I need to determine broad outlines ahead of time. The decisions I make on the fly, which happens because I run a sandbox, I'm still basing my decisions on my outline. I don't need to know that the king's guard is named George, all I need to know is that the king is dictatorial and unforgiving but also generous to his supporters and employees. I've chosen to make my world's climate resemble that of earth. There are some areas blighted by magic but it's never changed basic weather patterns.
Sure, but just about every game has such outlines. It's more a question of how and when are they determined. For me... even when I play D&D... a lot is established by the entire group in session zero. This allows me to focus on the areas of interest to the players, and to incorporate their ideas into the setting. For example, if one player wants to play an elf, I'm going to ask him about what elf culture is like and we'll figure it out together. In situations where something is not established about elves... maybe some kind of cultural observance or holiday... I'm going to include that player in coming up with that stuff.
As far as "we're free to establish" that just sounds like you're trying to give collaborative world design an elevated value because it's your preference. My approach is no better or worse than any other, it's just different and one that I've enjoyed as player and GM for decades.
I've never said one is better or worse. In fact, I've been clear to say the opposite. It's all preference.
However, I do believe that one is more concerned with player-driven play than the other, and I think that's clear from the comments here.
I don't "hide" information, I also don't provide information I don't think the characters would have. I'm pretty generous with high level information, people will recognize the symbols of the major religions (even fairly obscure ones if they have training), everyone knows you need fire or acid to kill trolls.
Not providing information you don't think the characters have is very likely equal to hiding information. I say that knowing that there may be times to hide information that are valid no matter what kind of game you're playing. It's not about having NO unknown information... it's how that unknown information is used in play, and how it impacts the players and what they can and can't do.
My point is that the impact such information has should be considered by the GM.
Like take your "king that's dictatorial and unforgiving but also generous". These traits are likely known by the locals... and therefore learned by the PCs. To me, this would be common knowledge that I would provide to the players even if they were newcomers to the town. They could then incorporate that knowledge into their dealings with the king's men.
If it were not shared... it kind of seems like a trap... with the justification "you didn't look into it". But ALL the information the players get must come from the GM. So why not just give the info? Why call gate it behind a roll?
In old school play, that kind of play worked, because the entire game was about skilled play to navigate a dungeon. If that's not what you're doing, then just provide the information and move on with play.
There's no murder mystery if the players know everything I know as GM.
Well, that's debatable... but we had a recent long thread about that, so no need to rehash it here.
But it's not about telling them everything. If a goal of play is to find out who killed X, then of course the players shouldn't know that.
Rather, it seems like proponents of sandbox play don't accept your definition of player-driven play or player agency as any kind of objective truth.
Yes, that is what it seems like! I just think they are wrong!