You already know: time-series forecasting (D3), streaming aggregation (D16), experiment design and its pitfalls (D10), and the churn case's lesson that actions change the data you learn from. This case adds dynamic pricing in a two-sided marketplace: riders react to price in seconds, drivers in minutes, and every price you set changes what you will observe next. The design questions: what the price is for, how to forecast per cell, how to set prices that stabilise rather than oscillate, and how to experiment when riders and drivers interfere with each other.
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: "when requests exceed drivers, multiply the price by the ratio". A staff candidate first asks what the price is for: reliability (riders get a car within a few minutes), not revenue for its own sake. Then: at what geography and cadence prices change, how riders and drivers respond and with what delay, and what limits apply (caps, emergencies, law, trust). Those answers turn a ratio rule into a forecast-driven controller with smoothing and caps, evaluated with switchback experiments.
Assumptions you state: