You already know: two-sided matching (the job case), suggestions that can disclose something (the friend suggestions case), cluster-level experiments, and congestion control. This case adds a market where both people must say yes, attention concentrates sharply, and safety is personal: a like matters only if it is returned; the most-liked profiles receive far more interest than they can answer; scammers and harassers target users; and some data is sensitive. A further twist: success makes people leave the app. The design questions: how to apply preferences, how to rank for mutual interest, how to spread attention, and how to keep people safe.
How to read this lesson. No code, on purpose. Numbers marked assumed are assumptions you state out loud.
1. The prompt, and the questions that change the design
A mid-level answer: "predict who each user will swipe right on and show those profiles first". A staff candidate first asks what success is (likes, matches, conversations, meeting), how preferences work (stated, in both directions), how attention is spread, and what safety rules apply. Those answers produce a mutual-outcome objective, stated preferences applied in both directions, attention-aware ranking, and verification, scam detection and protection of sensitive data.
Assumptions you state: