Linkable/Pullable GitHub Version

Note: this FAQ is a literal FAQ. I did not write it to be an organized, readable document. The questions which are here, are here because they were asked very frequently. I copied each question here (so that it might be found with ctrl + f or a search engine) and wrote up a reference answer.

If your question is missing, please email it to me or use a pull request (link above).

PM = Prediction market.


What’s going on?

Hivemind allows users to use their Bitcoin to create and participate in markets for event derivatives (“prediction markets”). Like Bitcoin, Hivemind is a protocol, a set of P2P action-response relationships (like Bluetooth, spoken English, or the rules to Texas hold ‘em) for individuals to either use or ignore.

Although substantially more complex, Hivemind offers correspondingly substantial benefits, having potentially world-changing implications for science, taxation, corporate governance and politics.

Hivemind has users, just like Bitcoin, but it also has “employees”. These employees work for “Branches” (each Branch resembles a new InTrade), and vote on “Decisions” (election outcomes or price-feeds), which resolve “Markets” (the trading-arenas defined by one or more Decision). Users are free to create any Decisions or Markets they like, although they compete economically for scarce “Decision slots”.

What makes Hivemind different?

(In the author’s opinion) Hivemind is the only Bitcoin 2.0 project (defined as a blockchain for e-payments + something else) which is:

  • Free: To use Hivemind’s advanced features, you need to own Bitcoin. That’s it!

  • Authentic: Hivemind, unlike all other 2.0 projects, proceeds as though its creators actually believe in it. We pay for everything upfront (design, development, review, testing), with no guarantee that the project will even make it through all of these (risky) phases without encountering a fatal problem. As a result of this, you know that we literally can’t be dishonest with you (we have nothing to gain either way). Other projects hype up an “Order now, while supplies last!” message, followed by a shakedown for money! Such an approach, while admittedly effective at generating a community (with skin-in-the-game), creates a huge problem: even otherwise honest project-directors face a temptation to lie to the public (especially “greater good” lies powered by groupthink and wishful-thinking), particularly about obscure long-run problems. With Hivemind you don’t even need to be emotionally invested.

  • Reputable: Hivemind was designed by a Yale Statistician with degrees in the blockchain-relevant fields of econ, psychology, and mathematics, who has been passionate about prediction markets since long before Bitcoin was created. Hivemind was reviewed (favorably) by Andrew Poelstra, sidechains-co-author and skeptical author of widely-circulated technical papers on cryptosystem stability. Gregory Maxwell has been aware of the project for months, Peter Todd is actively reviewing it. Unlike 100% of all Bitcoin 2.0 projects, there have (so far) been no standing complaints from these elite-skeptics. Roger Ver, who has famously invested-in and donated-to some of the most promising organizations and experiments in the Bitcoin space (from original Ripple, to BitPay, to, and ignored many others, has endorsed this project financially.

  • Legitimate: Satoshi designed Bitcoin to be resistant to coercion. He noticed that value-storage was an enterprise which allowed bankers, politicians, and lawyers to extract (what I call) “ledger rents”, and that competition (e-gold, Liberty Dollar, etc.) failed to erase this anticompetitive rent. But he also noticed that the diffuse and regenerative bittorrent (gnutella, et al) were able to compete against the outdated film and record industry; hence a new e-cash which was P2P. Hivemind, similarly, reflects on the tragic fate of and the (neutered) ForesightExchange and Iowa Election Markets. Hivemind’s second coin type, the VoteCoins, take aim at a second type of “ledger rent”. Other 2.0 projects follow nothing like any of this logic at all, many have simply Pavloved the words “decentralization” and “good”, for no reason.

  • Useful: Hivemind already has “killer apps” of its own, things which it can do that nothing else can. It solves a real world problem of info-aggregation, which is the problem that annoys you when you [1] watch journalists lie on television, [2] have to deal with obviously dishonest politicians rising through an obviously-dysfunctional “democratic” process, [3] get beaten down by office politics at work, or [4] participate in any serious argument with anyone. The fiber optic cable of the 90’s gave the global community a “nervous system”, called the internet, but despite advances in filesharing (“memory”), and search / auto-complete (“pre-processing”), we don’t yet have a “central” nervous system to advise our decisions, because info-aggregation does not scale without markets, and internet-markets have never before been possible.

If this project is so much better than Ethereum, Bitshares, …, then why haven’t I heard of it?

Firstly, I do not do any promotion / “active marketing”. I feel that “talking” about Hivemind goes against the spirit of the project: that talk should be ignored (in favor of action). I have more important work to do than to deal with the (socially dysfunctional, in my view) Bitcoin community.

Secondly, I don’t need any money, or help from the general public. Those other projects don’t already have funding from reputable investors, and so they need to have crowdsales. These crowdsales grant a large number of people “skin in the game”, at which point they become slavishly devoted to promoting the project. My project requires no money, and uses Bitcoin (whether the Bitcoin community, or any other community, likes it or not). Once the Bitcoin community learns of Hivemind (which might be a while, as they’ve been so overwhelmed with terrible ideas named “-coin”), and that it is free (does not require any other coins), I’m confident that they’ll be promoting “their” new project as much as they promote Bitcoin.

You really think this project is better than all of the other Bitcoin 2.0 projects?

Yes I do. Like Bitcoin, this is the only project which aims to solve an actual problem in computer science (P2P Oracle), it is the only project which had economically viable non-blockchain implementations (real life “trial runs”), which were later censored (namely, it avoids the deadly CS sin of rewriting code from scratch, and imports Bitcoin’s (crucial) monetary network effects. Roger Ver hired the most skeptical and technically knowledgeable people in the area (you know their names) to review the design. So far, no design flaws have been brought forward by these reviewers.

While other projects employ multi-million-dollar budgets, only to struggle to meet their (multi-year extended) deadlines, this project is free, and we’ve already nearly finished.


What’s in it for you? And what’s in it for Roger Ver?

I understand why this question comes up (despite the fact that, as open source software, the motivation or even identity of the software-author is completely irrelevant). Given the frequent occurrence of scams in this community, even more skepticism than usual is warranted with even completely free projects.

It is true that there may be an auction of the initial “VoteCoins”, which might raise a nice amount of money for whoever administrates the auction (probably Roger). However, should one even occur, you are under no pressure to participate in this auction: VoteCoins are themselves not required to use Hivemind (BTC will work just fine, VTC are used to “work” for Hivemind), and I would imagine that many of those most excited about Hivemind would never own any VoteCoins in their lives. The New York Stock Exchange (NYSE) is itself owned by Intercontinental Exchange, which is actually listed on the NYSE (as NYSE:ICE). Thousands of people trade in the NYSE every day without owning any shares of ICE, and millions of people around the world benefit from the operation of the NYSE without even being aware of ICE’s existence.

Or, the VoteCoins could be privately owned and slowly leak onto the marketplace. It makes almost no difference to the user, just as the Walmart stock price has almost no effect on the individuals who shop at Walmart stores.

I don’t believe that either of us are motivated primarily by this potential financial return. I can only speak for myself: I feel I owe a debt to society for the comfortable life I have enjoyed so far (particularly where the internet and FOSS are concerned), and I feel society must change as the result of injustices dealt to me and my interests in the past.

I hear you are using sidechains. Isn’t Blockstream an evil syndicate, destroying everything beautiful in the world in their march to unearth profits for their corporate masters? I heard somewhere that Adam Back is literally Adolf Hitler.


OK, first of all you can use Blockstream’s code / research, even if you have nothing to do with their organization.

Secondly, this criticism is of the form: “has corporate job” therefore “everything they say/do is suspect”. In rhetoric this is known as Poisoning the Well, a brutally unfair 1-hit KO which can’t be blocked or countered (the Avada Kedavra of debate) except by an un-poisoned friend or by one’s (commonly known) past reputation. One of the most ridiculous things about this fallacy is its symmetry: by the logic submitted, anyone criticizing Blockstream would have to prove that they (the critic) do not themselves work for a corporation, or else the critic should be completely ignored (according to them).

Why won’t anyone explain how those investors plan to get a return on their $21 million investment? Because it can’t be spoken aloud. You’ll have to figure out yourself, like I did, Good luck!

Perhaps Blockstream is evil, or biased (clearly, like everyone, they prefer their own work to that of others). But no evidence of any wrongdoing has been brought to me.


Why would the miners secure this sidechain?

From the Miner’s point of view, their choice to secure Hivemind results in more tx-fees (BTC fees) for them, at a cost of zero (once configured, merge-mining has marginal operating cost of zero). It is a no-brainer.

They are also unlikely to attack Hivemind, for reasons similar (but not identical) to those motivating their non-attack of Bitcoin: a failed attack results in the destruction of the pool operator’s buisiness, and a successful attack destroys the future stream of tx-fees for all merged-miners.

How do you know that people won’t claim an outcome was whatever will benefit them personally?

(It would be best to consult the whitepaper on this question, particularly the figures on pages 24-26 of Whitepaper 1.4).

A parallel question would be, why wouldn’t (a website which once provided USD prediction markets) simply steal their client’s money? Why don’t banks steal their client’s money?

The major reasons are that: [1] this would result in an overnight failure of their business, and [2] law enforcement would catch up with them (punishing the attacker for breach-of-contract, and possibly even reversing the stolen funds [if it’s not too late]).

It is clear that all businesses must operate for the long run (think customer service departments, or “sending back a meal” at a restaurant), making reason [1] enough to keep these agents in line. However,some businesses are uniquely vulnerable to short-term temptation because they not only “accept value” but they are also custodians of their customer’s value, necessitating the extra enforcement implied by [2].

[1] Reputational Capital

To mimic the value-added by beneficial voting-activity, Hivemind has “Branches”, which are analogous to corporations. Each would be like its own PM-business, generating profits (from their customer’s trades) over the long run, and resolving each market as its event takes place.

Branch-owners are compelled to report on a (capped) number of questions called Decisions. This restriction on the supply of reporting allows Branches to differentiate and thereby own and capture the economic value produced by their good behavior. Through something called Branching, each Branch can choose to split into two daughter Branches. Economic incentives are such that differentiation is always possible / some monopolistic-rents are always extractable, yet there is flexibility to scale as needed (these details will be discussed in Whitepaper 1.4).

It is a common misconception that the SVD-resolution algorithm is the main innovation behind Hivemind. While important in its own right, it is this concept of a second “share-token” with owner-operators (laborers who must buy the coin to work, and who lose the coin if they misbehave), as well as the careful operationalization of all the economic costs and benefits of each agent, which drives the blockchain-logic.

The SVD-Resolution Algorithm

Many of the details of this algorithm are in this presentation and the whitepaper.

Try and break my Resolution Demo, and let me know what you find!

Major points:

  1. Voters have a strong incentive to vote “the way they believe others will vote”. More deviant individuals are punished more.
  2. Voters must vote on many Decisions, many more than they could practically be trading on.
  3. Voters must buy reputation tokens, use them to vote regularly, and gain an economic return from them proportional to the usefulness of the network, hence Voters and Traders are likely to be different groups of people.
  4. Even if a substantial majority (>80%) votes unrealistically, as long as these votes are random, there are enough Decisions in a ballot (50 or so), and there is a tightly coordinated core group of identical truth-tellers, this 80% coalition will not affect PM Outcomes at all, and instead only lose their reputation-tokens. Liars must form a tightly coordinated coalition.
  5. Voters have an incentive to trick other voters into voting unrealistically (regardless of the proportion voting realistically). Thus, a tightly coordinated coalition cannot form.
  6. Voters face a penalty for not voting, which is high when the proportion of missing votes is high (a few missed votes won’t hurt you, but in a low-turnout environment you will be penalized more severely). Thus voters must always vote.

Votes cross-validate other votes, in addition to supplying the data used for Outcome-calculation.

Why don’t you…?

The idea I published was not my first idea for a P2P PM, it was my best. I probably did think of whatever you have in mind, but feel free to ask anyway (if you’d like to know why I didn’t select it).


  1. …have a single Judge collect fees for his Judging services? Good judges would gain a good reputation, and charge more for their services. Bad judges would lose the stream of future income payments. We could even force Judges to put up collateral so that they would have more to lose.
  2. …have a few ‘trusted feeds’, where the outcome is determined as the median feed, or outcome is a draw if feeds disagree too much?
  3. …publish a hash of every event description combined with the outcome “true” or “false” and sign the real outcome? Thus we can build up reputations and predictions can be placed completely independent.
  4. …store bets in a 2 out of 3 multi sign address? That would make it impossible for the central PM to run away with the money - just to choose the winning side.
  5. …have both sides put 10% more money in a multi-sign address? No the person that loses the bet can admit the loss and sign the transaction to the winner but he gets back the 10%. If none of both admits to be the looser we sign a transaction to the winner but take the 10% of the looser.
  6. …have a preselected Judge decide, but allowing users to force an audit of the Judge if they disagree with his ruling?


  1. Obviously, the Attack Profit here is defined by 3 inputs: the cash influx for attacking today (call this “Defect”), the stream of payments in the future for honest judges (call this “Conform”), and the discount rate used to reconcile the fact that these payments are taking place at different times (call this “r”). There are severe problems with all 3 inputs: ‘Defect’ can skyrocket unexpectedly as a market becomes popular, ‘Conform’ can collapse on news about the future of the protocol or judge-industry-competitiveness, and r will change with the Judge’s preferences (if a Judge wants to retire and leave the Judging industry, r can become infinite). This uncertainty means that it is practically impossible to guarantee today that a given Decision will have negative Attack Profit (i.e., not be attacked) at the time of Judgement.
  2. The first problem with this is that there is little to truly prevent the feed operators from colluding to lie (which is unacceptable). By buying cheap shares and reporting falsely, feed operators can reap substantial profit (InTrade’s Barack Obama market expired with 20 million USD volume). If feeds are anonymous this attack is easier to finish, if they are non-anonymous this attack is easier to start. With a low number of feed operators, that have brands/customer-support-contact info, a clear reason to work together, and a clear payoff, they will either collude, or create chaos by falsely accusing each other of attempting-to-collude. Secondly, non-cross-validated/centralized sources are fragile, and can be corrupted, hacked, closed by authorities, etc. Thirdly, this attack does not allow the feed operators to ‘retire’ and choose to do something else. In Hivemind, one wants to keep one’s reputation-tokens as “clean” as possible, so that one can eventually sell them, but with systems like these, there is no safe way to cash out, so one profits by making one’s last report a lie, and lying about when one plans to make a ‘last report’. Fourth, it is complex to choose a method for determining the feed-sources and deciding what to do if they break or start lying, in a way that cannot itself be gamed (ie with me falsely saying that you are a bad feed operator and you should be ignored).
  3. This requires trust and provides no method for cashing-out that trust. Again, a 20 million payday today for one lie is very persuasive, against an uncertain future stream of small fee payments. This strategy is already used by the service Reality Keys, and may be effective in a low-cost environment where many specific contracts are needed.
  4. Same problem, attacker simply trades on the losing side before lying. This is unblockable, and attempts to block it will let clever traders ‘fake out’ rivals by buying up losing shares last minute to tinker against the countermeasure.
  5. Although I was working with this idea for awhile, its not as good as one might think, because in Nash Equilibrium you still need an incorruptible way of determining outcomes (for the holdouts [even if it were never be used on-path, the game theory requires that the off-path reasoning be persuasive]). Furthermore, one holdout can cause you to have to research the issue, and once you’ve done that, it costs nothing to re-use that research for all other traders in that market. So a losing trader can easily force the administrator to research his claim, if the administrator is not prepared for the possibility that he will be required to research ALL claims, than this can be gamed strategically by buying a small amount of every losing claim.
  6. This is another idea I considered, but there’s a strange paradox where the Judge and the Auditor don’t really gain reputations (which, again, they can’t sell) at the same time: either Judges are lying (and Auditors are proving their worth), or the Auditors are never needed (and hence their reliability cannot be assessed). Who decides who becomes a Judge and Auditor, and who Judges/Audits what (to avoid the obvious problem of Judges being their own Auditors? Who bears the economic costs of establishing these selections? In a software environment these selections and associations will be strongly anonymous, individuals may have several accounts. If auditors are to be assigned randomly, what will be the source of randomness? Why are Auditors superior to Judges? Again, all parties involved may still be able to collude (if they are lucky enough to be paired). Moreover, if miscreants force needless Audits, it just delays everyone for no reason. Mandatory fees for auditing could prevent this, but could also weaken confidence. My proposal assumes that all Judges and Auditors would lie, and that they therefore must audit each other always. This pessimism results from the incentive to lie today (which is very strong).


