[RFC] Betcoin

[ben-abuya]

I suggest you look at, for instance, the "Mike Huckabee to Announce" contract on Intrade.  Even though Huckabee said live on his show that he will not run, his contracts are not selling for .01.  Odd, especially since Intrade *will* settle these contracts at zero in December.

The question is simply where the big money is at.  If the big money is wrong, it's still right!

I think the problem with that bet is that there isn't any big money. There's not much volume there, and it apparently isn't a very interesting wager for people to get in on. Also, since Intrade will close this at 0, how is that supposed big money going to be right?

Here's on of the discussion on the subject:
http://forum.bitcoin.org/?topic=6900.0;all

Awesome thread, thanks. I'd never had expected that OP_BLOCKNUMBER would work in the sense of "the current block bitcoin is up at time of verification". But it appears that OP_BLOCKNUMBER in the sense of "the block number at the time of block insertion" could still work, with the caveat of the re-org problem.

With the oracle approach you don't really need to go through all the trouble. Start an oracle service that signs arbitrary with data with the current chain height.
If you and another party think its safe then just use that. That allows you to base transactions on chain height without risking everyone else.
The same can be done for a time based transaction. Using the oracle approach is much safer. If the oracle signs arbitrary data with the time then it doesn't matter whether the chain gets reorged or not. Your transaction will still be valid even if you don't/can't broadcast until after the time has passed.

Yes, but I have a big problem with known centralized entities like this, even if they're not that centralized and they're pseudonymous. Here's why: suppose the market in question is a bet on the heroin crop, used by drug vendors to determine and hedge prices. The oracle or oracles might become the target of a well-funded DEA investigation. They would use block chain analysis, search warrants, CI's and anything else they could think of to uncloak the pseudonymity. In an oracle-less scheme there is no such high value target. You might not think a heroin market is a good idea, but I'm envisioning a global financial market built on this that's immune to regulation. The SEC might go after oracles in stocks, bonds, gold, dollar and pig bellies with the same gusto.

[1714793211]

1714793211

[1714793211]

1714793211

[o]

I have read the awesome contract proposed recently and I think it is a must for the Bitcoin in the future. However, I do not understand how your proposal can avoid the mediation if any conflicting occurs?

You proposal seems need the network to access external data. So, who decide whether condition is matched? If the external data is used, how can you sure that the network determine the result are honest? and I dont know why you said that your network is more resistant from the regulation.

[cartman]

dangerous idea: an anonymous censor resistant marketplace of user-created bets is also an implementation of the assassins-market. one could bet that some target person is still alive until date xy and an assassin could accept the bet and make the event happen...

[ben-abuya]

dangerous idea: an anonymous censor resistant marketplace of user-created bets is also an implementation of the assassins-market. one could bet that some target person is still alive until date xy and an assassin could accept the bet and make the event happen...

Yeah, bitcoin is a dangerous idea, too. Any powerful technology is dangerous.

I have read the awesome contract proposed recently and I think it is a must for the Bitcoin in the future. However, I do not understand how your proposal can avoid the mediation if any conflicting occurs?

You proposal seems need the network to access external data. So, who decide whether condition is matched? If the external data is used, how can you sure that the network determine the result are honest? and I dont know why you said that your network is more resistant from the regulation.

There's no conflict possible, everybody agrees to the terms ahead of time. The network doesn't have to access external data, since payout is based solely on closing prices, which is internal data. It's an unorthodox approach so you might want to reread the original proposal. It should be more resistant to regulation because regulators like a few high profile players, and they like to prevent everybody else from getting in the game. With this, there are no high profile players. Like bitcoin, if you want to shut it down, you have to shut down everybody.

[TierNolan]

dangerous idea: an anonymous censor resistant marketplace of user-created bets is also an implementation of the assassins-market. one could bet that some target person is still alive until date xy and an assassin could accept the bet and make the event happen...

The fact that it relies on social pressure could also help here.  A contract about someone who was murdered could very easily have a close price as if the person was still alive.

[ben-abuya]

dangerous idea: an anonymous censor resistant marketplace of user-created bets is also an implementation of the assassins-market. one could bet that some target person is still alive until date xy and an assassin could accept the bet and make the event happen...

The fact that it relies on social pressure could also help here.  A contract about someone who was murdered could very easily have a close price as if the person was still alive.

Heh, kind of like throwing an opposing home run ball back. Interesting.

[luv2drnkbr]

I am very disappointed that this thread has nothing to do with poker.

