Robustness and bug fixes. KHARMA should now gracefully handle a wide range of unphysical evolutions of the conserved variables (inevitable in the jets of accretion simulations). Every attempt is made to preserve the momenta, and they can optionally be preserved in all cases with the option floors/use_rho_to_slow. Documentation and paper on the details forthcoming. These improvements should reduce the need for reconnection and cancellation at polar boundaries, but those remain available if you face issues or for better accuracy in simulations which are obviously interacting badly with the polar boundary.
The new recovery code works by far the best with floors/frame = normal, i.e. floors applied in the normal observer frame. If you're specifying frame = drift or something else and seeing instability, try using NOF. I'd consider this something of a regression, except that you can also reduce the floors by a few orders of magnitude when you make the switch, so the drag and other side effects associated with normal observer frame floors should be minimal. If you can't recover acceptable behavior with the normal observer frame, let's discuss how to integrate your scheme of choice into the existing recovery code in a stable way.
True transmitting polar boundary conditions (i.e. via "excision" during the flux calculations, allowing fluxes through the polar faces) and internal static mesh refinement levels are generally working quite well; I know of one production use of each. They are easy to enable to experiment but your mileage may vary, I'm happy to work on them as usage grows.
Resizing simulations should be considered very experimental. It works in tests but seemingly nowhere else. I'll be fixing it in the dev branch ASAP and backporting the fix if possible.