An approach to evaluating feed/ranking design decisions by tracing a proposed rule all the way through: what mechanics does the rule create → what incentives do those mechanics create for participants → how do people actually act on those incentives → what outcome results.
Mosseri's illustration: a purely chronological feed sounds like it just reflects a stated preference for recency, but tracing it through reveals an incentive for publishers to post as often as possible (e.g., 50x/day) to dominate recency-ordered feeds, degrading the feed for everyone else. The fix isn't to take the stated preference at face value — it's to model second-order incentive effects before shipping, and measure resulting usage/satisfaction over time rather than trusting immediate sentiment.
The "designing a system is like designing a city" framing emphasizes that ranking/feed design is infrastructure design: rules shape behavior at population scale, not just individual experience.
Contrast with Exploitation vs. Exploration Ranking — a different lever within the same design space.