They forgot to run a (same) model to optimize article for reading speed