That's actually the best way to create a model that works. The problem is that their models so far don't "predict" the past correctly. So they don't talk about the assumptions they've built in, and stay focused on their empty assertions and ad hominem arguments.
“That’s actually the best way to create a model that works.”
It’s a start, but if the processes underlying the data are not properly understood, then they’ll never do better than to “predict” the past. As you suggest, they’re not even accomplishing that correctly.