Generally, many ideas ignore…

  1. …the fact that trusting a user with outcome-resolution is no different from trusting a user directly with money. Escrowed funds are already “not yours”.
  2. uncertainty. If you can’t guarantee the mechanism will always work, people will respond strategically to this, and avoid using it. Eventually, fate will drift everything into its weakest point, if that point is weak the idea will break.
  3. …the Chicken and Egg Scale Problem, that there’s no way to know if something is secure until people begin trusting it with large amounts of money. This makes all empirical assumptions about the project (51% honest, presence/effects of competition) permanently uncheckable, and disqualifies projects with many assumptions or even one potentially-untrue assumption. As a project grows, it may be attacked in extremely expensive ways. If it would not survive such attacks, it will never grow to that size.
  4. differences between brands/individuals and digital identities. In the real world, you can lose more than you’ve established: you can be sued, shamed, thrown in prison, or (traditionally) whacked. Online, you have little to lose.
  5. …the Exit Scam, i.e. they tell Judges that the most profitable way to retire is by attacking on their last turn (can’t be sued/killed, see above).
  6. …how easy it is to collude. If Judges think the mechanism is going to fail soon, they’re actually much more likely to betray it. They could think this for any reason, including irrational panic, Keynesian-beauty-contest-panic, or induced-reasons such as “some guy says he’s got 5 Judges and will kill their families unless they vote X”. This implies Early Retirement and #3.
  7. profit-equilibrium. If Judging is a profitable industry, more Judges will join, increasing competitiveness and decreasing profits. This implies Early Retirement.
  8. the opportunity cost of the attack, ie that one could potentially make $100 million dollars just by moving a switch here rather than there. You intend to discourage this how, exactly?
  9. …the Double Mechanism Reputation Problem, that proposing two ideas at once (such as a Judge being checked by an Auditor) necessarily means that one idea will go completely unused, making it irrelevant. This is fine strategically (“off-path reasoning”), but not economically (how do you pay Auditors who are never used? What if 10 million people sign up to be auditors?).
  10. Sybil attacks, i.e. that one person can be pretending (even from the very beginning, or for a long period of time) to be several people.

I did not base my own choice on trial-and-error, I instead based it on my day-to-day real-world truth-finding experiences.

In practice everyone [may] pull their data from some kind of external data feed. Presumably they’ll use whatever is simplest and cheapest, then they’ll be stuck with that until it completely breaks…

This concern has only been voiced for scaled claims, not binaries (which probably are novel and don’t have a feed). If Voters are instructed to use a few external sources (for comparison), they can double-check their answers (of course, if they believe others will NOT double-check, they will not either, so they might not), or more importantly switch to a new source if the first is broken. I personally find it highly unlikely that Yahoo or Google will start publishing wildly disparate financial index data, as they have every reason to ensure that the data is accurate.


