Currently, index bounds checks are done on BSON in both classic and sbe. For both engines this is inefficient. SBE should try to avoid materializing BSON as much as possible. In Classic ixscan, materializing the BSON in an index scan is often unnecessary, since there may be a FETCH that only requires the RecordId

Once bounds checking is done as keystrings, SBE will not have to materialize BSON, and Classic can lazily materialize the BSON. This saves significant work and has shown up to 30% lower latency in benchmarks