The resultset logic does not know how to map fields, and relies on an explicit project to perform the mapping:

{ $project: { author: 1, _id: 0, vintage: 1, publishYear: 1, comment: 1, title: 1 } }

For example, the metadata could be dynamically created as results are read. (If the first record has fields A and C, they would be placed in metadata under 0 and 1. Then, if an ensuing record with B is read, this would be placed in metadata under 2.)

Another possibility is to dynamically generate the $project from the entity class.

All project stages would need to be removed from tests.