue criterion]\label{prop:crit} Assume $t_L'>0$ and $t_R'>0$ on the lap (a rate reaching zero is closure of the throat, outside this statement), and let $T$ in this proposition denote the proper time of mouth $L$, which is exterior time when $L$ is at rest; for a moving mouth $E_L$ is defined with its proper time, and the lap interval is to be read in that parameter. $\Phi_{LR}(U)\ge0$ iff the operator $-\dd^2/\dd T^2+E_L(T)/2\beta$ with Dirichlet conditions on $[U,F(U)]$ is non-negative, i.e.\ iff \begin{equation}\label{eq:qi} \int_U^{F(U)}E_L\,\psi^2\,\dd T\ \ge\ -2\beta\int_U^{F(U)}\psi'^2\,\dd T\qquad\text{for all }\psi\text{ with }\psi(U)=\psi(F(U))=0 . \end{equation} \end{proposition} \begin{proof} The functions $t_L'^{-1/2}\sin\frac{t_L-t_0}{2}$ solve $\psi''+\tfrac12S_L\psi=0$. The solution vanishing at $U$ vanishes next exactly when $t_L$ has advanced by $2\pi$. By Sturm oscillation theory the Dirichlet operator $-\dd^2/\dd T^2-\tfrac12S_L$ on $[U,F]$ is non-negative iff that solution has no zero in $(U,F)$, i.e.\ iff $t_L(F)-t_L(U)\le2\pi$. Use Lemma~\ref{lem:lap} and $-\tfrac12S_L=E_L/2\beta$. \end{proof} With $w=\psi^2$, \eqref{eq:qi} reads $\int E_L\,w\ge-\tfrac\beta2\int w'^2/w$, which has the form of the Flanagan / Fewster--Hollands inequality~\cite{Flanagan,FH} with $c'/24\pi=\beta$. Since $\beta$ is dimensionful, this comparison is formal. The precise statement is: $E_L/\beta=-\{\tan(t_L/2),T\}$ is the flux of a unit-normalized $\kappa=1$ cylinder vacuum uniformized by $t_L$, and~\eqref{eq:qi} says that a lap is no longer than one period of that vacuum. The stream is the $\kappa$ vacuum whose period is exactly one lap, which is why it saturates its own version (lowest Dirichlet eigenvalue exactly zero). \begin{lemma}[Pointwise sufficient condition]\label{lem:sturm} Let the stream be on the $\kappa$-vacuum orbit, with flux $-\{\tan(\kappa w/2),\cdot\}$. If $E_L(T)/\beta\ge T^{\rm out}_{UU}(T)$ throughout a lap, then $\Phi_{LR}\ge0$ for that lap (and likewise with the incoming flux for $\Phi_{RL}$). \end{lemma} \begin{proof} $T^{\rm out}_{UU}=-S_b$ with $S_b=\{\tan(\kappa b/2),T\}$, and the Dirichlet solution of $\psi''+\tfrac12S_b\psi=0$ from $U$ vanishes next at $F(U)$. If $S_L\le S_b$, Sturm comparison puts the next zero of the $t_L$-solution at or after $F(U)$. \end{proof} Statically the condition is $-\tfrac12\ge-\tfrac{\kappa^2}2$, i.e.\ $\kappa>1$. \begin{remark} Lemma~\ref{lem:lap} and Proposition~\ref{prop:crit} are exact for arbitrary mouth motion and arbitrary matter state; Lemma~\ref{lem:sturm} needs the stream on the vacuum orbit. They are kinematic: they restate the question as a bound on $E_L$ over a lap and do not show that the bound holds. In the equations it can be violated by prescribing a negative-energy-first flux into a mouth; Proposition~\ref{prop:agents} shows that the profiles that do so are not stress profiles of any state. The question addressed below is whether \emph{moving a mouth}, which only Doppler-reparametrizes the streams and does positive work, can violate~\eqref{eq:qi}. Proposition~\ref{prop:noshort} answers this in the negative, as long as the clocks run and the streams are unsourced. \end{remark} \section{The theorem} Use bulk time $t$ as the variable at each mouth and let $q_i(t)=\dd t/\dd u>0$ be the rate of mouth clock $i$ per unit of its proper time. Let $T^+(t)$ be the bulk stress of the stream that enters the throat at $L$ at bulk time $t$, $T^-(t)$ that of the stream arriving at $L$ at $t$, and $\tau^\pm=T^\pm+\tfrac12$. Define \begin{equation} J^\pm(a)=\tfrac12\int_a^{a+2\pi}\tau^\pm(s)\,[1-\cos(s-a)]\,\dd s . \end{equation} $J^-(a)$ refers to the right-moving null geodesic from $(L,a)$ to $(R,a+\pi)$ and $J^+(a)$ to the left-moving one from $(R,a+\pi)$ to $(L,a+2\pi)$. The latter meets the stream element that entered at $s$ at $\sigma=(a+2\pi-s)/2$. The affine measure is $\dd\sigma/\sin^2\sigma$, and the integrand $T_{\lambda\lambda}\,\dd\lambda=T\sin^2\sigma\,\dd s$ carries the weight $\sin^2\sigma=\tfrac12[1-\cos(s-a)]$. So $J^\pm$ is the averaged null energy, with the $\tfrac12$ subtracted, along a complete boundary-to-boundary null geodesic. \begin{lemma}[Linear mouth equations]\label{lem:lin} $\{\tan(t/2),u\}=\tfrac12q^2+qq''-\tfrac12q'^2$ (primes are $\dd/\dd t$), and the mouth equations are \begin{equation} \beta\,(q_L'''+q_L')=\tau^-(t)-\tau^+(t),\qquad \beta\,(q_R'''+q_R')=\tau^+(t-\pi)-\tau^-(t+\pi). \end{equation} \end{lemma} \begin{proof} Writing a dot for $\dd/\dd u$ and a prime for $\dd/\dd t$, $\dot t=q$, $\ddot t=qq'$, $\dddot t=q(q'^2+qq'')$, and one finds $\{t,u\}=qq''-\tfrac12q'^2$, and $\dd S/\dd u=q^2(q'''+q')$. The Schwarzian balance is intrinsic to the boundary: in terms of the \emph{bulk} stresses on the throat side, $\dd E_L/\dd u=q_L^2(T^+-T^-)$, with no further term, for a mouth at rest or moving. The term $2\eta''$ of Section~1 arises only when the same balance is rewritten with the fluxes measured in the exterior frame (if the fluxes are evaluated directly in the proper time $u$ there is no extra term, and adding $2\eta''$ would count it twice): the two streams are related to the mouth's proper time by $\dd U/\dd u=e^{-\eta}$, $\dd V/\dd u=e^{\eta}$, whose Schwarzians differ by $2\eta''$. It is therefore already contained in the equations below and is not dropped. Divide by $q^2$. At $R$ the roles of the streams are exchanged and their arguments shifted by the crossing time $\pi$. \end{proof} \begin{lemma}[Sum rule]\label{lem:sum} Assume (i) the two linear mouth equations of Lemma~\ref{lem:lin} hold, with $q_i$ of class $C^3$, throughout the history up to the anchors, (ii) both clocks exist and are smooth at $a$, $a+\pi$, $a+2\pi$, and (iii) the $SL(2)$ gauge constraint $G^1=G^{\cos}=G^{\sin}=0$ defined in the proof. Then \begin{equation}\label{eq:sum} J^+(a)=-\frac\beta2\big[q_R(a+\pi)+q_L(a+2\pi)\big],\qquad J^-(a)=-\frac\beta2\big[q_L(a)+q_R(a+\pi)\big]. \end{equation} In particular $J^\pm<0$ when the clock rates at the anchors are positive. A throat that was static in the past with $2\beta=\pi(\kappa^2-1)$ satisfies (iii), for any subsequent history of the exterior, as long as the mouth equations hold in between (no closure). \end{lemma} \begin{proof} For $\xi$ with $\xi'''+\xi'=0$ define the boundary charge $Q_i^\xi=\beta[\xi(q_i+q_i'')-\xi'q_i'+\xi''q_i]$; then $\dd Q_i^\xi/\dd t=\xi\,\beta(q_i'''+q_i')$. Take $f\in{\rm span}\{1,\cos t,\sin t\}$ at $L$ and $f_R(t)=f(t+\pi)$ at $R$, and for a slice joining $(L,t_1)$ to $(R,t_2)$ let \[G^f=Q_L^f(t_1)+Q_R^{f_R}(t_2)+\int_{t_2-\pi}^{t_1}\tau^+f\,\dd s+\int_{t_1}^{t_2+\pi}\tau^-f\,\dd s .\] By Lemma~\ref{lem:lin}, $\partial G^f/\partial t_1=f(\tau^--\tau^+)+f(\tau^+-\tau^-)=0$ and $\partial G^f/\partial t_2=f(t_2+\pi)[\tau^+(t_2-\pi)-\tau^-(t_2+\pi)]-\tau^+(t_2-\pi)f(t_2-\pi)+\tau^-(t_2+\pi)f(t_2+\pi)=0$, since $f$ is $2\pi$-periodic. So $G^1,G^{\cos},G^{\sin}$ are three constants, the total $SL(2)$ charges of throat plus streams. For $f_a=1-\cos(t-a)$ one has $f_a=f_a'=0$, $f_a''=1$ at the anchors, so $Q=\beta q$ there. The slice $(L,a+2\pi)$--$(R,a+\pi)$ gives $2J^+(a)+\beta[q_R(a+\pi)+q_L(a+2\pi)]=G^{f_a}$ and the slice $(L,a)$--$(R,a+\pi)$ gives $2J^-(a)+\beta[q_L(a)+q_R(a+\pi)]=G^{f_a}$, with $G^{f_a}=G^1-\cos a\,G^{\cos}-\sin a\,G^{\sin}$, which vanishes under (iii). In the static solution $q=1$, $\tau^\pm=\tfrac12(1-\kappa^2)$: $G^1=2\beta-\pi(\kappa^2-1)=0$ and $G^{\cos}=G^{\sin}=0$. \end{proof} The mouth equations alone give only the conservation of $G$; that $G=0$ is the gauge constraint of JT gravity and must be imposed on the data. Off the constraint the right-hand sides of~\eqref{eq:sum} acquire $\tfrac12G^{f_a}$, e.g.\ the constant $\beta-\tfrac\pi2(\kappa^2-1)$ for a stiffened static throat. Sources acting on the streams in the exterior do not change $G$; an agent term added to a mouth equation does. For the static two-sided solution, \eqref{eq:sum} reduces to eq.~(2.10) of Maldacena and Qi~\cite{MQ}, $-2\phi_r=\int T$ along a null line. The sum rule is that relation with dynamical boundary rates; we do not claim it as new. \begin{remark}[Positioning] With the Poincar\'e-type coordinate $X=-\cot\frac{s-a}2$ one has $\dd X=\dd s/[1-\cos(s-a)]$ and $\{X,s\}=\tfrac12$, so $2J=\int T_{XX}\,\dd X$: the ANEC integral along a complete null geodesic, and \eqref{eq:sum} says that it equals $-\beta(q+q')$ with $q,q'$ the clock rates at its two ends. The logic of this note is therefore the familiar one: negative averaged null energy is allowed on a complete null geodesic only if that geodesic is not achronal (self-consistent achronal ANEC~\cite{GO}), and a wormhole supported this way is longer than the outside~\cite{GJW,MMP}. What is added here is that in this model the achronality step is proved (Rayleigh quotient plus lap identity) instead of assumed, for arbitrary mouth motion, in a form that can be checked exactly. The sum rule itself does not need conformal matter. The constraint $G^f=Q_L^f+Q_R^{f_R}+M^f=0$ uses only that $M^f$ is the Killing charge of the matter and $\dd Q_i^f/\dd t$ the Killing flux through the mouth, and on the null ray $t=a+\sigma$ from $(L,a)$ to $(R,a+\pi)$ the weight of $f_a$ is the affine measure, $\lambda=-\cot\sigma$. Hence, for any matter in the throat whose stress tensor is conserved on the fixed AdS$_2$ background, \[\beta\,[q_L(a)+q_R(a+\pi)]=-\int T_{kk}\,\dd\lambda\] along that ray: a positive sum of the two anchor rates is equivalent to a negative affine ANEC integral across the throat (one clock alone is not controlled by one ray). Conformal free transport enters only afterwards, in the Rayleigh step and the lap identity. For massive or interacting matter of that kind the no-shortcut question in this model is therefore exactly the achronal ANEC question on the complete ray~\cite{GO,Wall}, which we do not answer. The crossing segment is affinely complete in the throat geometry, so it is already a complete null geodesic; achronality is a statement about the glued spacetime (throat against exterior), and exterior matter enters the argument only there. In operator language the left side is the parabolic $SL(2)$ generator $Q^{f_a}=Q^1-\cos a\,Q^{\cos}-\sin a\,Q^{\sin}$ of a mouth evaluated as $\beta q(a)$; a parabolic generator is positive in a positive discrete-series representation, so if the mouth states lie in such representations the positivity of the clock rate is automatic there. The classical identity was checked symbolically; the operator positivity is inferred from the representation content, is not proven for the coupled system, and we draw no conclusion from it. (The general form has two internal derivations in the sense of Section~6. A numerical check exists for conformal matter only: 60 anchors per run at rest after an excursion, residual below $5\times10^{-5}$, $5\times10^{-4}$, $10^{-4}$ for $\kappa=1.2$, $1.5$, $1.05$ with the integral ranging over $0.57$--$3.24$, $2.37$--$6.24$, $0.08$--$1.32$.) It does not extend to matter coupled to the dilaton, such as the $s$-wave reduction of four-dimensional fields: there the background is no longer AdS$_2$ and the charge acquires a history-dependent term of no definite sign, of relative order $\phi/\phi_0$, the order of the other corrections to JT gravity (an estimate, not derived here). \end{remark} \begin{proposition}[No shortcut]\label{prop:noshort} In the model of Section~1, on the constraint surface $G=0$ (in particular for a throat that was static in the past), with the streams unsourced (each stream phase is carried along the light rays around the loop), no timelike motion of the mouths produces a shortcut in either direction while $q_L,q_R>0$. Precisely: the throat ray with anchors $a$, $a+\pi$, $a+2\pi$ is not a shortcut if both clocks exist with positive rates through that window and until the stream element of its lap has returned. Windows reaching past a stopped clock are outside the statement. \end{proposition} \begin{proof} For an unsourced stream $\kappa$ is a conserved monodromy: the phase gains exactly $2\pi/\kappa$ per lap of a ray. The bulk stress of a stream with phase $w(t)$ is $T=-\{\tan(\kappa w/2),t\}$, so $w'^{-1/2}\sin\frac{\kappa(w-w_0)}2$ solves $\psi''-\tfrac12T\psi=0$ and vanishes each time $\kappa w$ advances by $2\pi$. The Rayleigh quotient of $-\dd^2/\dd t^2+\tfrac12T$ on $[a,a+2\pi]$ with the Dirichlet trial function $\sin\frac{t-a}2$ is \[\frac{\int(\tfrac14\cos^2\frac{t-a}2+\tfrac12T\sin^2\frac{t-a}2)\,\dd t}{\int\sin^2\frac{t-a}2\,\dd t}=\frac{J(a)}{2\pi}<0\] by Lemma~\ref{lem:sum}. The lowest Dirichlet eigenvalue is therefore negative, so the solution vanishing at $a$ vanishes again before $a+2\pi$: the stream phase $\kappa w$ advances by $2\pi$ in less than $2\pi$ of bulk time at $L$. By Lemma~\ref{lem:lap}, applied to the stream leaving the throat at $L$ ($J^-$) for $L\to R$ rays and to the stream entering at $L$ ($J^+$) for $R\to L$ rays, $\Phi>0$. \end{proof} \begin{proposition}[Several species]\label{prop:multi} Let species $i=1,\dots,n$ (or a continuum of field lines), with weights $f_i>0$, $\sum f_i=1$, circulate unsourced through the same throat and the same pair of clocks, each along its own exterior path. On the constraint surface $G=0$ and while both clocks run, every throat ray is beaten by at least one exterior route that carries a field (for a continuum, by a set of field lines of positive weight). Hence the throat is never faster than every exterior route. \end{proposition} \begin{proof} The mouth equations are those of Lemma~\ref{lem:lin} with $\tau^\pm=\sum_i\tau_i^\pm$, $\tau_i=T_i+\tfrac12f_i$, so Lemma~\ref{lem:sum} holds for $\sum_iJ_i$. The lap identity holds for each species with the same window $[a,a+2\pi]$, and since $T_i=-f_i\{\tan(\kappa_iw_i/2),t\}$ the Rayleigh quotient of $-\dd^2/\dd t^2+T_i/2f_i$ is $J_i/(2\pi f_i)$. As $\sum_iJ_i<0$, some $J_i<0$, and for that species $\Phi_i>0$. \end{proof} \begin{remark} The statement does not hold species by species: a light species on a short path can have $J_i>0$, and this occurs in the runs. Each lap identity is transport along that species' own path, so paths whose lengths change differently under the motion are covered by the argument, but that case was not simulated. Species confined to the throat (reflecting at the mouths) drop out of the mouth equations but enter $G$ and the sum, and for them $J\ge0$: a confined species completes one lap in exactly $2\pi$ of bulk time, so by the inequality used in Proposition~\ref{prop:agents} the Dirichlet operator $-\dd^2/\dd t^2+T_i/2f_i$ is non-negative on $[a,a+2\pi]$ and its Rayleigh quotient $J_i/(2\pi f_i)$ is non-negative. On the vacuum orbit, $T_i=-f_i\{\tan(\phi/2),t\}$ with $\phi(t+2\pi)=\phi(t)+2\pi$, this needs no quantum inequality: $\phi'^{-1/2}\sin\frac{\phi(t)-\phi(a)}2$ is a zero mode without interior zeros, so the lowest Dirichlet eigenvalue is exactly zero (this short argument is due to one of the external automated checks of Section~6). Hence the circulating species alone have $\sum J_i<0$ and the proposition holds with confined species present. This extension has had less internal checking than the rest of this section. The argument needs one common pair of clocks. It was tested on about 1800 rays with no violation; in the hardest multi-species run the sum rule holds to $1.2\times10^{-3}$. \end{remark} \begin{proposition}[Local exterior agents]\label{prop:agents} Let agents act on the conformal streams inside a region of the exterior (absorbing, delaying, re-emitting, measuring and feeding forward), so that the streams are no longer pure transports of a vacuum phase. Assume: the agents are confined to the exterior and cannot signal faster than light there (they may use the throat); outside the agent region the streams are the chiral conformal fields of the model; the evolution is smooth; and the throat is on the constraint surface $G=0$ with positive margins initially (e.g.\ a static past). Then no throat ray is a shortcut while both clocks run. \end{proposition} \begin{proof} \emph{The inequality.