Hello Erik,
This is almost possible. For puncture data, one typically evolves lapse, shift, and a quantity B (the time derivative of the shift). For Kerr-Schild data, one also needs to evolve A, the time derivative of the lapse. Otherwise, Kerr-Schild data are not stationary. (One could instead add an offset alpha_0 to the evolution equations for K, but that is quite non-standard and not implemented in McLachlan.)
Ok so it is possible.
Additionally, I find it convenient to smooth all quantities near the singularity, and to choose to advect both lapse and shift. I don't know whether the latter is necessary in theory, but I am always using it.
Concerning the smoothing it seems that some smoothing was necessary to not crash the code (via the Exact epsilon parameter to transform r -> r+epsilon). An alternative may have been to use NoExcision to fill the interior of the AH with smooth data.
I can send a sample parameter file if that helps.
That would be great if you had something that you could send around easily.
Yours, Roland