This optimization is done for $match on the time field (the value in $match must be of type BSONDate, the optimization doesn't kick in for doubles) and avoids applying the event filter to each measurement in the bucket. The optimization is implemented via an optional state on $_internalUnpackBucket.

According to Arun Banala the optimization improved the eligible queries by up to 30% so we don’t want to lose it but if it ends up tricky to implement in SBE we would need to block these queries from being lowered.