This is completely untrue. While the index only stores the hashes, the table itself stores the full value and postgres requires both the hash and the full value to match before rejecting the new row. Ie. Duplicate hashes are fine.