Context

At Meta, teams running experiments on surfaces like Facebook Feed or Instagram Reels need schemas that support both reliable analysis and efficient SQL querying. A weak design makes it hard to answer basic questions like who was assigned, when they were exposed, and whether they converted.

Core question

Explain how you would design a PostgreSQL schema to track A/B test assignments and outcomes. Your answer should cover:

1. The core tables and keys needed to represent experiments, variants, assignments, exposures, and outcome events
2. How to handle assignment grain, late-arriving events, and users with multiple events
3. How analysts would join the tables to compute experiment metrics such as conversion rate and guardrails
4. What constraints or modeling choices help prevent common data quality issues

Scope guidance

The interviewer expects a practical schema design discussion grounded in SQL analytics. Focus on table structure, join paths, event timestamps, deduplication strategy, and trade-offs between normalized event tables and pre-aggregated outcome tables.