{$group: {_id: "$shardKey", accum: {$bottom: {sortBy: {notShardKey: 1}, output: "$shardKey"}}}} 

results in a COLLSCAN when ran on a collection with a compound index on shardKey and notShardKey. 

It looks like the issue here is that when we check if a sort can be provided by an index, we don't account for the distinct key. Normally the sort spec would have to be a prefix of the index key, but for queries like this, the index key prefix we want is actually `{...distinctKey, ...sortSpec}`.