What is the hybrid approach in MongoDB?

Normalization stores each fact once and references other collections (join on read). Denormalization copies data where you need it (no join). Normalize for data that changes often; denormalize for high-traffic reads of rarely-changing data.

When data rarely changes (author name on a message), reads are high-traffic and slow with joins, you need fast reads for UX, and the duplicated data is small. MongoDB rewards denormalization for high-traffic reads.

When data changes often (user profile, prices), you need consistency (one copy of each fact), reads are not high-traffic, or the referenced document is large. Use refs and populate.