[JoelKatz]

It seems pretty simple to do this with no third party verifier:

1) To place a bet, you place a transaction that can be claimed either:
 A) By you, up to a certain time in the near future.
 B) By the person you're placing the bit with, after a certain date just before the event.
 C) By you, at any time well after the event. (1 day, say)

2) To accept the bet, your partner places a transaction that can be claimed either:
 A) By him, any time after a time just before the event.
 B) By you, at a time well after the event. (1 day, say)

3) If he doesn't accept the bet, you claim back your money per 1A. Make sure to take back your bet if it's not accepted. If you take back your bet after the bookie places his, he takes back his bet using 2A (he'll have to wait a bit).

4) If you win, you claim both transactions using 1C and 2B, a bit after the event.

5) If you lose, he claims both transactions using 1B and 2A as soon as he knows you lost.

Yes, your bookie can cheat you. But that's much less of a problem than trusting a central authority.

[Michiko]

There are many of interesting problems and ideas around prediction markets with Bitcoin connection. The idea of integrating betting (conditional transactions sounds more professional...) into the core of the system, e.g. the block chain, naturally causes a lot of controversy. Unfortunately this discussion overshadows other interesting points raised in the original post.

If only for my personal benefit - could I please encourage further discussions of other aspects of prediction markets? I'd be especially interested in reading some more discussions with posters like dacoinminster or ben-abuya etc. who have voiced related ideas in other sections.

(Sorry if this is considered thread-jacking; I've mostly joined & posted 5x newbies to restart discussion around prediction markets. Maybe a more established member could start a new thread if preferred. I'd like to understand better how would the block chain cope with even a moderately successful implementation - cf Betfair's daily transaction numbers? Or, how can this 'escrow' service handle offsetting of opposite bets since it won't be able to do margining? How does the market communicate to the 'oracle' part which transaction-bets are in demand? Etc. ) 

[just_someguy]

Re-reading some of this thread I realize my grammar goes way down hill when I type from my iphone.

Yes, your bookie can cheat you. But that's much less of a problem than trusting a central authority.

There isn't a central authority just an entity that you and your counter party agree to trust for a single transaction.

Competing "authorities" is way better than trying to use a block chain for external data. Lets use the intrade example.
(I'm not an internet gambler so I may not describe how intrade works correctly)

Intrade has a prediction market with enough saturation in the market that it reaches a critical mass become a reliable indicator.
Alice wants to place a wager with Bob on an intrade event. Alice does not trust Bob and vice versa.
They want the bet to be settled in bitcoin and both will cheat the other player if given a chance since that's the way it seems to go on the internet.
Ben starts an anonymous intrade settlement service on TOR. There are many, this is Ben's.
Ben publishes a public key and claims to honestly sign and report the outcome of every intrade event. (This could be done by intrade directly but lets assume they aren't interested.)

To decide whether to trust Ben Bob and Alice each examine the Bitcoin block chain proper looking for all script contents containing Ben's public key. They each independently verify the results against their known list of intrade outcomes. Each is satisfied that Ben reports results honestly. They also decide on a time oracle using the same process.

Using a combination of the techniques on the contract wiki entry Bob and Alice exchange a transaction with the following properties:
If the event happens Alice wins, if not Bob wins. If the agreed upon time has passed each gets their money back. (This handles event cancellation but it could be done directly by Ben without the time oracle.)

Benefits of this scenario:
Neither Alice nor Bob can cheat each other.
Ben cannot collude with either because he doesn't know about the bet.
Bitcoin becomes the escrow agent.
Charlie and Dan don't trust Ben's intrade service. They are not forced to use it.

Variations of this scenario:
All miners are owned by governments that do not allow gambling. Because of this Ben charges a fee for each event pairing. The fee covers the generation of a unique public key for each matched bet. Ben does not publicly publish paring results until a few days after the event has occurred. In the interim he only provides event results to Alice or Bob. The bitcoin network still validates all betting transactions because they appear no different than regular transactions. By the time event results are published the transaction is too far down the chain to reverse.

[JoelKatz]

Neither Alice nor Bob can cheat each other.
Ben cannot collude with either because he doesn't know about the bet.

I see how you can have one or the other of these properties, but I don't see how you can have both.

In order to rig it so that nether Alice nor Bob can cheat each other, the transactions must be published in the hash chain. Otherwise, Alice or Bob could simply post a conflicting transaction to the hash chain and empty the account that was supposed to be used to pay the other. But if the transactions are published in the hash chain, how can Ben not know about the bet?

