Redesign Currently traverse is designed to not change traversable object. This is strong assumption, and should be removed. Aggregate functions should not change range (because it makes things complicated). Add clone-and-traverse (like traverse, but will never alter passed range). It is called across remove morep turn xpr into correct forward range specialized RRB-vector! metric-space dictionary metric-space set egnat range… egnat near function egnat grow functions egnat shrink functions make-bucket should create just single value tuple fix clone for RRB range Validate input to make-egnat-tree (scan for duplication) some sort of filter algorithm (aka: without) Use whole content of node as bucket for egnat Programmable layer symbols should be in separate package from API Add having function (for use with group-by). Add function for changing final result of aggregation. Add batches function (divides range into partitions of fixed size for aggregation) Add some sort of data frame. closure compilation Rewrite all multistage aggregations into single stage aggregations so the whole concept can be scrapped started today, finished with math <2020-01-14 wto> finished with clara <2020-01-15 śro> Add additional argument: element-type to the aggregation functions. remove aggregator protocol, design the new protocol based around closures.