It would be self-optimizing if they were just waiting in Whole Foods parking lots, not hanging phones in trees. It's a case of optimizing for the metric, not th

It would be self-optimizing if they were just waiting in Whole Foods parking lots, not hanging phones in trees. It's a case of optimizing for the metric, not the objective, by exploiting the divergence between the two. This is generally pretty distinct from a market self-optimizing. The objective is "closest driver", the metric is "closest phone that connects with Amazon's servers", and the divergence is exploited by this multi-phone syncing/dispatch setup. (In this case, the divergence is unavoidable, short of unspoofable, unremovable microchipping of anyone who wants to pick up gig work) Though it seems like a pretty easy fix: just lower the granularity of the location query.