} In units $c/24\pi=1$ the quantum inequality of Flanagan and Fewster--Hollands~\cite{Flanagan,FH} reads $\int T\,g^2\,\dd z\ge-2\int g_z^2\,\dd z$, with $z$ an inertial null coordinate. It is an operator inequality for the chiral stress on an interval, valid in every state in its domain, so it does not matter how the agents prepared the stream. Under a reparametrization $z(x)$ with $g=f\,z'^{1/2}$ one finds $\int g_z^2\,\dd z=\int(f'^2-\tfrac12\{z,x\}f^2)\,\dd x$, while $T_x=z'^2T_z-\{z,x\}$, the same law by which the stream is glued across a mouth; the Schwarzian terms cancel and $\int T_x f^2\,\dd x\ge-2\int f'^2\,\dd x$ in any coordinate along the stream. So the Dirichlet operator $-\dd^2/\dd x^2+\tfrac12T_x$ is non-negative on any interval of the stream that lies on one spacelike curve in the agent-free region. Here $T=\tau-\tfrac12$ is the bulk stress at the mouth, in bulk time. \emph{When it applies to a ray lap.} Let the agent region extend from light time $a$ after $L$ to light time $b$ before $R$, and let $m(U)=B(U)-(U+d)$ be the exterior-time margin of the throat ray leaving $L$ at $U$, with $B(U)$ its arrival time at $R$; $B$ is increasing. The stream passing $L$ during the ray lap that starts at $U$ can be crossed by one spacelike curve in the agent-free region iff $m(U+2a)\ge-2(a+b)$. Since $m(U+2a)\ge m(U)-2a$, this holds whenever $m(U)\ge0$. For the other chirality the roles of $a$ and $b$ are exchanged. \emph{Continuity.} Suppose $m(U)\ge0$. Then the inequality holds on the closed lap, as a limit of sub-intervals. The mouth equations involve only bulk stresses, so Lemma~\ref{lem:sum} is unaffected by the agents, and the Rayleigh quotient on $[a',a'+2\pi]$ is $J^-(a')/2\pi<0$ for every $a'$. If the clock of $L$ gained $2\pi$ or more during the lap, $[t_L(U),t_L(U)+2\pi]$ would lie inside it, a contradiction; so the clock gains strictly less than $2\pi$, i.e.\ $m(U)>0$ by Lemma~\ref{lem:lap}, whose proof uses only the rays. Thus $m\ge0$ implies $m>0$. As $m$ is continuous while both clocks run and positive initially, it has no first zero. The other chirality is treated in the same way with $J^+$. \end{proof} \begin{remark} (a) For unsourced streams this is a second proof of Proposition~\ref{prop:noshort} that does not use the vacuum-orbit form of the stress; the unsourced stream saturates the inequality on a full lap, $T=-\kappa^2/2$ on a lap of length $2\pi/\kappa$. (b) No closed causal curve is assumed; its absence follows. A direct argument by contradiction from $m(U)\le0$ would not do for extended or communicating agents: once a shortcut exists, agents on either side of the free strip could measure early rays of a lap, signal through the throat and act on later rays of the same lap, and the inequality would not constrain that lap. (c) Defining a Hill phase $W$ of any stress profile by $T=-\{\tan(W/2),x\}$, the inequality says that $W$ gains at most $2\pi$ per achronal ray lap. In the equations a source that discards the incoming stream and emits a phase-advanced one does produce a shortcut, but such prescribed profiles are not stress profiles of any state: in 219 source runs (24\,332 ray laps) all 1789 shortcut laps have $W>2\pi$ and none has $W\le2\pi$. For a delay implemented as a transformation of the incoming stream the Hill phase rides the rays, so the gain per lap is exactly $2\pi$ and the bound is saturated; this is a derivation and has not been simulated. What was run are monotone step delays with the discarding source (saturated, no shortcut) and delay bumps with that source, which do violate the bound on the rising part when a ray returns while the source is still on, an artefact of that source model. (d) With inequality-violating driving the classical equations of the model do reach a closed null curve with both clocks still running and only logarithmic back-reaction on the clocks. We read this as saying that in this model all the protection resides in the quantum inequality; it is not a protocol. (e) Covered as agent regions: a causal version of a coupling between the mouths (measure at one, signal through the exterior, act at the other) and a localized non-conformal object in the exterior, static or moving, provided light crosses it no faster than $c$. Excluded by assumption: an instantaneous coupling between the mouths, which is how the deformation of Gao, Jafferis and Wall~\cite{GJW} would appear in this setting (in their construction no exterior connects the two boundaries); agents inside the throat; agents that convert one species into another when the species travel different paths; an exterior with a faster side channel. (f) The achronality condition $m(U+2a)\ge-2(a+b)$ has one derivation and one internal check and no external one; the same holds for the statement that the inequality applies in every state the agents can prepare. This proposition is the part of the note with the least checking. \end{remark} \begin{remark} The statements hold while both clocks run; what happens when a rate reaches zero (closure) is outside them. The theorem gives the sign of the margin; Proposition~\ref{prop:size} below gives partial information on its size. The theorem explains the decoupled-throat control below: for $\beta$ far above its on-shell value, \eqref{eq:sum} has the positive constant $\beta-\tfrac\pi2(\kappa^2-1)$ added, and $J>0$ needs only a slowdown of the clocks of order $\tfrac\pi2(\kappa^2-1)/\beta$. \end{remark} \begin{proposition}[Size of the margin]\label{prop:size} Under the hypotheses of Proposition~\ref{prop:noshort}, let $[a,a+2\pi]$ be a window of bulk time at $L$, $Q=q_L(a)+q_R(a+\pi)$ the sum of the anchor rates, and $W=\kappa w$ the phase of the stream leaving the throat at $L$, as a function of bulk time, so that $T=-\{\tan(W/2),t\}$. \begin{enumerate} \item[(i)] (Two laps.) Let $\Phi_1$ be the margin of the ray lap that starts at $a$ and $\Phi_2$ that of the ray lap that ends at $a+2\pi$. Then \begin{equation}\label{eq:twolap} \Phi_1+\Phi_2\ \ge\ 2\pi-\frac{\pi^{3/2}}{\sqrt{\beta Q}} . \end{equation} \item[(ii)] (One lap.) Let $\Phi$ be the margin of the ray lap that starts at $a$. Then \begin{equation}\label{eq:hill} \Phi\ \ge\ \frac{\beta Q}{(\max W')^{3}}, \end{equation} the maximum being taken over the window. More generally $W$ may be replaced in~\eqref{eq:hill} by $M\circ W$ for any lifted M\"obius map $M$ of the circle $W$ mod $2\pi$ (i.e.