Ask HN: What to do with historical aggregated data after algorithm change?

[janantala]

Hello there, I would like to hear your advise. What do you do with existing aggregated data in lambda architecture after algorithm change? Lets say when you improve it.

Do you reprocess all historical data with a new algorithm? Do you use versioning for algorithms and computed metrics? What about clients who already used aggregated data in the past?

Thank you!