There was an incorrect assumption that template output would be grouped under a single container node. This is not the case as demonstrated in T364925. In order to find all parts emitted by a template, we need to walk forward from the first node and include anything with a matching about attribute.
Unfortunately, this also invalidates many of our April 1st result columns.
Update the changelog after fixing, identify affected columns, and re-run for dewiki. Discuss whether to continue with a full run.