:) I may be wrong, but it's because my intuition suggests that hash join 80046 rows from COWS_TCUE_STS_SCHEDULE with 24M rows from COWS_TCUE_STP_PROG_IDX01 and hash join 680K rows with 10M rows from COWS_PURPLE_PART have to produce not too much rows than greatest number of rows from related rowsource.