If you built something like this, what use would people get out of it? Most people/organizations aren’t interested in improving their forecast accuracy through PMs. Who would pay to create (let alone subsidize) these markets?

Excellent question. Firstly, Authors (who bear the economic cost of Market-Creation) are rewarded with a slice of transaction volume. Recreational speculation is likely in markets covering sports and politics, arbitrage transactions are likely in markets tracking a price index, and in many cases, individuals will just disagree with each other passionately enough to begin wagering (global warming, gun control, etc.).

Secondly, the public might just pay for publicly useful information.

Thirdly, although the information revealed by a PM is public, the effect of that revelation may be privately beneficial. For example, consider Robin Hanson’s Wish, where conditional PMs are introduced to advise firing an incompetent yet entrenched CEO. With 1000 USD per firm to kickstart this idea, we might then collect data on firm/BoD/investor response. However, if we did not have 1000 USD per firm ourselves (we don’t), is there not someone who would benefit from the creation of this information? Of course there is: the second-in-line for CEO! CEOs are typically paid so much that their salary can be twice that of even the CFO. 1000 bucks is nothing for a decent shot at such a nice raise, to say nothing of the competitive spirit which overwhelms those who seek the top job. Even better, individuals would not try this if they suspected their CEO of actually being quite competent. So there is actually an economically-efficient self-selection in the creation of these markets: when unneeded, they are not created. Better still, competent CEOs may themselves create these markets to protect their job (although I doubt this type of activity will happen for some time).

Fourth, individuals could collaborate via assurance contract to efficiently pool their info-demand. In fact, in my applications paper, I describe a way of using PMs to create extremely incentive-powerful Trustless Dominant Assurance Contracts. To my knowledge nothing similar has ever been conceived.

Isn’t this just gambling?

No. Gambling is a game of luck…the return is unalterable by skill. In a casino, the odds do not change, expressing the static uselessness of the betting activity there.

In a prediction market, the “odds” (the prices) change as a result of information, and as a result of which bets are placed when. The variations in price produce information which is supremely reliable, as well as universally common. This information, produced nearly for free by the voluntary actions of individuals, has tremendous social value.

What about people with gambling problems?

Sadly, around 2% of the population seem to be unable to control themselves when gambling. This rate has been stable over time and is independent of casino availability, suggesting that this is some kind of biological disposition. It is interesting that the DSM-V classifies Problem Gambling under Substance-Related and Addictive Disorders. Similar to alcoholism, treatments for this unfortunate condition include 12-step programs and CBT.

The problem of addiction is quite a serious one…it is very difficult to prevent people from doing what they want to do. Even large nation-states, with well-funded law enforcement and court systems, are unable to dissuade many citizens from using drugs. Even when these individuals are actually incarcerated, 1 in 50 still can get access to their drugs (in prison).

Fortunately, just as individuals can choose to take drugs, they can also choose to join a support group or check themselves into rehab.

Problem gambling is exacerbated by fast, high payouts at highly improbable odds. PMs don’t offer that at all, as the payouts resemble those of the modern stock market (roughly 1% volatility per day). Even far into the future, when Branches mutate into faster “Casino Branches” (should this ever happen), a PM specifically designed to offer some kind of “gambling market” would be, in practice, unable to reliably offer bets at highly improbable odds, as the trades would excessively alter the odds themselves.

Finally, the market environment does guarantee that these gamblers, at the very least, are always getting fair odds and charged minimal fees.

And, most finally of all, Bitcoin already enables a great deal of gambling. Hivemind seems unlikely to enable much marginal gambling, let alone marginal problem gambling.

If PMs can be used to finance Public Goods, can they also be used to finance Public Bads (for example, to assassinate someone)?

This question has grown in popularity to the degree that it warranted inclusion in the applications paper (Appendix 1, forthcoming).

Can’t PMs be driven by people who are oblivious to how little they know?

Yes, at first. However, these people will quickly start losing large quantities of money, and thereby lose their ability to influence the market. Moreover, the existence of these ignorant “sheep” will attract “wolves” who profit by correcting the mistakes of the sheep.

