Currently our indexes use null for both literal null and document doesn't contain such a value. The document has to be fetched to make a distinction between these two things. This means that what should be a simple use of an index turns into scanning many documents. Not needing to fetch documents (for instance, having a

bson type) to determine this would allow for indexes to be used efficiently in both $exists : true queries and other queries when using compound indexes without have to create both sparse and non-spare indexes to service different types of queries.