Update: Ahh, I figured out a way to make it work. Alice creates four transactions:
1) A transaction to lock her money in escrow such that it can only be claimed by transactions 2, 3, or 4 - by hash.
2) A transaction to take her money back if Bob does not accept the bet.
3) A transaction to pay Bob if he wins.
4) A transaction to take her money back if she loses.

She posts '1' to the hash chain. She holds 2 and 4 for herself. She passes 3 to Bob.

Now, Ben cannot tell from just seeing 1 what it means, since the transactions that involve him (3 and 4) are not posted yet. Bob, if he wins, can post transaction 3 himself to claim his winnings.

[just_someguy]

I see how you can have one or the other of these properties, but I don't see how you can have both.

You are right. I meant more in the practical sense that he could collude once but then it would essentially kill his business since the falsely reported outcome would be openly visible in the block chain.

As far as locking the money up so it can't be spent I'm assuming the technique from the contracts wiki would work:

Two transactions are used: one (the contract) is created and signed but not broadcast right away. Instead the other transaction (the payment) is broadcast after the contract is agreed to lock in the money, and then the contract is broadcast.
This is to ensure people always know what they are agreeing to.

[ben-abuya]

It seems pretty simple to do this with no third party verifier

I dunno, to me trusting the other side in the bet is like trusting the other side in a bitcoin transaction. Calling him a bookie is like calling a bitcoin recipient a merchant. It's p2p, just two anonymous dudes. I don't like any sort of entity that has to build up trust and reputation, because that makes them a big target for the government. It also makes it much harder to do these bets. You can just go out and trade, you have to find entities you trust.

The original proposal tries to avoid trust issues. What's wrong with it?

[just_someguy]

I dunno, to me trusting the other side in the bet is like trusting the other side in a bitcoin transaction. Calling him a bookie is like calling a bitcoin recipient a merchant. It's p2p, just two anonymous dudes. I don't like any sort of entity that has to build up trust and reputation, because that makes them a big target for the government. It also makes it much harder to do these bets. You can just go out and trade, you have to find entities you trust.

The original proposal tries to avoid trust issues. What's wrong with it?

The biggest problem is figuring out what a valid trade is. Using pure block chain activity to determine what is going on will be extremely unreliable.
There is nothing stopping people from trading with themselves to make it appear like any price and volume they want.
If you could make intrade bets at any price and volume without risk by trading with yourself then it would not be a very reliable indicator.
It would be akin to placing a bet that there would be at least x transactions on the bitcoin network on a given day. Anyone can push that number as high as they want.
Also, since the binary event will be decided without real world measurements the person with the most money can force the outcome to whatever they want it to be.

Prediction markets work because at the end of the day an arbiter will make a final call on whether the event did or did not occur.
If there is no arbiter of the event then you are just in a competition to see who can most effectively game the system.

[ben-abuya]

The biggest problem is figuring out what a valid trade is. Using pure block chain activity to determine what is going on will be extremely unreliable.

There is nothing stopping people from trading with themselves to make it appear like any price and volume they want.
If you could make intrade bets at any price and volume without risk by trading with yourself then it would not be a very reliable indicator.

It would be akin to placing a bet that there would be at least x transactions on the bitcoin network on a given day. Anyone can push that number as high as they want.

How is this different than trying to manipulate the MtGox price by trading with yourself? When you make a bid, you're automatically matched with the best offer, not any offer you want. You've now made that bet and you're on the hook for it to someone else. Also, there are transaction fees. This may not have been explained properly in the original post, but making bids and offers firm and matching them correctly is important.

Also, since the binary event will be decided without real world measurements the person with the most money can force the outcome to whatever they want it to be.

Prediction markets work because at the end of the day an arbiter will make a final call on whether the event did or did not occur.
If there is no arbiter of the event then you are just in a competition to see who can most effectively game the system.

There is no person with the most money. They're going up against the entire market. Even the world's most powerful governments get p0wned by the market when they try that. There's reason to believe that since everyone has an external event to rally around and synchronize to, that's where the market will head, and anybody who tries to game it will get crushed, because they're going against the only thing that signals the market. Whether this will work in practice is an open question, but I think it's too early to dismiss it.

[Michiko]

How is this different than trying to manipulate the MtGox price by trading with yourself? When you make a bid, you're automatically matched with the best offer, not any offer you want. You've now made that bet and you're on the hook for it to someone else. Also, there are transaction fees. This may not have been explained properly in the original post, but making bids and offers firm and matching them correctly is important.

