Snowflake's loop skews harder than average on coding (graphs, DP, concurrency) and leans heavily on SQL and window functions, plus data-platform system design (micro-partition metadata, columnar storage, query pruning) and a behavioral round centered on ownership.