When an index includes all the fields that a query requires:

coll.createIndex({a: 1, b: 1});
coll.find({a: 5}, {_id: 0, b: 1});

Then we want to generate a plan that only touches the index, with no FETCH / Seek stage. But indexes don't distinguish null and missing, so we do need to check the collection for any index entry where b is null/missing. This is unsatisfying because maybe b is actually never null/missing, in which case we're doing many expensive Seeks unnecessarily.

One solution could be to construct a plan like this:

Union
|   |
|   Scan {a:1, b:1} where a == 5 and b != null
|
Fetch
Scan {a:1, b:1} where a == 5 and b == null

That way we only fetch the documents where b is null/missing.

In this example we would expect the b == null predicate to produce tight index bounds, so if that predicate is always false then the Fetch Scan branch will do very little work.