A slow query log entry for a DELETE statement showed the following:
We can see that Rows_affected is 0. Supposedly, r_filtered is 55.55%.
The documentation has the following definition of r_filtered:
r_filtered is an observation-based counterpart of the filtered column. It shows which fraction of rows was left after applying the WHERE condition.
If 55.55% of rows were left after applying the WHERE clause, then how were 0 rows affected by the delete? It seems likely that either r_filtered is completely wrong or the definition of the column is wrong.