turns out that's nearly impossible, in most cases (businesses change)
I definitely take a more iterative approach now. There's a short spike window to architect the rough plan, get buy in from other engineers, and as long as we feel like we're directionally going the right way and we're not digging ourselves into a corner, we ship it.
Sometimes that has resulted in redoing things (we made a mistake in our thinking), but those redos are minimal compared to the weeks/months we may have spent over-architecting something
It depends on the whole project and circumstances. Usually I go with specifics and later refactor. The reason is simple: too often did I experience the case, that in order to change something in view, we had to alter the "generic layer". On the other hand, how can you build something generic, when you have not at least 2-3 use cases?
"But we will never refactor" - I am one of the very few, who do just that. I worked my way up from dev to senior manager in order to give people the freedom I always missed.
Getting things to glue together in the right way is a challenge though which is probably why you want it to be data driven. But inevitably you need some flexibility or logic in your data processing so you end up building an expression engine and we get into "creating a inner system/platform effect".