Option B is correct because it accurately describes Snowflake's two-stage optimization process. First, Snowflake uses micro-partition metadata (min/max values, distinct counts) to prune entire micro-partitions that don't contain relevant data for the query. Second, within the remaining micro-partitions, Snowflake filters rows based on query conditions. This approach minimizes I/O by eliminating unnecessary data reads at both the partition and row levels. Option A is incorrect because pruning happens at the micro-partition level first, not by column. Option C is incorrect because column pruning happens after micro-partition pruning, not before. Option D is incorrect because Snowflake doesn't sort data within micro-partitions by rows during query execution - the data is already stored in sorted columnar format.