Hmm I d say better granularity could

Hmm, I'd say better granularity could have great value, for example per index if you're expecting one index per customer: if free users are in one dataset with say 30-day retention, then someone upgrades, do I lose their old data because of the dataset switch, do I copy everything over (probably getting sampled again so not viable imho), or do I write cross-dataset queries (no) just to retain their data for longer?