If some large number of those peer-reviewed journal articles' conclusions are (or appear to be) undermined by future discoveries having the raw data to confirm or refute such a development taking into consideration assumptions about the data that might have since changed or been proven incorrect seems like a pretty important thing.

What is in your high school geography notes was dictated in part by political decisions about school curriculum - you cannot assume that the future of even high school geography notes will match your experience.

That is a bit like asking what you need source code for after having compiled the binary program.

Just like you have new versions of a program building on changes to source code, you can run newer and improved analysis on the scientific data (whether to see if you get the same result, a new additional insight or whatever).

Journal papers are often no better than having the binary (or perhaps more accurately the stdout dump of running the binary...)

Ah. Never thought of it that way, makes perfect sense. Thanks for the analogy.

Why would you not? With storage costs being as minimal as they are I can't see a good argument not to store data that's under threat of deletion.

A contrived analogy for this might be something like "Why wouldn't you delete the data once you've rendered a graph of it in PNG?". Data is always important. Just as much so as it's original interpretation and constraints.

For [climate] models for example.