In SQL databases, NULL represents an unknown value — not the absence of a value. When a value is simply non-applicable for a given entity, the correct relational modeling approach is normalization: the entity gets no row in the relevant table at all, rather than a NULL in a column. This distinction becomes tricky with OUTER JOIN results, where the absence of a row is surfaced as NULL across all columns of the unmatched side, including key columns — making it easy to confuse "unknown value" with "no row existed." MongoDB has its own subtlety: a field can be explicitly set to null or simply not exist in the document at all. In the BSON representation, these are distinct — one is a key with a null-typed value, the other is the absence of the key entirely. The schema is flexible: you can define a field or not. But in indexes, this distinction disappears. Except for partial indexes, indexes must have a key value for every document it covers.…