Also, since the binary event will be decided without real world measurements the person with the most money can force the outcome to whatever they want it to be.

Prediction markets work because at the end of the day an arbiter will make a final call on whether the event did or did not occur.
If there is no arbiter of the event then you are just in a competition to see who can most effectively game the system.

There is no person with the most money. They're going up against the entire market. Even the world's most powerful governments get p0wned by the market when they try that. There's reason to believe that since everyone has an external event to rally around and synchronize to, that's where the market will head, and anybody who tries to game it will get crushed, because they're going against the only thing that signals the market. Whether this will work in practice is an open question, but I think it's too early to dismiss it.

Mt Gox does not offer really illiquid markets. Is the final closing price of your local college swimming team competition really of interest to enough participants to create a deep, 'non-movable' market? You might have won according to your local newspaper, but someone's bot has gathered sufficient trades (maybe with itself) to make a gain after paying fees? That would according to your rules not even be illegal or wrong, if I understood you correctly?

[ben-abuya]

Mt Gox does not offer really illiquid markets. Is the final closing price of your local college swimming team competition really of interest to enough participants to create a deep, 'non-movable' market? You might have won according to your local newspaper, but someone's bot has gathered sufficient trades (maybe with itself) to make a gain after paying fees? That would according to your rules not even be illegal or wrong, if I understood you correctly?

That's the interesting thing about this stuff. Illegal and wrong become irrelevant, because there's only a priori enforcement. If you can do it, it's ok.

Yes, a lot of the thinking here concerns massive liquid markets with healthy doses of speculators. I'm not sure it would be a good fit for small local bets. But it might work anyway. There's not a lot of money on the table, so is it worth going to great lengths and using sophisticated techniques to try and game it? If there's a sophisticated cheater, why wouldn't there be a sophisticated speculator going up against him?

You can't trade with yourself without going through the whole market first. If you're much bigger than the market, I suppose you could manipulate it, but who knows, maybe there will be speculators who specialize in making easy pickings on these small local bets by gathering online data about them and going against the cheaters. If you were a speculator with a bunch of capital, and you observed all these local bets going wrong, wouldn't you want to get in on the action?

In general, I'm really more interested in the big markets, because those are the ones that are choked by regulation today.

[Michiko]

That's the interesting thing about this stuff. Illegal and wrong become irrelevant, because there's only a priori enforcement. If you can do it, it's ok.

Yes, a lot of the thinking here concerns massive liquid markets with healthy doses of speculators. I'm not sure it would be a good fit for small local bets. But it might work anyway. There's not a lot of money on the table, so is it worth going to great lengths and using sophisticated techniques to try and game it? If there's a sophisticated cheater, why wouldn't there be a sophisticated speculator going up against him?

You can't trade with yourself without going through the whole market first. If you're much bigger than the market, I suppose you could manipulate it, but who knows, maybe there will be speculators who specialize in making easy pickings on these small local bets by gathering online data about them and going against the cheaters. If you were a speculator with a bunch of capital, and you observed all these local bets going wrong, wouldn't you want to get in on the action?

In general, I'm really more interested in the big markets, because those are the ones that are choked by regulation today.

I don't necessarily agree with you here. Conceptually I'd prefer to treat all prediction markets the same whenever possible, giving up somehow artificial classifications. Also, I don't suscribe to that liquid  or large (not deep) markets suffer more from regulation or avoid tinkering if you sit on Digital payoffs. As an example, market making on something like basis spread swaptions might currently be harder than getting matched for a popular horse race on Betfair - still the former quite possible is of more interest to you and  already subject to more regulation.

But to gently move the subject away from storing bets in the Bitcoin block chain - how can an anonymous exchange or escrow agent for conditional transactions perform margining or offsetting? I haven't been able to come up with a clear concept and  I'm not sure how to interpret earlier suggestions (could be in the context of the separated oracle and escrow set-up)?

Most traditional financial exchanges and OTC deals offer some kind of margining or collateralisation. That's not only good for counterparty risk but also helps you fund the bet throughout the life. How can that be dealt with in an anonymous setup, simply running huge offset positions until maturity? I can't even say how the simpler set-up sport betting exchanges take would work: they offer only bets with pre-known maximum loss/win. Then the bettor has to fully fund upfront that maximum liability. The mitigation is that they allow offsetting: you could enter an exactly offsetting bet  a different stakes and they'd recognise you've reduced or removed exposure to the outcome and release your funds prematurely. Both systems - margining and offsetting - seem to rely on knowing the customer - how can that be done for anonymouse exchanges?

