Hello Greg -
I would think you’d be right about possibly wanting to store values that you want to aggregate at query time as metrics, if you want to save space, and won’t filter on those values (so you don’t need the indexing).
Regarding the other sentence, I recently heard Gian say something about that, and it stuck in my head because it surprised me. If you want to use something like a theta sketch at query time, it might be faster (at query time) if you store a sketch as a metric during ingestion - even if it’s only a sketch for 1 value. I’m not recommending this per se, as I’ve seen some people try this and get pretty bad performance, but I guess in some cases it could help. I don’t remember all the details, but basically, it’s a question of the operations needed to create data sketches from scratch on the entire resultset (if you don’t during ingestion) vs the operations needed to combine all those existing data sketches.
Hope that helps!