Matt Nathanson, "Romeo and Juliet"
At one point I walked past a table at Dexcon where Chad Underkoffler was running a game of Swashbucklers of the Seven Skies. He had just called for a perception check, and folks had rolled and announced their outcomes. He said what was needed to succeed and then asked, as if it were the most natural thing in the world "Ok, how do you fail to notice the assassins sneaking into the room through the windows?"
Now, I had previous experience with the idea of player's describing their failures, and a great intellectual appreciation for the idea, but Chad laid it out so easily and organically that it's stuck in my head since then as a perfect example of how smoothly the technique could be deployed. At that point I pretty much converted from "this is a good idea" to "this is my default mode" and I haven't looked back. Obviously, there's some refinement that's required from game to game, but the one thing it does that keeps me coming back is that it maintains a level of respect for the character which disarms a lot of common problems.
In my experience, players don't mind the actual act of failing as much as they object to looking stupid (or breaking concept) when they fail. They understand that even very competent character fail in fiction, but they also understand that those failures don't make those heroes look like chumps. Giving the player the chance to describe their failure allows them to save face.
There's a lot of synchronicity between this and the idea of rolling before describing an action (fortune-at-the-beginning, in jargon). The ability to know the outcome lets your characters action sync with that outcome - contrast that with a social roll where the player's speech is excellent, his skill is high and the situation is in his favor, but the dice betray him. Even setting aside the problems that come with the failure, the narration needs to take a fast left turn to explain how Charisma McCharisma just dropped the ball.
All this came up in the comments of a very interesting post about the dice and predictability over at gameplaywright.net. It's worth a read, and it's prettymuch convinced me to try rolling the dice before making any social checks in the future as well as for some other non-combattey rolls. Why? Because the flow of it totally appeals to me. Social rules tend to involve a lot of play and very little in the way of mechanics, even in systems that support social rules, because people like to talk. Talking can take on a life of its own, and can be enjoyable enough that you can reach the point and startle yourself with the reminder that you probably should roll some dice. If the dice then fail to jibe with the direction play was going, it tends to demand something abrupt and drastic to disrupt the flow. That's great once or twice, but do it a few times and it can become a running theme. If the player knows at the outset how its going to end, he still has reason to engage (social contact has enough nuance that you can still benefit even if you don't get what you wanted) but he can also arrange to lose gracefully. As a GM, this also makes my life alot easier, because the alternative is having every NPC be cagey all the time.
This is not a technique for everyone, and as such I would definitely be leery of any attempt to systemize it. You can write a game to work this way, but I'm not sure you benefit from doing so, versus letting the GM choose the style that suits her table. Of course, that level of division of technique from system gets us into Rule Zero very quickly, and that's dangerous ground, and I'm far enough into this as is, so let's call it a wrap there.
1 - The most common objection I've heard to this (or any technique that gives the players narration rights) is that the player may overreach and narrate things which are outside of the scope of the game, either in the form of technicalities (grabbing a torch from a wall when there are not torches) or in a large way ('I fail because I slip on the Million Dollars on the floor'). I have never actually seen the latter happen, but it's wasily enough dealt with by the all purpose, "Dude." The former's trickier, but honestly, it's on the GM's head anyway for failing to describe things as completely as is clearly important to her. My solution tends to be "Unless there's a good reason for there not to be a torch on the wall (or whatever) then sure, of course there is!" and if there is a good reason? A small nudge is usually all it requires.
2 - I have often resisted fortune in the beginning because most of the examples of it I've seen are on a scene level rather than on a task level. On a scene level i find it stifling, but I'm much more comfortable on a task level.
3 - People with MUSH training handle this with well placed ellipses, but that's some esoteric and crazy stuff.
4 - What my brother somewhat brilliantly called '"Roll before Role'"playing'.
5 - But if the conversation was so good, why bother with the dice? It's a good question, and some don't. Heck. sometimes I don't. But it's unfair - it favors the players who are personally engaging without respecting what abilities the characters may have.
6 - I somewhat suspect something like this could also help save the "I Fight Harder" problem that comes out of emulating fighting anime, specifically that they seem to center around protagonists who just need to grit their teeth a little more and spike their hair a little higher to be able to win a losing battle. Depending on players to give up is a losing proposition, unless they've _already lost_.