You already know: learning to rank and position bias (the e-commerce and ad cases), query understanding (the autocomplete case), and fraud and integrity layers (the spam and fraud cases). This case adds search in a marketplace where ranking is worth money: developers compete for the top slot, some stuff keywords, buy fake reviews or run install farms, and some imitate famous brands. Half of searches name a specific app, so the first result is the whole product. The design questions: how to understand intent, what to learn from, how to resist gaming, and how to keep ads honest.
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: "text relevance plus popularity, trained on installs". A staff candidate first asks what people search for (brand names, categories, features), what makes an app good (installs or installs people keep), who tries to game the results, and how ads appear. Those answers produce intent detection with verified developers for brand queries, labels of installs kept after a week, integrity checks against stuffing, fake reviews and copycats, and ads in labelled slots.
Assumptions you state: