\section{Introduction} \label{sec:intro} Di\'{o}si~\cite{Diosi1987,Diosi1989} and Penrose~\cite{Penrose1996} proposed that a mass in a superposition of two locations loses coherence at the rate \begin{equation} \Gamma=\frac{C\,\EG}{\hbar},\qquad \EG=\frac{G}{2}\iint\frac{\Delta\rho(\mathbf{x})\,\Delta\rho(\mathbf{y})}{|\mathbf{x}-\mathbf{y}|}\,d^3x\,d^3y, \qquad \Delta\rho=\rho_L-\rho_R , \label{eq:dp-rate} \end{equation} where $\rho_L$ and $\rho_R$ are the mass densities of the two branches and $C$ is a dimensionless coefficient ($C=1$ in Di\'{o}si's normalization). The rate is first order in $G$ and refers to no environment. Reviews of gravitational decoherence~\cite{BassiGrossardtUlbricht2017,AnastopoulosHu2022} survey the models that contain it and the experiments that test it. \paragraph{The question.} The models that contain the rate~\eqref{eq:dp-rate} postulate it. We ask what kind of physical system could produce it, in the simplest case: one body held at rest in a superposition of two positions, so that both branches carry the same mass. \paragraph{Why it is open.} Two candidates present themselves. The first is the quantized gravitational field. Master equations that treat gravitons as an environment have rates of first order in $G$, but they decohere superpositions of different \emph{energies}~\cite{Anastopoulos2013,Blencowe2013,OnigaWang2016}. For a spatial superposition of one body the field gives a bounded, reversible loss of coherence and no rate. The same holds for any bath in a ground state whose response is continuous at zero frequency (Section~\ref{sec:known}; these results are standard~\cite{CallenWelton1951,Kubo1966,Wightman1956,Unruh1995}). The second candidate is a bath that is not in a ground state, for instance one at finite temperature. Dissipative Di\'{o}si--Penrose models~\cite{Bahrami2014,DiBartolomeo2023} postulate such a bath, with a temperature $T_\beta$ as a new parameter, and laboratory data require only $T_\beta>10^{-13}$~K, or $6\times10^{-12}$~K for small smearing lengths~\cite{DiBartolomeoCarlesso2024}. Black-hole horizons are a different case: they decohere superpositions through free, on-shell quanta~\cite{DSW2023,DSW2024local,BiggsMaldacena2024}, which lack noise of Di\'{o}si--Penrose form. \paragraph{The answer.} A bath of the second kind is bounded by celestial mechanics. If it couples to the mass density through a Hamiltonian, the noise that decoheres a body also slows the body when it moves, at a rate fixed by the noise strength $C$, the fluctuation--dissipation temperature $\Teff$ of the bath and the density of the body (Section~\ref{sec:idea}, Theorem~\ref{thm:drag}). The orbital decay of the double pulsar PSR J0737$-$3039A/B then requires \begin{equation} \Teff/C>3.3\times10^{16}~\mathrm{K} \label{eq:headline} \end{equation} (Corollary~\ref{cor:pulsar}), more than $5\times10^{27}$ above the laboratory bound on the linear-friction dissipative Di\'{o}si--Penrose model. The bound leaves the limit $\Teff\to\infty$, noise without back-action~\cite{Diosi1989,TilloyDiosi2016,TilloyDiosi2017}: Di\'{o}si's original model with a smearing length $R_0$. Its rate for a uniform sphere does not depend on $R_0$ at any separation, so a visibility measurement determines $C$ (Sections~\ref{sec:survivor} and~\ref{sec:prediction}). The mechanism, fluctuation--dissipation, is textbook. What is new here is the bound: Theorem~\ref{thm:drag} applied to binary pulsars, and Corollary~\ref{cor:pulsar}. Oppenheim \emph{et al.}\ remark that a friction term in classical--quantum gravity could conflict with the orbital decay of binary pulsars~\cite{Oppenheim2023}; we know of no earlier quantitative bound from orbital decay on a collapse bath. Neutron-star \emph{heating} bounds exist on collapse models~\cite{TilloyStace2019} and on a hybrid classical--quantum model of Newtonian gravity~\cite{Piccione2026}, and binary-pulsar orbital decay has been used to bound the non-conservation of energy--momentum in unimodular gravity~\cite{ChakrabortyJanaMohanty2024}. \paragraph{What is not shown.} The bound is conditional. First, a laboratory rate tests the noise only at wavelengths up to the size of the body. That the noise keeps the Di\'{o}si--Penrose form, stays white and passive, and has one temperature at the kilometre wavelengths a neutron star resolves is an assumption. It holds by construction in a model that postulates its kernel at every scale and damps momentum linearly, as that of Ref.~\cite{DiBartolomeo2023} does. It does not hold for every reciprocal medium: baths that differ between the two scales escape (Section~\ref{sec:gap}). Second, the published dissipative classical-channel models~\cite{DiBartolomeo2021} are not bounded, because their friction does not act on the mean orbit (Section~\ref{sec:models}). Third, Eq.~\eqref{eq:headline} is an order-of-magnitude statement: the stars are modelled as Newtonian uniform spheres, and strong-field corrections of order one are not computed. Fourth, the ground-state results hold at second cumulant, for an external classical source, and for gravity only for the free linearized field; they are not a theorem about full quantum gravity. Finally, nothing here fixes $C$ or $R_0$, or shows that the rate exists.