(I've mostly unsuccessfully tried around with chains of bets on bets; basically synthesizing Futures by construction sequences of forwards with wide initial margins rewarding the non-breaker, but that creates all kinds of additional problems)

Too many aspects around anonymity/non-regulation and transactions I don't understand, and too much daytime distractions. Please share share more of your thoughts around match making on exchanges and preventing timing issues (prevent double spending by double-signing with trusted 3rd party?), multi-node conditional transaction escrow, advantages/disadvantages of multi- versus single- entity market data fixing etc. I guess you guys have a lot more thoughts on things needed to create a proper prediction market, apart from scripting in the block chain!

[ben-abuya]

I don't necessarily agree with you here. Conceptually I'd prefer to treat all prediction markets the same whenever possible, giving up somehow artificial classifications. Also, I don't suscribe to that liquid  or large (not deep) markets suffer more from regulation or avoid tinkering if you sit on Digital payoffs. As an example, market making on something like basis spread swaptions might currently be harder than getting matched for a popular horse race on Betfair - still the former quite possible is of more interest to you and  already subject to more regulation.

Yeah, I agree that all prediction markets should be provided for, but they don't all have to work well with every system. You could have betcoin for bets that work well with it, oracle betting for others, whatever works. And as I said, it's possible that less liquid bets will do fine on betcoin.

But to gently move the subject away from storing bets in the Bitcoin block chain - how can an anonymous exchange or escrow agent for conditional transactions perform margining or offsetting? I haven't been able to come up with a clear concept and  I'm not sure how to interpret earlier suggestions (could be in the context of the separated oracle and escrow set-up)?

My suggestion was to escrow the full value-at-risk into the trade. Of course that means that the contract must define a clear trading range. The way I see it is you're entering into a binding contract with the other side and the system has to guarantee payment. The only way to do that is to escrow the full amount. You want margin? Convince someone that you know what you're doing and get a loan. Or maybe another market will open up for that.

Most traditional financial exchanges and OTC deals offer some kind of margining or collateralisation. That's not only good for counterparty risk but also helps you fund the bet throughout the life. How can that be dealt with in an anonymous setup, simply running huge offset positions until maturity? I can't even say how the simpler set-up sport betting exchanges take would work: they offer only bets with pre-known maximum loss/win. Then the bettor has to fully fund upfront that maximum liability. The mitigation is that they allow offsetting: you could enter an exactly offsetting bet  a different stakes and they'd recognise you've reduced or removed exposure to the outcome and release your funds prematurely. Both systems - margining and offsetting - seem to rely on knowing the customer - how can that be done for anonymouse exchanges?

Offsetting bets is an interesting question. Clearly it would be beneficial for the system to be able to reconcile offset bets on the same account. I'm not sure if that's possible though, since the contracts would be different and the system would have no way of knowing if the two contracts really do offset each other at all. Again, if you can convince others, or another market, that you have offset bets maybe they'll front you some cash. Maybe there'd be a bet about how closely two other bets are correlated.

Too many aspects around anonymity/non-regulation and transactions I don't understand, and too much daytime distractions. Please share share more of your thoughts around match making on exchanges and preventing timing issues (prevent double spending by double-signing with trusted 3rd party?), multi-node conditional transaction escrow, advantages/disadvantages of multi- versus single- entity market data fixing etc. I guess you guys have a lot more thoughts on things needed to create a proper prediction market, apart from scripting in the block chain!

Bitcoin's greatest achievement is preventing double spending in a p2p system. Unless you know of a better way, might as well copy it. Bitcoin addresses timing issues partly by slowing things down. A block every 10 minutes is pretty slow. I think we can slow down prediction markets too. Prediction markets are made for hedging, not for day trading, so you don't necessarily need millisecond trade updates. Ordering identical-value bids and offers that go into a pool can be done randomly to ensure fairness and prevent gaming.

[Michiko]

ben-abuya - your time to think through these issues and share your ideas with the forum is much appreciated. I find your earlier exchanges in this thread with dacoinminster etc. very enlightning, In my opinion more of these types of 'visions' are required. It looks like rather sooner than later the first prediction market or Futures exchanges etc. annonunced on the forum will hit beta. Without thinking the bigger picture I fear we end up with 'Betfair Bitcoin edition', or 'BTC settled CME Futures' etc. and that would really be a shame, given all this potential right now!

