Maybe we are just thinking of different examples in our head of what that means, but applying realism to the setting is in my experience a big expectation in sandbox (one I often push back on because I do think it can lead to play being too dull for me).
But what kind of game doesn't have that expectation? That things make sense within the context of the game world?
This is why I think what we're talking about is the priority of such considerations. I think as you describe sandbox play, realism/consistency is a higher priority for you than gameplay.
Which then leads me to the conclusion that sandbox play is less player-driven than people are stating.
It sounds like more than that to me--like you should be shaping the world in response to what the PCs desire. I.e., if the PCs want to bribe a guard, the GM should make the guard bribable and communicate that to the players because this gives the players the control. Is that characterization wrong?
It's not entirely accurate, no. I'm saying that such a trait need not be decided ahead of time. In the event that it is, then it should be either communicated to the players, or they should be able to discover it.
In the context of the admittedly simple example of the guard... I don't think that the scenario involving the guard is likely so consequential to gameplay that it warrants such detail ahead of play, so I'd leave it up to the dice. Nor does it warrant many steps of interaction... like, gathering information on a single guard? Unless it can be resolved quickly, I don't see the need to spend a lot of table time on this.
In the event it should be important, or perhaps the loyalty of said guards has already been established in play for some reason, then it should be communicated clearly to the players.
If I've understood you right, then I think it should be clear why this makes the resulting experience less of a game, in the same way changing the rules to make rooks move diagonally midway through makes chess less of a game.
I'm advocating for open and clear processes that would prevent the equivalent of a rook moving diagonally.
There's nothing that prevents me from making interesting and engaging scenarios for play. There's no reason that I can't do so without relying on hidden information.
What it boils down to is that you can create situations that have information hidden from the players, or you can share the information with the players. Given those two options, I know which I prefer and which leads to more engaging gameplay.