How do you address the problem of low liquidity?

For PMs which are bounded, it is possible to design them such that traders can always update the market price, by trading with an automated market maker powered by a market scoring rule. I threw together an Excel sheet in the docs folder LogMSR_Demo.xlsx for additional clarification. Feedback is appreciated.

Won’t the BTC/USD rate be so volatile that it overwhelms any change in the prediction’s accuracy? Doesn’t this project need stablecoins?

This project creates a “stablecoin” for every Decision in it. The Decisions define ‘dimensions’ of the Markets, and partition them into states.

So it is possible to bet “in” any currency for which there is a Decision (USD, Gold, DJIA, etc.), see Example 5.


Does this really need its own blockchain? Why not do it “on top” of Bitcoin?

Yes. The design requires a way to conditionally pay out money (i.e. pay out if this happens or this), as well as VoteCoin based on voting behavior. Even if drastically simplified, the design would fundamentally require those things. A blockchain cannot sign transactions, and we cannot introduce a ‘signer’ without introducing trust.

In some of your notes you suggested timelines of a couple of weeks for voting. Would it be fair to say this wouldn’t be suitable to horse racing etc. where the events happen very fast, there are many of them and payouts need to be made fairly quickly?

The price will not fix at 1 or 0 until voting occurs, but it will converge toward one of those values as the event info is revealed (ie just as the horse wins). If voting is in two weeks, there may be some time-value-of-money/time-preference/liquidity effect, but shares of the winning horse could still sell for 99.0 or 99.9. Those buying at 99 would be Wall Street / banker types who pick up the shares purely to earn 2 week’s worth of above-market interest on their capital. So fast cashouts shouldn’t be a problem.

Why don’t you perform SVD-resolution more often?

Three reasons:

  1. For security reasons, we need a large group of multiple Decisions per Judgment Period.

  2. Judgment is inconvenient, and we would like to minimize inconvenience. Because of setup costs, it is likely easier for judges to sit down once per month, at their leisure, and do all Decisions at that time, than it would be to force them to do this every week or every day. Monthly JPs allow for a two week vacation, for example.

  3. There’s really no benefit to doing so (see the horse racing question above).

Do strategic decisions change if one node started publishing (potentially biased) votes immediately upon noticing them? One could publish 10% attack-votes, and the 11th % user would be tempted to conform, leading to a cascading failure.

Ballots are encrypted, and contain a new destination (public key), for this reason. Votes are cast in one period, and unsealed in a later period (during which no new votes are cast). Because private keys are required to decrypt, it is always impossible to prove that you’ve voted a certain way (and recall that you have an incentive to vote honestly yet say that you are voting dishonestly).

Is SVD computationally complex, to continue doing on such a large scale and frequency?

That depends on your definition of complex. In python, on an i5 processor, svd solves instantly for a matrix of dimensions 10000 x 100 (my expectation for the steady-state requirement). It is a common misconception that svd is performed frequently in my scheme. It is only performed once per Voting Cycle (per month or so) per Branch, upon the maturation of large batches of Decisions.

It sounds a lot like you’ve reinvented…

So far, people have written in ‘Bayesian Truth Serum’, “Thirteen theorems in search of the truth.”, EigenTrust, and a few other things (that weren’t quite right).

I probably did. I don’t think any of my core building-block ideas are that complicated (although the assembly was), especially compared to the library of academic work published each year since forever. BTS allows you to vote on how others will lie, this is a little better but doesn’t scale as easily (and furthermore, in a coordination game, all the votes would theoretically be equal to each other).

Do you feel its necessary to penalize miners for not voting? In my mind it would be better to rely on positive rewards as we wouldn’t really want to encourage voting by those with less interest/knowledge in the outcomes.

