I'm on the trail of another new game idea. I've refined things (relatively), and there are some procedural things that I'm sure I'm missing. Again, we're on first draft.
I've been reading through the Forge's researchable depths for games using the Prisoner's Dilemma. Surprisingly, there are only a handful.

Don't you love how spies talk?
It's awesome. It's what I imagine when I roll an intelligence or charisma check.
I'm out to give players the chance to play the role, instead of playing the roll.
The premise of the game is to pit all of the characters in a modified prisoner's dilemma to answer some of these questions: "Can these people fall in love with trust problems?" "Can spies stay in love?" "Is there honor among these thieves?" "How far will family ties stretch?"
Players use dialogue and choices, staking their credibility and relationships for getting the job done.

So, this game is set in a corporate espionage backdrop: two companies are viciously competing over a Product (a GM-defined MacGuffin).
(If you've seen the film Duplicity, move on to the older "Spy Game" and then dip into every John LeCarre book you can find. Those are my inspirations -- in that order.)

Characters are expert spies.
There are no traits (other than the ones the players write down for themselves to remember).
Characters are a combination of a handle (ie: a name) and a voice (ie: a way that their character talks. Attitude.. not accent.)

Play is not exposition narration ("I grab my guns and go to the door.")
Instead, play is In-Character dialogue. ("Did you just lie?" "It's not lying. It's story-telling.")
These plays occur in dialogues. It's just talking, but when is it ever just talk? It's trust-building and information-getting. These are framed sometimes externally, but they have a structure of "entrance, talk, exit"... not unlike a loose improv scene. Acting is optional.

Once players define their name and their attitude, they define an in-character negotiated* relationship with all the other players.
A relationship with other players should be something 'familial': Lover, Marital Spouse, Sibling, Parent/Child, Extended Family, or BFF.
After that definition, players have their first dialogue with each other.
The goal is describing how they first met/how they discovered they're both in spies.
Then, at the dialogue's exit, each player fills in this blank for each of their relationship partners.

"They will lose my trust if I catch them [insert gerund phrase]."
Some examples of gerund phrases:
getting my goat
cheating on me with another man
lying to someone I know and trust
standing on their/my head
burning evidence/my house down
being angry with me for no reason
using torture
letting revenge get the best of them

These are each character's trust issues with the other characters. Each player judges the severity of their trust issue, based on a number 1-5, and give that many "Credibility Notes" (ie: Monopoly money) to the player.
Each player shares their trust issues with the GM.

Now, every player has some history with everyone else. Every player has 'earned' their credibility.
The GM reveals a stack of cards labeled plot cards. Each card describes one stage of the unfolding drama between the rival companies -- and how each player is involved. The GM tells players that there are five possible end-cards that are shuffled into the bottom half of the deck.

In a silent note-passing with the GM, players interact with the first plot card. They may need to spend credibility, based on the GM-specific negotiations on their part of the plot. It's all silent. Then, when everything is settled, the GM plays the NPC of a handful of dialogues, framed by a combination of the trust issues, the plot card and the GM.
Likely, at least one trust issue has been violated in front of the other character/player -- for the sake of the job.

The GM draws the next plot card when all the reveals are done. Not every player gets a reveal.

At this point, players can ask to initiate more history with their characters. If the involved players agree, they determine whether it's present or past, then draw a card out of another available stack. This gives objectives in that scene for each player - and players are encouraged to do "everything" to meet that objective. (If they don't meet it, they can't be awarded credibility. "Everything" can include speaking out of character.)
Players & GM collaborate to frame the scene (spending Credibility for influence over setting/circumstances). Entrance, chat, exit.
Then, the involved players write out another trust issue against each other and reward the appropriate credibility notes.

In this way, play proceeds through 15-20 scenes describing their roles in the rival company's espionage.
The game ends when a player takes advantage of an end-game plot card, when an absolute end-game plot card appears, or when all the players take advantage of an end-game plot card.
When the last reveals are done, the GM narrates a closing on how the companies' little war ended. A final tally of credibility does count toward the pay off.
Then, the players do a dialogue of their characters parting ways.

----

Now.
1) I'm devising most of those plot cards. I do aim to limit the number of players (2-4 + a GM), but has anyone seen something already created that performs the same job in the game. I'd like to be inspired improve an existing system if possible.

2) I'm excited by the idea of dialogue-only play. I've only experienced it once -- and in a very structured D&D forum. I've tried doing RPG searches for that form of play, but it's almost impossible to find anything relevant. Have any of you come across a game using that dialogue-only? Any rules governing it? I'm thinking of including a printed version of the actor's code ('no running over people's lines.' 'no gainsaying what's established.' etc.)

3) I've spent hours reading the Nobody-Gets-Hurt versus I-Will-Not-Abandon-You, and I fear this model has equipped some players with legitimate in-game abandonment. (Seriously, what if your dad does backstab you and runs off to patent the Product? "Well, then he's an arse. Everyone do the 'good game' handshake", right?)

4) I'm still not sure how much I wanna trust that the GMs imagination can do. Yes, it's a ludicrous worry. I have prompts for the players on the table, the cards and other notes to help the GM, and the trust issues to keep everything grounded. Still, wiser folk than I have created better games than mine (read: Ron, Mike, Eero, etc.), and I would like to know if my design has a major hole.

First of all, here's a working link to Prisoner's Dilemma.

Next, yes, this sounds cool. I can't quite tell how you're using Prisoner's Dilemma in your play procedures though. I assume that, somewhere, there's an opportunity (and incentive) for players to betray one another. One thing that makes PD work as a game is that there's no negotiation. Decisions are made blind, or in the case of iterative PD, blind except for past results. That is, even in iterative PD, the players are not discussing things between decisions. How does your game do that? I suppose you don't have to prevent discussion, since the opportunity for betrayal is still there, though.

I assume that Credibility is the currency being wagered in the PD simulation? What payoff matrix are you using? Is it constant throughout play or does it change from scene to scene?

Thanks for the working link! I don't know what I missed in the mark-up...

After spending about an hour with my notes and Game.net's iterative Prisoner's Dilemma, I'm actually convinced that I fail Math..

[sub]In casual usage, the label "prisoner's dilemma" may be applied to situations not strictly matching the formal criteria of the classic or iterative games, for instance, those in which two entities could gain important benefits from cooperating or suffer from the failure to do so, but find it merely difficult or expensive, not necessarily impossible, to coordinate their activities to achieve cooperation. -- Wikipedia[/sub]

I think that may be my only claim to PD.
If you look up the AP on “Swiftly Tilting City,” you’ll see a real use of PD: red stone vs. blue stone.
Me, I’ve just turned all of the kibitzing between plays into untrustworthy communication.
(“So, is he just asking me to get a Coke for him so that he can talk to someone else without me here?”)

Onto the questions I can answer.

Adam wrote:
That is, even in iterative PD, the players are not discussing things between decisions. How does your game do that? I suppose you don't have to prevent discussion, since the opportunity for betrayal is still there, though.

I assume that Credibility is the currency being wagered in the PD simulation? What payoff matrix are you using? Is it constant throughout play or does it change from scene to scene?

... and I just, just finished reading Zero to the Bone. (Like in the last 5 minutes.)

Superficial differences aside, now I'm not sure I can finish this game without someone saying to me (earlier is better than sooner): "You're just banking on that Narr-guy's 'ultimate' Narr-game."
Then I'll just look overly defensive trying to correct them on two-- no, three accounts.