Well, it's about the results. If the approach produces bad results, it's bad. Starting from the fiction, i.e. deciding everything that's true in the world first, then trying to design classes (or any other element for that matter) will give you nothing new. You'll just get the same old same old every time. The mechanics might be minor variations on what's gone before, but they generally won't be new. Write new mechanics, then write new fiction to justify those mechanics.
Awesome. Let's do that. I loved that about 4E. Anyone can cast ritual magic. Sounds perfect. Great way to boost up the boring non-casters. Now just strip out rituals from slot-based casting. Love it.
Again, mechanics then fiction. Why? Because, again, if you start from the fiction you established years ago you're artificially restrained to following that fiction, i.e. nothing new gets to happen because it doesn't follow the old fiction. That's bad.
Only if they produce good results. The only thing that matters is the results. Does your approach produce nothing new? Then it's bad.
"But the established fiction..." is just a roundabout appeal to tradition. We can do better.