I've noticed that for sharded collections, we report executionStats under the shards field for each sub-pipeline that gets executed on the individual shards. But we don't have any information about what is happening on mongos during the $mergeCursors or the stages after the $mergeCursors. For example, a $group or a $sort could spend a significant amount of time after the $mergeCursors and we provide no information about that in the explain. There could also be a network latency between the mongos and mongod communication that would be easy to identify if we report executionStats on mongos.