So, this is my latest work, an rpg that can be played on the internet over email, simulating a "story told in letters".

Questions:
1) Is the play document comprehensible? What do you need to know before playing the game? What do I need to add before you can understand the text?

---
Epistolary: The letter writing RPG

Concept: A story game played entirely through written communication that simulates the writing of the players characters.

Play:
1.Create a setting document.
1.Setting document explains the presumptions of the setting.
2.Setting document sets up the kinds of characters in the story
3.Setting document creates the setting question, a single yes/no question that will be answered in play.

2.Create characters
1.Each player picks one (or more than one) character to be in charge of.
2.Write a short capsule description of your character, and explain their core competencies, social role, and relationships. All characters should have an oppositional relationship to at least one other player.
3.Create a single, personal yes/no question that will be answered in play.
4.Roll 2 six sided dice. Play goes from highest scoring to lowest scoring, then repeats.

3.score questions
1.Give each question a pacing number, which determines how long the question will be left unanswered. Pacing number = Number of players * (Number of rounds you wish to explore the question-1) * 3.5

4. Write letters
1.Write a letter in the voice of your character. It can be to another character, to some other non player character.
2.Letters cannot contradict things in earlier letters, or the setting presumptions, but can state false beliefs, contextualize earlier statements,etc.
3.Questions cannot be answered definitively in a text without the question being settled mechanically first.

5.Score letters
1.Then, find a passage in the text which supports either a yes or no answer to one of the questions that arise in play. Cross off that passage, then assign a six sided die to the outcome to the question supported by the passage.
2.Each other player does the same to your letter.
3.Repeat, until each player has assigned two dice to your letter.

6.Roll dice
1.At the end of the letter scoring, roll all of the dice assigned to the various outcomes, and add them to the outcomes previous score.
2.If the questions total score (yes+no) equals the pacing number, then the question is reached it's climax. Each player has one more letter to contribute to the question. Once each player has contributed, the conflict is over and the question is answered.
3.If all of the questions are answered, the game ends (if somebody wants to write a postscript, thats cool) otherwise, play continues to the next letter writer.

I'm intrigued, but I need more information. Is there a gamemaster, or is this 'GM-less'? What are some examples of yes/no questions? Who is rolling the dice? 

Speaking of dice: how do you feel about bidding or token systems, instead of dice? Look at Baron Munchausen, Mortal Coil, or Nobilis, for examples of how to have uncertainty in conflict resolution without randomness. I.e. other players bid on how much they want your paragraph to add to the 'yes' total, versus the 'no' total, with the tokens being redistributed for later use.

I'm thinking of "I bid two tokens for 'yes', from the pool that everyone knows has 12 tokens in it" versus "Cool. I rolled a 12. Again. Take my word for it."

But that's a side-issue -- the main issue is still the need for more information. I'm a big fan of epistolary novels and games, and ma deeply curious to see how you solve the problems of remote access to players.

The game is GM less. That is, each person creates one player, the players conflict with one another. Usually, one person in a game like this takes it as their responsibility to insure the players are creating conflicts,ect.

I imagined people would play online, using the same dice rollers people usually use for PBEM. I think their are on line verifiable dice rollers out there.

An example of a yes/no question would be "Is the Kingdom of Bagravad conquered?", "Does Susan marry Bob?" or "Will Sgt. Smith return to his wife?"

These questions are used to structure the games conflicts around. If the setting question is "Is the Kingdom conquered?" the games central conflict will revolve around the invasion.

Very interesting, it seems to have a very bizarre approach to narrative theme: Themes are expressed as a series of distinctions, I'm guessing like "Is Geoff the son of the king?" Which then become the battleground for narrative control. You can't explicitly give the answer, but you could do stuff like this, I gather:
"The king asked after young Geoff the servant today, which I thought a little strange, but he has always been fond of him." or something that implies he was away on campaign months before the boy was born etc.

Now here is where it could get tricky; what if only one player is interested in exploring a certain question? The fact that you cross off lines as you use them to imply stuff means that people could try to grab the juiciest lines for their own question, it's a sort of scarcity situation. But that's probably not what you have in mind.

To make the dice mean more, perhaps you could instead roll for the relevance of a statement, as at the moment the dice seem to effectively give a randomised time limit on the game. I'm not sure how this counts as settling questions mechanically.

An alternative approach would be to say, "How much evidence do we need to decide?" and then for each statement you think is a clue for or against, you roll a d6 and on evens it is relevant on odds it is not. This works for both yes and no, and decides in what way statements must be consistent with it. If it appears to be relevant to the question but is not, then how can we make that work? People would still pick two sentences per letter, so then we have a new system for random game length:

If players provide (on avg) 1 piece of evidence per other player per round, and spread their evidence equally between player's questions (no guaranties) then:
game length = [sup]evidence required[/sup]/[sub](players - 1)[/sub]

But as I said, there are no guarantees. Players may jump on one question and finish it really fast, the fastest they could finish it would be [sup]evidence[/sup]/[sub](2*players)[/sub], but as you want random length that seems no great problem. You'd just have to make sure that players had got sufficiently invested in the initial setting to keep answering other players questions, as it is not required that they would be interested in the other peoples questions themselves.