From the linked BF:

Instead of verifying that the aggregate times out as expected, this test would be not only more robust, but also more correct if we verified that the contents of the output collection are what we expect. More precisely, the purpose of this test is to verify that when $merge outputs to the collection that is being aggregated over, it can trigger an infinite loop of updates, so in the control case (i.e. the aggregate which outputs to a different collection), it doesn’t matter whether the aggregate takes under 2500ms or not, rather, it matters that each document has a value of ‘a’ that is double to its original value. This would confirm in the control case that each original document was updated exactly once as expected.