When one uses the evolution method ADMBASE::evolution_method = "static" (which is very tempting e.g. for Cowling), then no boundary conditions are applied. This means that the ADMBase variables remain unset on the boundaries after regridding, leading to all sorts of (rather unexpected) problems.
The easy solution is to use thorn Exact instead, which applies an exact solution at each time step, including on the boundary.
I suggest to remove the ADMBase functionality to provide a "static" evolution method, since this evolution method does not know which boundary condition to apply, and hence is bound to fail in non-trivial situations. Alternatively, ADMBase needs to synchronise in postregrid (and probably a few other bins as well). ADMBase should then also offer to apply an outer boundary condition, in this case probably a Minkowski Dirichlet boundary condition or a von Neumann condition.