\ $\tan(\tilde W/2)$ a real M\"obius function of $\tan(W/2)$ with positive determinant), chosen separately for each window, the maximum again being over the window: \begin{equation}\label{eq:hillinv} \Phi\ \ge\ \frac{\beta Q}{\inf_M\,(\max\,(M\circ W)')^{3}} . \end{equation} \end{enumerate} The same holds for the other chirality with $Q=q_R(a+\pi)+q_L(a+2\pi)$. \end{proposition} \begin{proof} Let $H=-\dd^2/\dd t^2+\tfrac12T$ with Dirichlet conditions and $\lambda_1$ its lowest eigenvalue on the window. The Rayleigh step in the proof of Proposition~\ref{prop:noshort} and Lemma~\ref{lem:sum} give $\lambda_1\le J^-(a)/2\pi=-\beta Q/4\pi$. (i) The two laps are $I_1=[a,a+2\pi-\Phi_1]$ and $I_2=[a+\Phi_2,a+2\pi]$, and $H\ge0$ on each, with a zero mode. If they do not overlap, $\Phi_1+\Phi_2\ge2\pi$ and there is nothing to prove. Otherwise the overlap has length $\ell=2\pi-\Phi_1-\Phi_2$. Let $\theta$ be $0$ to the left of the overlap, $\pi/2$ to its right and linear across it, and $\chi_1=\cos\theta$, $\chi_2=\sin\theta$. For $\psi$ vanishing at the ends of the window, $\chi_i\psi$ is a Dirichlet function on $I_i$, and the localisation identity \[\langle\psi,H\psi\rangle=\sum_i\langle\chi_i\psi,H\chi_i\psi\rangle-\int(\chi_1'^2+\chi_2'^2)\psi^2\,\dd t\ \ge\ -\frac{\pi^2}{4\ell^2}\int\psi^2\,\dd t\] gives $\lambda_1\ge-\pi^2/4\ell^2$. Hence $\beta Q/4\pi\le\pi^2/4\ell^2$, which is~\eqref{eq:twolap}. (ii) Write $\psi=W'^{-1/2}\chi(W)$. Since $\tfrac12T=-\tfrac12\{W,t\}-\tfrac14W'^2$, one finds $\langle\psi,H\psi\rangle=\int(\chi_W^2-\tfrac14\chi^2)\,\dd W$ and $\langle\psi,\psi\rangle=\int\chi^2W'^{-2}\,\dd W$. Over the lap $[a,a+2\pi-\Phi]$ the phase $W$ gains exactly $2\pi$, so over the window it gains $2\pi+\Delta$ with $0<\Delta\le\Phi\max W'$. The Poincar\'e inequality on an interval of length $2\pi+\Delta$ gives \[\langle\psi,H\psi\rangle\ \ge\ -\Big[\frac14-\frac{\pi^2}{(2\pi+\Delta)^2}\Big]\int\chi^2\,\dd W\ \ge\ -\frac{\Delta}{4\pi}\,(\max W')^2\,\langle\psi,\psi\rangle ,\] using $1-(1+x)^{-2}\le2x$ for $x\ge0$. So $\beta Q/4\pi\le-\lambda_1\le\Delta(\max W')^2/4\pi\le\Phi(\max W')^3/4\pi$. For~\eqref{eq:hillinv}: the argument used only that the potential is $-\tfrac12\{\tan(W/2),t\}$, which is unchanged by $M$, and that $W$ gains $2\pi$ on the lap, which is unchanged because $M(W+2\pi)=M(W)+2\pi$. \end{proof} \begin{remark} (a) Static values. In~\eqref{eq:twolap}: $2\pi-\pi/\sqrt{\kappa^2-1}$ against $2m_0=4\pi(\kappa-1)/\kappa$; for $\kappa=1.2$ this is $1.547$ against $2.094$. The bound~\eqref{eq:twolap} is non-trivial only for $\beta Q>\pi/4$ (static throats with $\kappa>1.118$) and is empty near $\kappa=1$. In~\eqref{eq:hill}: $W'=\kappa$ and the bound is $\pi(\kappa^2-1)/\kappa^3$ against $m_0=2\pi(\kappa-1)/\kappa$; the ratio $(\kappa+1)/2\kappa^2$ tends to $1$ as $\kappa\to1$, so~\eqref{eq:hill} is sharp where~\eqref{eq:twolap} is empty. (b) Kinematics behind~\eqref{eq:hill}. Assume in this item that both mouths are at rest in the exterior, so that proper time is exterior time and the exterior light time is the constant $d$. Then $\Phi_{LR}(U)=\pi+t_L(U)-t_R(U+d)$, so $\dd\Phi/\dd U=q_L(U)-q_R(U+d)$ and at an extremum of the margin the rate of $L$ at departure equals the rate of $R$ at the arrival of the exterior ray. The lap map in bulk time is $e(a)=a+2\pi-\Phi(a)$ with $e'(a)=q_R(U+d)/q_L(U)$, and $W(e(a))=W(a)+2\pi$, so $W'(e(a))=W'(a)\,q_L(U)/q_R(U+d)$: along a chain of laps $W'$ is $\kappa$ times the accumulated bunching factor of the rays since the static era. The bound~\eqref{eq:hill} is therefore a statement about clock rates only. (c) Numerical test (one code base, one step size per run): \eqref{eq:twolap} on about $520\,000$ windows ($\kappa=1.1$ to $1.5$, delay kicks and moving mouths), no violation, smallest slack $+0.353$; \eqref{eq:hill} on about $60\,000$ windows of nine moving-mouth runs ($\kappa=1.05$ to $1.5$), no violation, $\Phi$ above the bound by a factor $1.16$ to $1.40$ near static windows and by factors $56$ to $512$ at the smallest margins, where $\max W'$ is $3.9$ to $9.0$ (the largest value in any window of these runs is $18.9$). Part (ii) was also tested as mathematics on 300 random profiles $W'$, with no violation. So~\eqref{eq:hill} is weak at small margins. (d) What is not shown. Neither bound gives a positive floor on the margin for clock rates bounded away from zero: \eqref{eq:twolap} is empty when $\beta Q\le\pi/4$, and in~\eqref{eq:hill} $\max W'$ is not bounded by the rates and depends on the choice of phase, which is why~\eqref{eq:hillinv} is the form to use. A drive at the lap frequency shows the difference: in linear theory the clocks do not respond there, and in a run of 40 laps the margin and the rates stayed flat while $\max W'$ grew from $2$ to $197$, so~\eqref{eq:hill} becomes empty (of order $10^{-7}\beta Q$) without the margin moving. In that run the product of the maximum and minimum of $W'$ stays at $1.34$ to $1.41$ against $\kappa^2=1.44$, i.e.