Hopefully you take my continue nagging below in that spirit, hope people on this forum find some more time to explain their ideas.

I personally would love an open market system without accounts, without building reputation or trust (on the user side), without unreasonable regulation by unsuitable regulatory bodies, and most of all without all the time somebody telling what is best for me. Technologically, with the arrival of Bitcoin and all these ideas here, there seems to be quite a bit of progress going on.

[...] The way I see it is you're entering into a binding contract with the other side and the system has to guarantee payment. The only way to do that is to escrow the full amount. You want margin? Convince someone that you know what you're doing and get a loan.
[,,,]
Offsetting bets is an interesting question. Clearly it would be beneficial for the system to be able to reconcile offset bets on the same account. I'm not sure if that's possible though, since the contracts would be different and the system would have no way of knowing if the two contracts really do offset each other at all.

For me margining/offsetting/netting is key to prediction markets in general. There's really no distinction between a hedger with 'legitimate interest' and the 'bad' speculators/investors/day traders just selling to gain. Most hedgers will take both sides over time, in most cases fire-n-forget static hedging is impossible. Also, without the liquidity provided by participants other than hold-to-maturity participants most markets will struggle badly. Markets where you cannot get out before maturity won't attract deep volume. Why forgo additional liquidity if there could be a (technological) solution to have it both ways?

The topic of integrating conditional-transaction settlement in the blockchain has found lots of interest but also a lot of resistance here. I'd love to learn more about ideas how to allow anonymous offsetting or margining. Surely somebody has thought of it?

My take for offsetting would be that maybe you could prove to the conditional transaction settler (who admittedly is then not the blockchain itself) via revealing access to recieving addresses that you will gain a guaranteed pay-off regardless of the outcome, say of long/short positions in identical bets or deals? Then they could release portions of the funds  to the recipients addresses prematurely.

And for margining, if you continue your thoughts about self-settling transactions in the blockchain: in a liquid market you could probably enter into forwards on forwards of ... of the actual deal? And each forward you would give a wide but finite initial margin built into the transaction. The side breaking the chain would automatically lose the remainder of its wide margin to the opponent, who then hopefully can close out in the liquid market.

Anyway, this are just crude and broken fragments. I really need the some of the smart people who discussed earlier in this thread  to come forward and explain things clearly for the benefit of my tired and slow brain...please.

Bitcoin's greatest achievement is preventing double spending in a p2p system. Unless you know of a better way, might as well copy it. Bitcoin addresses timing issues partly by slowing things down. A block every 10 minutes is pretty slow. I think we can slow down prediction markets too. Prediction markets are made for hedging, not for day trading, so you don't necessarily need millisecond trade updates. Ordering identical-value bids and offers that go into a pool can be done randomly to ensure fairness and prevent gaming.

On the same note, I cannot see slowing down a market being a good thing. Somebody trying to make a market on a slow prediction platform will be taken apart by bots arbitraging out on traditional markets I guess, reducing liquidity and increasing bid-ask spreads needed in the first place? I think everybody profits from fast markets - the ones who like it fast can use it, the ones who don't care about speed can trade at any slower speed they want. Isn't that win-win?

The problem is not in the match-making bit of the system (I will try to write a more legible post later today to explain what I understand the 4 core components of a traditional prediction markets are). That can be made arbitrarily fast without Bitcoin confirmations. But as the guy running the exchange how do you guarantee your client that the match that hit his bid will follow through (maybe some kind of double-spending problem in Bitcoin language?) By the time it becomes clear the other one - who doesn't have a fixed account with the exchange - didn't deliver a confirmed transaction the first guy is pissed off because he is now unhedged and the market has moved or whatever his problem now is. You probably don't want to end up absorbing the loss, and you cannot ban the offender because you don't require them to register with you. Waiting for blockchain confirmation is probably fine if you can hold off shipping that flat screen TV, but I don't think it will fly for exchanges?

I think there could be simpler solutions to this particular subproblems, there were loads of discussion long time ago (Snack Vending machine problem),  but maybe you don't need the nuclear option of an upfront escrow or similar techniques. Generally, I like to understand better how one could leverage of the possibilities offered by the existing Bitcoin we have now. I fully understand the resistance people have against changing it without need or without clearly thinking it through. If there is a half-way solution leaving core Bitcoin alone I already would be perfectly happy.