that is just framing. What’s at stake - the safe is keeping me from seeing what’s inside. Why do I try to crack the safe - To see what is inside. I’ve just defined the safe example as a conflict resolution without needing to resolve an external conflict.
Take a look at the Paragon system to see how they differ. With conflict-resolution, we're bound by the outcome, but not the fiction getting there.
Outcome resolution: Success = I find the dirt, failure = I don't find the dirt.
Possible narration: Success = I crack the safe and grabbed the documents inside, failure = I crack the safe but it's empty.
Benefit: resolving outcomes makes that binding on everyone. Disadvantage: some folk aren't comfortable with leaving how we got there in doubt - possibly leading to narrations we find jarring (maybe they dislike me saying I cracked the safe even though I failed my roll).
With task-resolution, we're bound by the fiction getting there, but not the outcome.
Task resolution: Success = I crack the safe, failure = I don't crack the safe
Possible outcome: the dirt is in the safe (could be the dirt isn't in the safe, even on success)
Benefit: resolving how we got there makes that binding on everyone. Disadvantage: some folk aren't comfortable letting someone else say what we found when we arrived - possibly leading to frustration even on a total success.
You raise the question of - just how far out can and must an outcome be? Can it be as near as - I want to crack the safe to see what's inside? Or as far as - the super-villain is defeated, her plans laid to ruin and and she forever incarcerated!? For me, questions like that indicate that regardless of how we structure play, we're still reliant on linguistic, cultural, social and perhaps other norms to judge what is fitting.
Conflict resolution: What is a legitimate outcome from this fictional position?
Task resolution: What outcomes can this fictional performance legitimate?
The route to bringing these together is - in short - declare your fictional-performance==outcome pairings up front. This is achieved in D&D by following the DMG237 rules for ability checks, and in PbtA through moves. The difference is that in D&D, DM assembles fictional-performance==outcome pairings either in their prep or on the fly, while in PbtA game designer preps fictional-performance==outcome pairings during design.
You don't need to resolve both fictional-performance and outcome: pick one. In PbtA, performance is not in doubt: roll indexes outcome. In D&D, roll indexes performance: DM tells you what outcome attaches to each. This has ramifications for design and for how the game may be played - the kinds of things going on in mechanics and modes of play the text will have clearest utility to.
I don't know if that all implies we could have a principle like this:
It isn't fun for a single player to assert both the fictional performance and the outcome associated with it. There ought to be resolution or someone else involved, on one side or the other.