\ the drive acts on the phase as a boost, and~\eqref{eq:hillinv} is estimated at about $0.62\,\beta Q$. The form~\eqref{eq:hillinv} was also tested on 60 random profiles boosted by rapidity up to $3$, with no violation. The frequencies at which the clocks respond resonantly are the real roots of the linear response ($1.563$ for $\kappa=1.2$, symmetric mode); a drive there saturates, with smallest margin $0.79\,m_0$ and slowest clock rate $0.78$. Drives at $2\kappa$ and $3\kappa$ lie close to further roots and end in closure by a stopped clock, with $\Phi\ge0.957\,m_0$ up to that point. These driven runs are single runs (two step sizes for one of them). Even~\eqref{eq:hillinv} does not bound the phase rate by the clock rates. A positive floor on the margin for bounded clock rates is not proven; this is open. The stronger statement $\Phi\ge c_0\,\beta Q$ with a universal constant is false ($Q$ is unbounded while $\Phi<2\pi$). In all our runs small margins occur only together with slow anchor clocks, $\Phi\ge0.51\,m_0\min(q_L(a),q_R(a+\pi))$; this is an observation on twelve undriven runs, and by item~(g) it is a statement in the frame of the static past. In the driven runs, a separate and larger set (about 100 runs), the corresponding statement holds only with the smallest rate over the whole history. We do not conjecture either as a bound. (e) With local exterior agents (Proposition~\ref{prop:agents}) the sum rule is unchanged and the Hill phase of the stress gains at most $2\pi$ on an achronal ray lap, so the gain over the window is still at most $2\pi+\Delta$ and~\eqref{eq:hill} holds with $W$ the Hill phase, and so does~\eqref{eq:hillinv} by the same invariance argument; \eqref{eq:twolap} holds as well, since it uses only $H\ge0$ on the two laps. (f) The two bounds have three internal derivations in the sense of Section~6 and no external check. (g) Frame dependence. The gauged $SL(2)$ acts on the bulk time of both mouths and of the matter together, $t\to M(t)$ with $\tan(M(t)/2)$ a M\"obius function of $\tan(t/2)$. Invariant under it are the sign of $\Phi$ (a causal relation), everything defined by null rays and proper times (the exterior-time margin, the round-trip time through the throat), the Schwarzians $S_i$ and the energies. The clock rates $q_i$, hence $Q$, the phase rate $W'$ and the \emph{value} of $\Phi$ are not invariant. With a static past the solution is unique, and all values of $\Phi$, $q_i$ and $Q$ in this note are in the frame fixed by that past ($t_i=T$ initially). After a disturbance the throat can be left ringing about the static solution in a boosted frame: in a run driven at frequency $1$ at $\kappa=1.2$ (one run) $q_L$ swings between $0.60$ and $1.67$ and $\Phi/m_0$ between $0.61$ and $1.61$, while the exterior-time margin stays within $10^{-4}$ of $m_0$, both Schwarzians within $10^{-3}$ of $\tfrac12$, and the energy is unchanged. Proposition~\ref{prop:size} is covariant and holds in every frame, but a small value of $\Phi$ is not by itself a physical statement. Two exact relations to invariant quantities, for mouths at rest (two internal derivations, no external check): $\Phi_{LR}(U)=\bar q_R\,m_{\rm ext}(U)$, with $m_{\rm ext}$ the exterior-time margin and $\bar q_R$ the mean rate of $R$ between the arrival of the exterior ray and the exit of the throat ray; and the round-trip time through the throat, $T_{\rm rt}=2\pi/\bar q_L$ with the mean over the trip, is the distance between conjugate points of $\psi''+\tfrac12S_L\psi=0$, so that $T_{\rm rt}\ge2\pi/\sqrt{2\max S_L}$ by Sturm comparison. \end{remark} \section{Numerical illustration and checks} All simulations are of the full nonlinear closed loop (both mouths, both streams). The theorem and the propositions do not depend on any run; the runs illustrate the size of the effect and test the formulas. They come from one code base, with a second-order and a fourth-order integrator, and have not been independently re-implemented except where stated. \paragraph{Validity of runs.} A run is accepted if the three constraint charges $G$ of Lemma~\ref{lem:sum}, which the integrator does not enforce, stay below $10^{-3}$. This test was applied to one representative of each class of run quoted in this note, not to every run. They pass with $G$ between $10^{-5}$ and $3\times10^{-4}$, with two qualifications: three outward runs at $\kappa=1.0002$ carry $G$ of $2\times10^{-3}$ to $8\times10^{-3}$ from mid-run and are marked as marginal where quoted, and three runs at $\kappa=1.0002$, $v=0.99$ fail the test and are not used (Section~5). The sum rule~\eqref{eq:sum} holds to $10^{-6}$--$4\times10^{-5}$ on kicked runs ($\kappa=1.02$--$1.2$, $v$ up to $0.9$), on a run with an inequality-violating source and a shortcut, on a closing run, and in its off-shell form on stiffened throats. The identities used in the proofs were also checked symbolically. \paragraph{Clock locking.} A mouth is taken on an excursion and returned, accumulating a twin-paradox lag. For slow motion the residual shift between the mouth clocks is zero to the accuracy of the run ($\kappa=1.2$, lag $0.612$: residual $0.0000$; lags up to $6.9$, exceeding the loop period, were tested), and the margin never falls below the static value by more than the transient ringing. The effect is due to the matter stream: multiplying $\beta$ by a factor $s$ (stiffening the clock against the stream) gives residual shift $0$ for $s=1,10$, a crossover for $s\sim10^2$--$10^4$, and the full lag imprinted for $s=10^9$ (invariant shift $\tau=+0.6123$ from circuit times), where the throat then has a genuine shortcut (minimum margin $-0.27$, stable under refinement). The decoupled case is the Morris--Thorne--Yurtsever behaviour and serves as a positive control. \paragraph{Slow motion, first order in $v$.