My best guess is that you have a transposed bvec table.
This format is specified as having three rows, one for each of the i, j, k components of the diffusion sensitisation direction, and one column per volume. In the C++ binaries, as of 3.0_RC1, bvec files that are erroneously transposed can nevertheless be imported. This is however not reflected in dwigradcheck, which is written in Python and does a straight numerical matrix data import. At my end I could modify dwigradcheck to provide the same support, but at your end I would be critical of whatever produced that file; there’s likely no problem, but whatever produced it is nevertheless non-conforming.