For practical purposes it is impossible for the GM to decide the state of the world, in all relevant respects, before each action is declared and resolved. Which guards know the password? Is their knowledge out-of-date due to a password change? Where is everyone in the castle? How many of them might recognise a disguised PC? Which accents will they treat as foreign or suspicious?
In my experience, the practical upshot of the sort of approach you advocate - once the fiction becomes richer than the artificially austere environment of a classic D&D dungeon - is that the GM has to make stuff up as they go along prior to actions being declared and resolved, or else introduce new fiction into the situation as part of success or failure narration. The first case puts the shape of the fiction primarily into the hands of the GM. The second case does likewise, unless (and to borrow @Manbearcat's phrase) the GM is on some sort of "budget". A skill challenge is exactly that sort of rationing device. The GM has no more than N moves (and maybe 2 fewer, if the players roll no failures) before they are obliged to narrate a final conclusion to the situation. No more than N troublesome townsfolk, password-forgetting guards, suspicious responses to accents, etc.
A GM can and should attempt to make all decision about the fictional setting without regard to the player's intentions. You're talking as though "this thing cannot be done perfectly" means "therefor, we should do precisely the opposite" follows. Why is it superior that a GM should pick a specific number of obstacles, instead of attempting to determine the situation in absence of the player's input?
* It produces more interesting play, in the sense that the players do not have to try to puzzle out the GM's - perhaps ad hoc - conception of the fictional situation, and instead can follow their own sense of the fiction and (if their checks succeed) make that part of the shared fiction;* It produces more interesting fiction, in the sense of fiction that more closely resembles the inspirational fiction (eg JRRT/LotR, REH/Conan, etc).
I don't think the first point much matters, if the alternative is what you're proposing. I'm arguing the game is improved by giving players agency to affect the number of points of interaction it takes to achieve their goals, "success" in SC parlance, you're suggesting that it's better to set that number beforehand, through an arbitrary system instead of leaving it up to human judgement.
We can certainly both agree a bad GM can cause havoc in either state, from setting up SCs where they don't belong, making them too complex or not complex enough, or in my paradigm, by arbitrarily adding additional complication to the fictional situation after it's been established, so let's assume a talented one. In that case, my paradigm will increase player agency at least some of the time, even if our poor GM occasionally does slip after a string of successes and decide there really should be another guard patrol in this castle in the heat of the moment.
Pretend for a moment that we did have a perfect simulation of a fantasy setting available, to go look at after each declared action, and the goal of our system is emulate that view whenever possible. What will have a higher success rate? A system whereby a human does their best to emulate knowledge of that world, or one where we pick a number ranging ranging from 4-15?
To me, these events seem to more closely evoke the feel of Viking myths and related fairy tales - spying the ox, and then trying to trick the dim-witted giant chief into trading back the horse for his own ox, and then narrowly avoiding being eaten when the deception is seen through - than would happen if the resolution was based on GM notes + discreet action resolution. (What happens if the GM doesn't think to note a barn in the steading, or a giant ox? Why should such an interesting fictional element be contingent on the GM thinking of it?)
I'm fairly willing to concede that if you want to emulate a specific narrative, you can better achieve that by using a closed scene resolution system, because you will have more control over the arc of the action. That's not really relevant to my point about agency, and could very while lie in opposition to the particularly flavor of gameplay enjoyment I'm talking about. I am less interested in having a compelling narrative emerge than in trying routinely to make the best choice in a bunch of strange situations.
Players will still find plenty of obstacles to founder on, even if they manage to ignore/obviate a few. If you're looking for the experience of trying to do your best in a fictional world, it's going to be better if that world rewards you for optimal choices.
And this is FUNDAMENTALLY why we won't agree, cannot agree, because we are not approaching the structure of play in the same way, at all. There is no 'world' and it is NOT being 'simulated' in any meaningful way. We are playing a game which involves telling parts of a story, or several interwoven stories. Nobody owns it, or has total control of what it is. There is certainly no 'deist god'. There are a group of participants who are playing the game and finding out what sort of world they are creating, and what sort of things will happen in it, together. There's no need for the GM to be in some sort of 'drivers seat'.
The "deist" bit is important there. The GM is absolutely not in any kind of driver's seat, they aren't advancing "the narrative" in any meaningful way, outside of hopefully setting up an interesting place to be in, with interesting problems to solve and interesting things to care about. I'm not the viking hat guy, I'm just also not the "story" guy here. Narrative is a retrospective view on the events that occurred, not an intentional act of creation by the group sitting at the table.
But how do you measure when it is no longer possible? Either you have to have closed scene resolution, or the GM invents that answer on their own (or maybe a player gives up or changes there mind I suppose).
This is entirely intrinsic to the player's goals. If your goal was the save the last native badgerhorse and then it dies, the goal is not longer achievable (or maybe it is, if you've got resurrection magic, it's a fantasy game, you might have pretty extreme actions available).
The GM can never be removed. In this classic paradigm the GM IS the world, removing them isn't even remotely possible.
The GM can be quite easily removed from resolution. Actions can just do whatever they say they do, and be absolute descriptors of themselves.
So, you are advocating for a radically different play loop then, where the GM describes an initial situation, players declare actions (I would say here that they would have to declare INTENT, not just what action the character takes), resolution takes place (IE dice or whatever), and then the PLAYER DESCRIBES what the resulting situation is. What if their intent fails (IE the dice go against them?). How does the sequence get reestablished, and how in fact does an overall goal get established to start with if there is no 'closed resolution' of overall (operational level basically) goals? I don't see how such a technique could be architected outside of something like SCs (though obviously something like BitD scores would also work). Honestly you are getting VERY close to the Heroes of Myth & Legend Challenge System here...
No, the description of the world is established by the resolution mechanism. "You jump 30 ft" is quite clear. We know where you started, we know where you landed, we know what kinds of things modify that number from a table of sensible modifiers, it's all self-contained and does not require anyone to adjust the fiction outside of the stated effect. Failure to activate an action (i.e. in the case of skill check with a DC where failure is possible) should be described as part of the action. I imagine for this simple jumping example, you likely jump less far.
"Overall goals" do not need system specification. Players can just want things, and then take actions to try and make them occur. If I want to be somewhere else, I will go look at the movement rules. If I want to disguise my appearance, I will check the rules for creating an effective disguise, use them to measure its effectiveness, and then use the rules for determining whether a disguise is seen through when the PC interacts with anyone else. Resolution does not need to be less atomic, from those specific actions, players can string together plans and desired specific changes in the world.
Really, I think what I'm talking about is not particularly radical. You can just keep the desired outcome in your head, and use it as a guide to determine what action you take next. The whole enjoyment is in picking an effective strategy to make that desired outcome occur.