We should also specify users can query passing in an arbitrary model; however, we do not recommend querying this as results look for an exact match where ordering matters.

This query could work with changes in progress for ArrayField + EMF (implementing EmbeddedModelField.get_db_prep_value()) but it currently crashes like bson.errors.InvalidDocument: cannot encode object: <Author: Author object (None)>, of type: <class 'model_fields_.models.Author'> since the model instance isn't transformed to dict. However, I'm very wary of this pattern as I think it will be a huge footgun (the queries work fine initially, then mysteriously break after a schema change... then what is the user to do?).

We should also emphasize that you can query using a dictionary, but we that's not

Customer.objects.filter(address={..."city": "New York"})

I don't think it works. e.g. Book.objects.filter(author={"name": "Shakespeare"}) gives db.model_fields__book.aggregate([{'$match': {'$expr': {'$eq': ['$author', {'name': 'Shakespeare'}]}}}]). I think you're thinking of Djongo's query syntax.