D&D (and most other non-"storygame" aka "Traditional" RPGs) don't even go that far.
Most of the "No" in D&D GMing isn't even "off the reservation" play. It's narrating past the attempt, then failing the roll. I strongly encourage players to stop just before the point where an effect is expected, then go to the roll. And, from there, on a success, they or I can narrate the reasonable action.
Long ago (40 years or so), I realized that it was better to tell players to "Tell me what your character is attempting to do" instead of "Tell me what your character does." Less player dissatisfaction, because expectations are better managed.
There's an interesting reflection on that in the Burning Wheel Codex, that I recently had my attention drawn to
When adjudicating the result of a failed test, we find it’s a useful exercise to look at the intent and task separately. You wanted to assassinate the duchess quietly in a crowded room and implicate your enemy? A daring plan. The intent is to lay the blame of the duchess’s murder upon your enemy’s shoulders. The task is her secret murder by stiletto. The obvious failure result is that you do not murder your target and are caught. However, that’s a bit of a roadblock and it leads to rather trite situations that lack nuance—fighting your way out or pleading innocence. A less obvious option has you succeed in the assassination, but you are caught. It’s certainly much harder to plead innocence when the blood is running down your knife hand. In this circumstance, the action is reframed: The duchess is dead and everyone knows you’re guilty. A still less obvious and more devious turn has you succeed, but your dear friend is wrongly blamed for the act and carted off to the gallows. Your actions have dragged an innocent soul down. What will you do?
Which direction to choose? Each is serviceable. The key is in the character’s [motivating] Beliefs (or to a lesser extent, an Instinct). Which option presents an interesting and difficult choice for the player based on his Beliefs? If the player has a Belief about never fighting in the open, then the first option is quite valid. If the player has a Belief about maintaining his pristine reputation, then the second option is a strong challenge to that Belief. If the player has a Belief about using the duchess’s murder as pretext to propel his friend to power, then a twist like the third suggestionisn’t an option—it’s a necessity.
One area for doubt is how far out to bound intent? In the duchess example, the intent was not solely to murder the duches, but to do so without being noticed. But if it's just intent, could the player have gone even further, e.g. they also intended to propel their friend to power... so why not that? It seems too far, but it's hard to articulate the rule or principle that says precisely why.
In D&D-ish play styles, it might be enought to encourage players to describe the desired effects of their action along with their approach. Focus on an action's effects should work to constrain any 'distance to intent' issues that might otherwise arise. In the duchess example, player would say that they want the effects of their action to be that the duchess is murdered and no one notices it was them.
I can think of two ways to manage that mechanically.
A) Combat is not occurring because the duchess has no meaningful defence, so GM does not call for initiative. It's certainly consequential, so they call for Dexterity (Sleight of Hand) DC 25 (it requires a subtle yet lethal stab) with disadvantage (the room is crowded). That seems at least very hard to me. A tier-2 rogue could have an odds-on chance of pulling this off.
B) Initiate combat just between duchess and character, giving the latter advantage on initiative and calling for an attack roll and damage as usual. An assassin (sub-class) would have a fair chance of murdering a defenceless duchess, but bad rolls for initiative, the attack, or the damage, could all send their plan awry. GM also calls for a Dexterity (Sleight of Hand) check for this to go unnoticed. A caution here is that the cumulative probability may be awkward to assess at the table and might give the assassin unintentionally terrible odds, although my guess is that any tier-2 or higher character attempting this will have a fair chance of pulling it off.
Both approaches contain the possibilities illustrated above. Following B) the character could very well murder the duchess (wins initiative, hits, damage is sufficient) but fail to do so without being noticed (flubs the Sleight of Hand). Following A) GM can narrate that same outcome.