My experience of this (in Rise of the Runelords) was that it really sucked, because it broke my immersion. The adventure listed all these goblin tribes, exposition NPC talks about them, I say "Let's investigate Tribe X", GM says "No, only Tribe Y is detailed in the adventure". That sucked.
Would it have been better if the GM had just used the descriptions about Y to handle your investigation of X (taking a punt that you would never bother coming back to the "real" Y, or at least giving the GM time to come up with new details for Y)?
If the players make an uninformed choice, then I see no difference as to whether the DM placed the tower there before the game began, during game play because they thought it was cool, or rolled it randomly on a wildnerness encounter table.
While I'm generally sympathetic to [MENTION=463]S'mon[/MENTION] and [MENTION=90370]Zak S[/MENTION] saying that giving illusory choices is a waste of time, I don't think it always is.
I think a lot of these illusory choices are just about creating a bit of colour. By choosing the desert or the forest, the players choose some colour. If the GM is any good, this also means that perhaps the dragon at the bottom of the dungeon gets changed from blue to green, or vice versa. If the PCs choose to go north then the weather at the dungeon is cold (assuming the typical northern hemispheric gameworld); if they choose to go south then the weather is warm.
Contributing to colour in this way isn't the greatest expression of player agency by any means, but I can see why some RPGers wouldn't regard it as a complete was of time. (I think this is at least part of what [MENTION=6801286]Imaculata[/MENTION] is trying to get at in this thread.)
IMC the PCs didn't chase some bandits, which would have led to Dungeon 1. So I decided to use Dungeon 1 in a different locale instead, and the bandit cave would then be a different dungeon.
I dunno what's so bad about giving the PCs a chance to explore Dungeon 1 by having it be one of the ones the PCs become aware of. I only have so many detailed dungeons, after all. I don't *force* them to explore Dungeon 1 (and in fact tonight they decided to explore Dungeon 2 instead), I just used it as a possible choice.
To me, this raises the question: what is the point of prepared backstory?
By "prepared backstory" I don't just mean prepared material (like a map, some monster statblocks, an NPC with a bit of a biography, etc). The reason for prepared material is mostly to save time at the table. But prepared material doesn't mean prepared backstory - Dungeon 1 can serve as a bandit lair, or something else, for instance.
Prepared backstory means authoring the gameworld in advance - Dungeon 1 is here, the bandits lair in it, is has these other inhabitants, this connection to gameworld history, etc. What is the point of that?
In Gygaxian D&D the answer is easy: collecting information (by way of rumours, detection magic, etc) is part of skilled play; but information is only reliable if the backstory is locked in. Furthermore, locking in backstory allows the players to use information to gain tactical advantages. (Eg the GM just can't author in additional reinforcements to negate the PCs' clever ambush following their use of an ESP spell.)
But once we get to more contemporary, "story"-style RPGing it's less clear to me what prepared backstory is for at all. In a sandbox it creates a backdrop against which the players make choices, but in the absence of player information the connection between those choices and the backstory is something that only the GM is aware of. Perhaps players are still expected to gain information, but given that most detection spells in the game are still set up with dungeon-style ranges/AoEs I'm not really sure how that's meant to work. It's also hard to detail a whole world in the level of detail that a Gygaxian dungeon relies upon if it is to interact properly with the mechanics (eg detection spells, rumours) that players use to get the information they need.
In your (S'mon's) case, for instance, it doesn't seem that the players acquired any information about the connection between the bandits and Dungeon 1, or were even expected to. So I don't see that would be any special virtue in sticking to your original plan for Dungeon 1, rather than changing it up for the reason that you only have so much prepared material.
Hinging the plot on something like meeting the wizard is perfectly fine, so long as you don't force the players to enter the wizard's tower. If they see the tower and the plot calls for them to enter and foil the wizard who is trying to create a particularly nasty monster, and they ignore the tower and continue on, so be it. The plot continues on without them and the monster is created and ravishes the countryside, possibly killing people that the PCs know and love or possibly not, depending on where the tower is. If the PCs go far enough, they may not hear about what is happening. If they are close enough, stories of the monster will reach them via rumor.
The key is to never force a plot, ANY plot on the PCs. Do that and you are not railroading. Don't do that and you are.
The key to what?
Given my desires as an RPGer, the key is to force the players to make choices, by framing their PCs into difficult situations. I don't know if you count that as "forcing a plot" or not (I'm not sure what you mean by "plot"). It's not forcing
outcomes. But it is forcing
situations.
In my most recent 4e session, for instance, when the players escaped Thanatos flying on their chaos skiff to The Barrens, when they arrived at their destination they were confronted by a choice: Oublivae, the demon queen of ruin, wanting to bargain information in exchange for their skiff. Had they not travelled to The Barrens but somewhere else, they would have been confronted by some different situation (appropriate to their destination) but it would still have been something that forced them to make a choice.