A useful test is to freeze the list snapshot, record the engine mix, and change only one variable in the next batch. Otherwise a higher submission count can hide the real constraint.
If you adored this write-up and you would such as to obtain additional facts concerning contextual placement quality kindly browse through the site.