Bill anatomy, order-of-magnitude cuts, engine decisions, and real-time OLAP — the corpus's deepest pool (90 companies).
From Foodpanda's −45% BigQuery bill to a 93% COUNT(DISTINCT) cut with HyperLogLog, from the Redshift→Snowflake→Databricks migration genre to Netflix's trillion-row Druid — warehouse cost and performance is the most universally interviewable material in data engineering. Includes the Data Modeling module (SCD2 mechanics, Picnic's data vault, warehouse-first architecture).
User-facing analytics is a bounded-work problem: the pre-aggregation ladder, upsert hygiene and GC as capacity axes, and fleets where contention — not capacity — sets p99.
Your product team ships merchant analytics: every page load fires four aggregation queries at the OLAP store. In the demo it was 200 milliseconds. At launch traffic the p99 is nine seconds, the on-call is you — and the cluster is 60% idle. Finance approves more hardware; it changes nothing. A store that is mostly idle, serving queries that are mostly slow, refusing to improve with capacity: something other than capacity is the limit. What?
The full week 4 brief is part of LeetData Pro.
