I'm not so sure that equating 'b' with the 'truncated cube length' is correct.

If you apply an OCTA lattice to a cube of water atoms, you'll likely get large energies; the user must arrange the atoms in the desired shape. The initial script in this post by Lennart (and a simplified followup) demonstrate one approach. I've found that one can also use a sphere of matching volume as a pathway (via minimization) to RHDO and OCTA lattices of water; there's an RHDO example here in the Script Archive.