How Long Should We Preserve Old Data?

When you're dealing with things that affect other people's finances, a lot of the models and predictions I'm making are like, how hot is this battery going to get? It's not that heavily regulated, right? I'm only upset about a week. But in insurance, and that could be an expensive acoustic wire, so it's not like that's not a big concern. And we have more questions. Another question coming in from with Sam, he wants to know your thoughts on data versioning. Is there a general rule as to how long we should preserve old data that generated old model versions? Any thoughts there?