} For $v\ll1$ there is a closed description (item (i) and the simultaneity argument of (ii) have two internal derivations in the sense of Section~6, item (iii) one; all numbers are from the one code base). (i) \emph{Adiabatic family.} For mouths at rest at separation $x$ the static solution with $\beta$ fixed has both clocks at a rate $q$ determined by \[2\beta q=\pi\Big[\Big(\frac{2\pi}{\pi+qx}\Big)^2-1\Big],\qquad m_{\rm ad}(x)=\frac\pi q-x>0 ,\] which follows from~\eqref{eq:sum} with the lap of bulk length $\pi+qx$; slowly moving mouths follow this family. For $x\to\infty$, $qx\to\pi$ and $m_{\rm ad}\to2\beta$: slow mouths far apart keep a margin $2\beta$, above $m_0$ ($1.382$ against $1.047$ for $\kappa=1.2$). For $x\to0$, $q\to3\pi/2\beta$ and $m_{\rm ad}\to2\beta/3$, so $m_{\rm ad}1$ is off the $SL(2)$ constraint surface, and no perturbative bound follows at the physical ratio $\beta/m_0\approx\tfrac12$. \paragraph{Test of the criterion.} On 960 laps from five runs, the sign of the lowest Dirichlet eigenvalue computed from $E_L(T)$ alone equals the sign of the margin computed from the clocks in all 960, including 15 laps with a genuine shortcut in the decoupled control and laps within $\pm0.03$ of threshold. Independently, for 400 random monotonic clock functions (251 with $\Phi<0$) the two signs agree in every case; this tests Proposition~\ref{prop:crit} as mathematics, independently of the dynamics. \paragraph{Consistency.} The $SL(2)$ charge $Q_0$ of the throat, which is not built into the integrator, is conserved to $1.3\times10^{-4}$ against individual terms of order $1$--$5$. $Q_\pm$ were not checked. \section{Limitations} \begin{enumerate} \item \textbf{Scope of the theorem.} Not covered: an instantaneous coupling between the mouths; agents inside the throat; conversion between species on different paths; evolution after a clock stops (closure); throats or species that do not share one pair of clocks; $1/c$ corrections to the classical Schwarzian mode; a gravitating exterior. The checks that have and have not been made are listed in Section~6. No physicist has checked this work. \item \textbf{No lower bound on the margin.} The infimum of the margin over smooth subluminal trajectories is not known to be positive, and Proposition~\ref{prop:size} gives no floor in terms of the motion; only the theorem (margin $>0$) protects that regime. Exterior-time margins for rays that cross while a mouth moves depend on the clock on which they are read (Section~4); this note does not give all clock readings for every run it quotes. \item \textbf{Runs near the marginal throat.} Three runs at $\kappa=1.0002$, $v=0.99$ gave non-converging negative margins. Using~\eqref{eq:sum} as a diagnostic shows these are integration failures: the formula holds up to a definite time and fails afterwards, and all negative margins lie in the invalid part; in the validated part the margins are about $0.51\,m_0$ or larger. Whether and when those throats close is numerically unresolved. These three runs also fail the constraint test of Section~4 with either integrator ($G$ of $0.1$ to $0.3$), so no number from them is quoted as a result. There $\beta\approx6\times10^{-4}$, outside the semiclassical regime of the model in any case. \item \textbf{Quantum fluctuations of the clocks.} The statements are about the classical Schwarzian mode, i.e.\ expectation values at large $c$. Whether the margin is positive as an operator statement is not shown: for $\beta\gg24\pi/c$ it would be equivalent to an achronal averaged null energy condition on clock-conditioned states, which is unproven, and the classical description first fails for $\kappa-1\lesssim24/c$. \item \textbf{Toy model.} Two dimensions, JT gravity, large $c$; the exterior is rigid and does not gravitate; mouth motion is prescribed. In four dimensions the relevant ray would be a complete achronal null geodesic and the generalized second law applies~\cite{Wall}; nothing here bears on that. \item \textbf{Single code base} for the dynamical runs. This note does not specify the runs (excursion profiles, Doppler map, initial data, the invariant defining the residual shift) in enough detail to be re-implemented from the text alone; an external automated check confirmed this for the erosion runs. No reproducibility appendix is included, and the numerical values of Section~4 other than those of the paragraph on the independent re-implementation should be regarded as unreproduced. \item \textbf{Prior art.} The sum rule is the dynamical form of Maldacena--Qi eq.