Conceptual Questions: What is the loss function for logistic regression? Why use logarithmic loss instead of mean squared error? SQL1: Examine the statistics of video upload events from non-US countries within a specific time period. The focus is on SELECT, WHERE, GROUP BY, ORDER BY, and handling the case sensitivity of the `country` field, requiring the use of the UPPER function to ensure correct filtering. SQL2: Count the daily active users in the US region on a specific date, excluding high-risk users. This mainly examines the use of JOIN and the details of COUNT(DISTINCT user_id), avoiding inflated data due to direct COUNT(*). SQL3: Count the number of viewing events from non-US countries today and rank them based on the average of fraudulent sessions over the past 30 days. The difficulty lies in using CTE modularization to handle two independent logics, proficient use of RANK() and OVER() window functions, and offset calculations for date functions such as CURDATE() and INTERVAL.

Walk me through how you'd design an A/B test for a new ranking model on the For You feed, and how you'd pick guardrail metrics.