Exploit natural rid order to order by { groupKey, rid } without composite index?

[Timo van Zijll Langhout (Timovzl)]

Automatic indexing on each property is so powerful. I'm wondering if we can leverage this to efficiently fulfill "top N per group" queries without composite indexes, e.g. find all objects with a given groupKey value in order of creation. As an example of a groupKey, imagine a customer ID on invoice objects.

If I'm not mistaken, elements with the same groupKey value should be encountered in rid order, and rid order is practically creation order.

By that logic, ordering by groupKey, rid should require no sorting, even when no composite index exists for the pair. So long as groupKey is indexed, which by default it would be, we should be able to efficiently find the instances matching a given value, and we should encounter them in rid order. I'd reason that translates to ordering by groupKey DESC, rid DESC too, as well as to variants that constrain the rid to a specific range - basically anything you could do with an index on groupKey, rid.

Is this reasoning correct? Does CosmosDB work this way?