I have fixed a bug in GRHydro's atmosphere handling after reconstruction. After reconstruction, the reconstructed plus and minus face values are tested for whether they drop below atmosphere level and are then reset. In particular, this means that plus and minus face values for cell i don't generally coincide with the minus/plus face values of the corresponding cells i-1 and i+1. If the reconstruction previously found that the Riemann problem was trivial between, say rhominus(i+1) and rhoplus(i), when changing rhoplus(i), it is not anymore! However, the mask "trivial_rp" testing for a trivial Riemann problem is not changed! This leads to inconsistent behavior at the boundary of the atmosphere.
The symptom was a drift in the center of mass of a static and perfectly symmetric TOV star.