Actually it is necessary. Firstly the concern you mentioned later is mitigated by (and one of the main benefits of) Branching, where voters stay only on the branches/trees where they are interested and knowledgeable. More importantly, however, if we assume voting is costly (which it certainly is), then the assumption I made on page 2 that voters are lazy implies that theoretically none of them will vote, in a sort of ‘tragedy of the commons’. Realistically, some might care enough to vote, but this would weaken the confidence in the coordination-ballot, as it takes a smaller % of votes to throw off. Thirdly, voters are not just contributing their point of view when they vote, they are also securing the network by validating the point of view of all other voters (via the coordination game). Notice that an individual’s penalty for not-voting is only high when many other individuals have also not-voted. You can miss a vote every once in a while, and as long as everyone else is voting it won’t really be a big deal for you.

Why is reputation (“Votecoins”, “Hiveminds”) tradeable? That seems overly complex, and/or allows rich people to attack the network.

See the “Reputational Capital” section of the “How do you know that people won’t claim an outcome was whatever will benefit them personally?” question.

Basic Details

What does this project aim to accomplish?

Traditional PMs have many problems:
1. Counterparty risk (must trust whoever is holding the money to use it as promised). CR is why one can’t make bets with certain friends.
2. The trustworthy PM-administrator must (coincidentally) share your taste in PM topics.
3. The PM-administrator would have to be expected to provide service for a length of time (and not be interrupted by an executive’s death or regulatory interference).
4. PM-administrator would have to offer competitive prices, be economically viable, not go out of business, etc.
This project aims to solve all of these problems.

Can we test your resolution-algorithm empirically? You are asking users to accept a few significant sources of uncertainty at once.

Someone is already trying to do that (and is using my notes here, if anyone else wants to try and get this set up). Although the folder is labeled Qualtrics, I believe the current plan is to use Mechanical Turk (following the Harvard study at PeerPrediction.pdf).

If two or more people could do it, that would be fantastic!


Mining / Coin Distribution ?

Any game-theorist worth his salt knows Pareto/Kaldor-Hicks improvements. Cryptocurrencies have the beautiful ability to preserve ownership by taking snapshots of the unspent-outputs set. I see no good reason not to do this for the currency-tokens, just as I see no good reason not to use Bitcoin’s proof-of-work mining scheme. I’ve never seen a proof-of-stake proposal that achieved Kaldor-Hicks, and in fact my guess is that it is this challenge (KH) that primarily motivated Satoshi to choose his mining scheme (which is Pareto). The reputation-tokens are more complex, but the obvious choice is to distribute them to individuals in proportion to what they sacrificed to create the software.

Have you seen a similar effort from some CS guys at Princeton? Wonder if they’re solving the same problems…

I had previously heard about the Princeton guys but I never heard any news from them and had forgotten about it. Would be great to collaborate, but I had already finished most of my ideas when I found that article. To date I have heard nothing from them about their project. As always: Nullius in Verba.
Update: The long awaited paper did actually come out. Unfortunately, we seem to disagree on the true nature of the problem of designing a decentralized prediction market. Namely, in Section 5.3 the paper assumes Prompt 1 above (in “Why don’t you…”), which I view as unacceptable (see above). The paper does mention two ideas, ‘PredCo’ (which solves the discount rate problem), and the Keynesian Beauty Contest, which, when combined, more closely resemble what I had in mind. The second major proposal in the paper is for an order book system. Instead, I chose to use market scoring rules, for a variety of reasons, one of which is that operating a MSR is extraordinarily similar to sending a normal Bitcoin transaction. The paper is not truly a proposal for anything, instead a sort of menu of different ideas and avenues for decentralizing prediction markets. A novel is easier to assess than a dictionary, so it difficult for me to critique the paper more directly than this.


How do I contact you? Please share your opinion, and your questions so the FAQ can grow!

How on Earth do you pronounce “Sztorc”?

In Polish, “c” is pronounced as the English “ts”. Check it out!
So if an English-speaker just said “Storts” they’d pretty much have it. It rhymes with “Sports” and “Quartz”. The z sound (“Sztorts”) isn’t as important and usually just distracts English-speakers (who have never needed an “Szt” sound).

Donation Address