Assume a table with the following field:
There are n data files, each of which has the statistics min(id) and max(id). ids are positive integers.
Querying by id < 0 would require an O(n) run on all data files in the manifest, querying whether min(id) < 0 < max(id).
Aggregating metrics and/or statistics to the Snapshot level would reduce such scans from O(n) (n being the number of data files) to O(1).