Function refuse_a_range_total_through_a_ranked_index 

pub fn refuse_a_range_total_through_a_ranked_index(
    document_type: DocumentTypeRef<'_>,
    index: &Index,
) -> Result<(), Error>

Refuses a range total (one aggregate over a range, or one per carrier branch) read through index when its path passes through a ranked level: a level whose property-name tree Drive lays out as an indexed tree, whichever of the type’s indexes ranks it (an index may continue below another’s ranked last property). It walks the type’s index structure along the index’s levels and asks each the resolver the write path uses ([property_name_tree_type_and_ranked_axes_for_level]), so it reads the layout Drive builds. grovedb’s range aggregates neither read an indexed tree (AggregateCountOnRange, AggregateSumOnRange and AggregateCountAndSumOnRange take provable trees only) nor descend through one when proving. An unproven read through a ranked ancestor would succeed (it opens only the leaf tree), but it is refused as well: every count, sum and count-and-sum range-total path builder calls this, so the unproven read, the proof and its verification refuse alike rather than answering only without a proof. Grouped by the last property, the same range reads each value.

Unversioned, so every protocol version reaches it: rankings exist only from protocol version 14 (meta-schema v3), so no level is an indexed tree before and it refuses nothing.