\section{The idea: noise that decoheres a body also drags it} \label{sec:idea} \paragraph{A rate needs noise at zero frequency.} A body in a superposition of two positions loses coherence when something acts differently on the two positions and fluctuates. Let that something be a field $\Phi$ that enters the Hamiltonian of the body through $\int\rho\,\Phi\,d^3x$, as a Newtonian potential does. A fluctuation of $\Phi$ turns the relative phase of the two branches. Fast fluctuations average out. The phase wanders steadily, and the interference fringes fade at a constant rate, only if the fluctuations have power at zero frequency. The Di\'{o}si--Penrose rate corresponds to one such noise: white in time, with the spatial correlation $C\hbar G/|\mathbf{x}-\mathbf{y}|$. \paragraph{A bath in its ground state has none.} A bath in its ground state can take energy from a probe but cannot give any. Its fluctuations at a frequency $\omega$ are then exactly as strong as its ability to absorb at $\omega$. Absorption is odd in $\omega$, so if it is continuous it vanishes at $\omega=0$. Such a bath dresses the two branches differently, which costs a fixed, small amount of coherence, and it produces no rate. The quantized gravitational field in its vacuum is of this kind (Section~\ref{sec:known}). \paragraph{A bath that is not in its ground state can have the noise, and then it drags.} A Hamiltonian coupling acts in both directions: if the field pushes on the mass, the mass pushes on the field. A body that moves through the bath stirs it, the stirred bath pulls back, and the body slows. Einstein's relation for a grain suspended in water expresses this~\cite{Kubo1966}. The molecular impacts that make the grain jitter are the same impacts that resist its motion, and the ratio of the two effects is the temperature: $D_p=M\gamma\,\kB T$, where $D_p$ is the momentum diffusion constant of the grain, $M$ its mass, $\gamma$ its friction rate and $\kB$ Boltzmann's constant. A hot bath jitters much for the friction it exerts, and a cold bath drags much for its jitter. In our problem the jitter is given: it is the Di\'{o}si--Penrose noise, of strength $C$. The friction then follows up to one number, the temperature $\Teff$ at which the fluctuations and the dissipation of the bath balance, and it grows as the bath gets colder. Moreover the kicks are gradients of $\Phi$, so the factor $k^2$ of the gradients cancels the $1/k^2$ of the noise spectrum and $D_p$ is proportional to $\int\rho^2d^3x$. The friction rate of a body therefore depends on the body only through its density, \begin{equation} \gamma=\frac{2\pi}{3}\,\frac{C\hbar G}{\kB\Teff}\,\langle\rho\rangle ,\qquad \langle\rho\rangle=\frac{\int\rho^2\,d^3x}{\int\rho\,d^3x}, \label{eq:idea-drag} \end{equation} when the noise has this form at every wavelength the body resolves; Theorem~\ref{thm:drag} gives the lower bound that holds when only the long wavelengths are known. \paragraph{Neutron stars in a binary are the test.} The densest bodies whose motion can be timed are neutron stars, and the best timed are the two stars of the double pulsar. Friction shrinks an orbit and shortens its period. The measured shortening agrees with the emission of gravitational waves to $1.3\times10^{-4}$, which leaves room for a friction rate below $6.1\times10^{-21}~\mathrm{s}^{-1}$. At the density of a neutron star this requires $\Teff/C>3.3\times10^{16}$~K (Corollary~\ref{cor:pulsar}). A bath at $3$~K with Di\'{o}si--Penrose noise would drain the orbit $10^{16}$ times faster than the timing allows. \paragraph{What is left, and where the argument can fail.} As $\Teff\to\infty$ at fixed noise the friction vanishes. The limit is noise without back-action: a classical random field that kicks matter and is not kicked in return. This is how Di\'{o}si wrote his model. There are then three cases. In a ground state there is no rate. At finite temperature there is a rate and a drag, and the pulsars bound it. At infinite temperature there is a rate without drag, and an interferometer must decide (Section~\ref{sec:prediction}). The weak point is the step from the laboratory to the star: an interferometer probes the noise at wavelengths of micrometres, a neutron star at kilometres. A bath whose noise is cut off, slow, or hotter at long wavelengths passes the pulsar test (Section~\ref{sec:gap}).