A coarser aggregate can't add information the finer one lacked, but a denser one can expose signal the sparser one buried

In other words, the sum of information held by granular features, can’t ‘magically grow’ after aggregating them, but this aggregation can surface signal that was hidden in the sparse columns that the model would have to find on its own. It’s basically encoding information that we derived before the model even exists, thus giving the model a good head start.