The problem is not what it looks like. The problem is that I forgot to include hdf5 and gsl as a runtime dependency, so the within-conda version of hdf5 wasn’t where the psi4 conda binary was expecting to find it. The larger problem is that you have to be subscribed to another (non-default, non-psi4) channel (asmeurer) to get a gsl library. For the moment, I’ve pulled out chemps2 again, so the problem you see should be gone.