Don’t make it an option.
Wandering the RPG blogosphere and forums, I’ve seen a lot of advice for GMs along the lines of “What to do when the party fails?” The advice goes on to detail some clever or not so clever ways of preventing the module or entire campaign from going down in flames, but it’s almost always from the point of view of picking up the pieces once the party has failed to notice the clue, bypassed the room with the key, alienated the noble who’s the only one with the information they need, and so on. Conspicuously absent, from my point of view, is a discussion of how the GM got the party in that pickle in the first place. And it is the GM that got the party to the point where everything hinged on a single action, make no mistake.
You don’t want a single point of failure in your business processes, and you don’t want it in your RPG scenarios. Unless, that is, you and your players are perfectly happy to fail (a possibility in some challenge-based games).
I’ve talked about “Scenario Breaker Rolls” in the past, so I won’t go into that again, but a botched die-roll isn’t the only way that the party can reach an impasse.
Take NPC interactions. Too many GMs (and I’ve been guilty of this myself) make NPCs basically inert in social situations until the PCs prod them, and then decide how the NPCs react based on the PCs’ approach (plus or minus a die-roll). Then they let the whole game get derailed when the PCs fumble the role-playing part of the interaction, perhaps by offering less deference to the King than the GM thinks the situation warrants. That’s fine if there’s no problem for your game if the NPC declines to offer the quest after the PCs have insulted him and/or stolen from him, or you regard them ending up in the King’s dungeon as a good adventure hook. If failure is going to be a big issue, then, as GM you should be taking charge and making sure that failure doesn’t occur. Don’t make the NPC a surly and suspicious bugger if you need him to trust the PCs, no matter how neat you think it would be if the PCs were able to jolly him around through brilliant roleplaying. Don’t put on your GM stone-face and wait for the PCs to start talking; have the NPC greet them with open arms and move the conversation along to where you need it to be (e.g. at least the announcement of the quest) before they open their mouths. Yes, this steps on the role-playing opportunities of the situation, but you know what? That’s what you get when you make an NPC a plot device. You can have all the other NPCs interact in a more naturalistic fashion, or even that NPC in other situations, but during the portion of the adventure where you need that NPC to convey certain information or offer a particular deal to keep the PCs from hitting a brick wall, it’s a mistake to leave it up to the RP of the players if you think there’s any chance that they’ll screw it up.
Again, let me emphasize that “screw it up” means ruin everybody’s enjoyment with a failure to get the information/come to terms…whenever “failure” can be just as fun and interesting for everybody as success, I strongly encourage GMs to let things fall out however the players direct it. But even GMs who are strongly committed to open-ended games without any rails can reach a point where the decisions of the players to that point have committed them to a course of action, at which point game-breaking opportunity for failures can crop up. My feeling is that unless you and your players are equally committed to challenge-based games, even in an open-ended sandbox campaign it’s the GM’s responsibility to minimize the single-points of failure. If the players have decided to solve a mystery, and successfully uncover the murderer, it’s a mistake to make it so the fact that they’ve antagonized the Chief of Police along the way turns the whole adventure into a failure; you either have to make the Chief honorable enough that given proof he’ll make the arrest anway, or there has to be somebody else they can turn the culprit over to and see justice done.
To sum up, fixing game-breaking errors is no substitute for not making them in the first place. There are all kinds of techniques you can use to recover from error, and it’s good to have some of them in reserve, but your first line of defense should be designing your scenarios so that there just aren’t any places where the PCs could fail unless you are willing for them to fail. You can’t generally make players happy to fail, and you can’t (IMO) make success inevitable without cheapening it, but I think you can and should make every effort so that even if they fail, the players regard it as time well spent; failures should never result in them saying “That was stupid. What a waste of time.” if you can possibly help it.