Two problems.

First, the following sentence is clearly broken:

An approach may be to create a unique index on a field (or fields) that should have only unique values (or unique combination of values) prevents duplicate insertions or updates that result in duplicate values.

Second, the entire "document-level atomicity" discussion does not make clear whether said atomicity is at the level of the leaves (the documents embedded within documents) or the roots (the outer, enclosing documents that are direct members of the collection).