As with #934, it would be useful to add one-sided least squares — repeatedly solve for the item matrix using pooled user matrices.