~(2.10)~\cite{MQ}, and integrated null energy is related to $SL(2)$ charges in the same way in~\cite{MSY} (which we did not study in detail). What may be new is its combination with the lap identity for a stream looping through an exterior. In Maldacena--Qi~\cite{MQ} (Section~4.1 and Appendix~B) the potential is for the common rate of the two boundary clocks; the relative shift is an $SL(2)$ gauge mode without matter and is fixed by the matter charges with matter. The locking found here is that known structure (the relative shift is constrained, not a free modulus) in a new setting with a finite-light-time exterior and moving mouths. The original four-dimensional construction is described as a long wormhole that does not lead to causality violations~\cite{MMP}; see also~\cite{FGM}. On a fixed classical background a potential difference between static mouths produces a growing time shift~\cite{FKZ} (of which we read the abstract, one figure and the discussion only); our runs are the back-reacting counterpart. The use of Hill's equation and Sturm theory for Schwarzian functions is classical, and~\eqref{eq:qi} has the form of known quantum inequalities~\cite{Flanagan,FH}. We did not find the lap criterion, the no-shortcut statement or the dynamical calculation in three keyword searches and in the citation lists we could check; this is not exhaustive. \end{enumerate} \section{Statement on the use of AI and on checking} \textbf{Production.} The derivations, the simulation code, the numerical work and nearly all of the text of this note were produced by an AI system (Claude, Anthropic) under the author's direction. The author is not a professional physicist, chose the question and the direction of the work, and takes responsibility for the note, but has not independently verified the mathematics. \textbf{Internal checks.} The work was organised as several separately prompted instances of the same AI system with different roles: one deriving and running numerics, one refereeing, one writing. An ``internal derivation'' or ``internal check'' in this note means a derivation or check by one such instance; two internal derivations are by two instances. These are not independent in the sense that matters most: the instances share one underlying model and could share its errors. The lap identity, the eigenvalue criterion, the linear mouth equations, the sum rule and the no-shortcut proposition have each been derived or checked by three instances. Proposition~\ref{prop:multi} (in particular the confined species), Proposition~\ref{prop:agents} and Proposition~\ref{prop:size} have had fewer checks, as stated where they appear. \textbf{External automated checks.} Two different AI systems from other providers were given an earlier version of this text, without the code, and asked to re-derive and criticise it. Both reproduced the kernel identity behind Lemma~\ref{lem:sum} and the Rayleigh step. One correctly objected that the sum rule was then stated for general histories but proved only for a static past; the present proof by conserved charges, with the constraint as an explicit hypothesis, is the response. The other wrote its own simulation and reproduced the zero residual clock shift, the accuracy of the sum rule, the conservation of the three $SL(2)$ charges and the shortcut of the decoupled throat, and gave the values compared in Section~4. Everything added after those checks has had internal checks only: the conserved-charge proof of the sum rule, Propositions~\ref{prop:agents} and~\ref{prop:size}, the general-matter form of the sum rule, the slow-motion formulas and the invariant-interval results of Section~4. \textbf{Human checks.} None. No physicist has examined the derivations. The prior-art search was done by the AI system with keyword and citation searches and is not exhaustive. \textbf{Numerics.} All runs except the external re-implementation come from one code base written by the AI system. During the work several numerical artefacts were found in it and corrected (a tabulated trajectory differentiated three times; loss of the constraint at step-size changes), which is why a constraint-based validity test is used; it was applied to representatives of each class of run and not to every run. \begin{thebibliography}{99} \bibitem{MQ} J.~Maldacena and X.-L.~Qi, ``Eternal traversable wormhole,'' arXiv:1804.00491. \bibitem{MSY} J.~Maldacena, D.~Stanford and Z.~Yang, ``Diving into traversable wormholes,'' Fortsch.\ Phys.\ 65 (2017) 1700034, arXiv:1704.05333. \bibitem{MMP} J.~Maldacena, A.~Milekhin and F.~Popov, ``Traversable wormholes in four dimensions,'' Class.\ Quant.\ Grav.\ 40 (2023) 155016, arXiv:1807.04726. \bibitem{GJW} P.~Gao, D.~L.~Jafferis and A.~C.~Wall, ``Traversable wormholes via a double trace deformation,'' JHEP 12 (2017) 151, arXiv:1608.05687. \bibitem{GO} N.~Graham and K.~D.~Olum, ``Achronal averaged null energy condition,'' Phys.\ Rev.\ D 76 (2007) 064001, arXiv:0705.3193. \bibitem{Flanagan} E.~E.~Flanagan, ``Quantum inequalities in two-dimensional Minkowski space-time,'' Phys.\ Rev.\ D 56 (1997) 4922, gr-qc/9706006. \bibitem{FH} C.~J.~Fewster and S.~Hollands, ``Quantum energy inequalities in two-dimensional conformal field theory,'' Rev.\ Math.\ Phys.\ 17 (2005) 577, math-ph/0412028. \bibitem{Wall} A.~C.~Wall, ``Proving the achronal averaged null energy condition from the generalized second law,'' Phys.\ Rev.\ D 81 (2010) 024038, arXiv:0910.5751. \bibitem{FGM} Z.~Fu, B.~Grado-White and D.~Marolf, ``A perturbative perspective on self-supporting wormholes,'' Class.\ Quant.\ Grav.\ 36 (2019) 045006, erratum 36 (2019) 249501, arXiv:1807.07917; ``Traversable asymptotically flat wormholes with short transit times,'' Class.\ Quant.\ Grav.\ 36 (2019) 245018, arXiv:1908.03273. \bibitem{FKZ} V.~P.~Frolov, P.~Krtou\v{s} and A.~Zelnikov, ``Ring wormholes and time machines,'' Phys.\ Rev.\ D 108 (2023) 024034, arXiv:2305.03887. \end{thebibliography} \end{document}