Currently, it is possible for a time-series collection to not be clustered by _id, either by the (undocumented) timeseriesBucketsCollectionClusterById server parameter or if using a storage engine that doesn't support clustered indexes, like ephemeralForTest.

When this is the case, any queries on the time field appear to use the _id index, even if there is also an index on the time field. See this test.