without you ever knowing about it, as well as (perhaps) swapping in a cheaper-to-operate model some percentage of the time, perhaps as request loads peak, hoping you’ll just roll the dice and try again.