You end up with a lot of potential options. To update everything you could end date everything and start sate new records
Later than the end date of the last thing. Some of the Esri feature service and app capabilities let you play data through time using this sort of method, so you can see changes in conditions or measurements over time.
It gets complex when you compare multiple "related" historical things, were they always related? What if somebody finds a record from the past that should have been a different value? You can deal with all this but you have to organize virtual history, real history, and reconcile why things changed over time, and how actual changes factor in to predictive analytics vs virtual changes.
An example we deal with a lot is crashes at an intersection, then the intersection is improved, then the factors contributing to crashes can really change significantly because of real changes. Say we change the name of the highway at the intersection, that doesnt change the actual history, just the relationship of the history to what the name was at the time vs now.