Snowflake keeps metadata about each micro-partition, including column ranges. A filter can exclude a partition when its range cannot satisfy the condition. The Query Profile reports partitions scanned and partitions total, which lets you distinguish a selective output from a selective scan. Returning one row can still require reading many partitions.
The worksheet displays a local profile after table queries. It groups the fixture rows into sixteen-row partitions and uses a single literal comparison with min/max ranges to decide which groups could match. A clustering key orders that local model by the named columns. This model teaches pruning; its bytes and partition counts are simulated estimates, not measured Snowflake storage or execution time. Try a broad query and a selective range over the same table. Keep result correctness separate from scan estimates, especially for unsupported expressions where the model conservatively scans every partition.