I'm no Bitcoin expert, but from skimming the paper it seems that the way they were able to achieve this (being "trustless") was by adding new opcodes. You record your position and commitment to the wager in the blockchain, at which time your coins are transferred to a "sidechain". Basically, the blockchain itself holds all the wagered coins.
The part that seems a little weak is they've created a new resource, Reputation, that allows you to report the outcome of a wager. Reputation is tradeable, and therefore gameable.
1) What adversarial analysis have you done to make sure that the reputation weighting mechanism can't be gamed, or the market making bot's funds emptied?
2) How do you handle miner censorship of vote revelations?
3) Why do you include floating point operations in the consensus code?
4) Have you considered separating the market-making aspect from the prediction market vote counting?
5) Have you considered implementing this proposal in an extended script directly, rather than monolithic opcodes?
1). Good question. Paul Sztorc -- the inventor of the Truthcoin concept, on which Augur is based -- has been helping us a lot with game theoretic calculations. The Reputation concept's goal is to create a Schelling point on the consensus. You can construct a thought experiment, where you look at how the Reputation payoffs align with a perfect coordination game (with Schelling points). The result was that there is around a 90% chance that the Reputation mechanism creates a Schelling point. I don't have the details of this calculation, but I'm sure Paul and/or Jack (our project lead) could provide them if you're interested!
As for emptying the funds, we have the B parameter in LMSR vary according to volume. However, we're currently talking with an expert in automated market making (PhD @ CMU in CS w/ thesis in market makers) and he's recommending that we do not use LMSR (because new, advanced, market making algorithms have came out that don't allow issues like the one you mentioned come up).
2). I'm not exactly sure what you're asking here. We currently plan to encrypt votes using time lapse cryptography and then decrypt them via distributed key sharing.
3). SVD 0 and 1 work fine for binary outcomes, however when you include scalar predictions you have to fit the scalar values between 0 and 1 (because that's what svd operates on), hence floating point decimals. We're currently writing and implementation using fixed point, however, and that's probably what we'll end up using.
4). Yes we have, that was the simplest way to do it in a bitcoin transaction, however...
5) Absolutely, we're currently looking at Serpent contracts as a way to implement this (it looks like the biggest hurdle will be keeping SVD costs low, but so far it's progressing nicely)!
Thank you, gojomo. We have talked to two of the authors of http://users.encs.concordia.ca/~clark/papers/2014_weis.pdf, and they were very helpful. Their proposed design differs from Paul Sztorc's in various key aspects. Scroll to the bottom of this page: http://www.truthcoin.info/faq/ for Paul Sztorc's comments on the paper to which you are linking above.
This sounds like a really well cushioned description of a generic system to implement Jim Bell's[1] Assassination Politics[2] system... "well cushioned" in that it uses the 'free market' rhetoric of the establishment to preemptively self-justify.
Hmm...this seems like a very interesting venture. I remember websites like this (there was one where you could bet on practically any event, but the name escapes me). I read the paper and found many of the concepts very interesting. The question is, capitalism (as we know it) is built upon trust; it is the essential value needed in capitalism, so how will a "trustless" prediction market work? I'm very interested to see how this works out.
Perhaps you remember InTrade? Or there's also a few UK websites like that. The problem is they're centralized, so they can go out of business or a person who judges / shows the outcomes of events could lie to get a payout.
As for capitalism, I would argue that greed is one of the most common elements. We rely on people being greedy and, generally, willing to do the least work for the most possible results - aka laziness. It's trustless in the sense that bitcoin is trustless, we're building a platform that unlike the traditional financial system, doesn't require you to trust others.
The way the trustless aspect is used is that people are incentivized to vote on the outcomes of events with the consensus, because if you don't, you will lose reputation in the system. This is explained in further detail in the game-theoretical explanation of this project, available here: https://github.com/psztorc/Truthcoin/blob/master/docs/Truthc...
Coolest Dapp I've heard of! It also sounds like a solution to an existing problem that could start running immediately. When are you guys planning to to go beyond the testnet and release the real thing?
The part that seems a little weak is they've created a new resource, Reputation, that allows you to report the outcome of a wager. Reputation is tradeable, and therefore gameable.