For real-time ingestions tasks the data is held in memory and directly queryable as soon as it is consumed from a stream. But like Kyle mentioned, batch jobs are all or nothing, you only see the data, in Native engine queries, when it fully completes the job including the load into historical's local storage.