It is all handled automatically in a lossless way (unlike Summary Indices). So your only concern, is to not search for (rely on) data that has happened more recently than the max duration of your input + indexer pipeline latency.
That's reassuring. Is there any documentation that shows how this "losslessness" works?
It is because all of the magic happens WHENEVER the event is indexed as PART OF the indexing process. Therefore, it is impossible for the acceleration to be missed, even if event/pipeline latency is very large.