The problem is not only that it could be a local Maxima. It might also measure the wrong metric all together.

Is the platform widely used and the business successful because content makes users happy and improves their lives?

Or is engagement high, because users have become accustomed to whatever a company is feeding them and look for something shallow to fill up their time.

If you only measure engagement as metric, without any deeper reflection and judgment, then heroin is a perfect product in terms of A/B testing.