I figured I'd start a new thread for this despite that there are a few similar threads already, as I wanted discussion of this exact design all in one place. I will incorporate feedback and put it onto the wiki in the coming week. These ideas are of course not really mine, they have come out of lots of different discussions and debates. But I want to try blending them in a different way, with more technical detail.

"Bitcoin passports" are related to fidelity bonds discussed elsewhere in this forum, but I think it's a slightly better sounding name (if a little misleading).

---

Services that can be abused (wikis, forums, webmail etc) need a blacklistable identity for each user so bad behaviour can be punished through exclusion. If no such blacklistable identity exists then account termination is pointless because the abuser will just immediately sign up again. This is why Tor tends to have a read-only view of the internet. Mesh networks like cjdns will have the same problem. Small sites tend to use IP address but this is weak; more advanced operations like the Google abuse team use phone verification and other techniques.

For anonymous users to take part, a cost on establishing an identity must be imposed. We can do this by requiring users to present a proof that they gave away some coins to miner fees. The proof is standalone and can be checked by anyone who has a copy of the block headers. The hash of that proof becomes the pseudonym, it can be checked against Spamhaus style DNS blacklists to discover past behaviour that has been associated with that nym and optionally blocked.

To create one, you take X coins and spent them to miner fees over the course of Y consecutive blocks. The outputs of the transactions are set to OP_FALSE so nodes can be taught to prune them. Each transaction and its merkle branch are included into a data structure, called the passport:


If a service or application wishes to verify a passport, it asks the operating system to open a URL of the form:


An application, call it the Passport Manager, handles this URI and then signs the given challenge using each public key in each sacrifice transaction (in order). The challenge is randomly generated by the server. The signatures are then bundled up together with the state and the passport itself:


The state parameter is arbitrary data looped through the client from the server. It allows for a stateless server implementation that does not need a database to track what challenge corresponds to what server action is currently suspended pending verification. For instance, the server could encrypt any state it requires into this field. It is useful for operations that use global load balancing.

Upon receiving an IdentityProof, the server parses the passport field into a BitcoinPassport message, verifies the signatures (the passport is presented by its owner), verifies that the sacrifice transactions link back to the given block headers based on the servers copy of those headers, and then verifies that the number and value of the sacrifices is sufficient. The transactions should appear more or less in consecutive blocks the chain, though there should be a bit of leeway allowed.

If the proof is validated, the server can then check against its own blacklists of abusers or look up the hash in a remote database. If no problems are found, the nym (proof hash) can be stored in the newly created account. If later the account turns out to be abusive, it can be blacklisted. The state variable can be decrypted and whatever data needs to be adjusted then can be. The server replies to the app with an HTTP redirect and the app then opens the redirect URI, which takes the user back to the browser. The redirect URI may complete registration with a welcome message, or alternatively explain to the user why their nym is not accepted.

Passports can be created using the Passport Manager app. It is a standalone program that maintains its own Bitcoin wallet. You send money to it, click the "Make passport" button and then go to bed. It ensures there are outputs of the right quantity and value before starting, and then sends transactions such that they get into consecutive blocks. This may take an hour or two, but once complete the passport can be written to a .passport file which contains the public data and of course, the private keys used to sign for them.

Design choices

(1) Why miner fees instead of a charity?

There are several problems with the idea of donating to charity:

• Charities are not politically neutral as miners are. For instance, if you wanted to donate to charities that operate in Iran that might be an issue. Nobody will be able to agree on one set of charities that makes everyone happy - even "no brainers" like open source foundations would be a turnoff to non-technical people who may wish to use the system.
• Charities massively prefer to have real relationships with donators who commit to the long term (or can be convinced to do so). Anonymous, unpredictable revenue flows can make their accounting much harder because they don't know when money might arrive, so it can't be relied upon. A charity that received money this way then didn't use it would lose a lot of goodwill through no real fault of their own.
• Some charities do not want to or cannot accept anonymous donations at all.
• Everyone would have to agree on the list of charities to use ahead of time, so there would be no connection between how well a charity does and how much it received in sacrifice fees. A charity may operate effectively for a few years and then become corrupt or incompetent, but upgrading everyones software to use a new list would be difficult.

Miners are politically neutral, anonymous, indistinguishable, the money is almost guaranteed to quickly re-circulate into the economy via exchanges and it helps support network security.

(2) Why require multiple sacrifices in consecutive blocks?

This is to avoid people mining their own sacrifice. It's very hard to mine 6 consecutive blocks but relatively easy to get transactions into them. A gap of 1 block or so would be acceptable, you don't need to hit every single block to have very high confidence.

(3) What are good values for the size of the sacrifice and the number of blocks?

Abuse is cheap. Let's imagine we deploy this for Gmail. The cost of Gmail accounts on the black market is roughly $100-$150 per thousand accounts. This is what it costs professional spammers to overcome phone verification and successfully avoid bulk termination.

Thus a single account costs about 10 to 15 cents. Assume a price of $20 per coin, that means 10 cents is 0.005 coins. If we say you need to spend 0.005 coins to fees in the space of 6 blocks, then this means 0.001 coins per sacrifice with an allowable gap of 1 block in 6 total. That sounds quite reasonable to me. Because you know when the sacrifice was made, calculating the value of it in the currency of your choice is straightforward despite floating exchange rates.

(4) Why a new URI scheme?

For a website to ask an app for data, there are only a few possibilities. Magic HTTP headers, JavaScript APIs etc all require browser extensions to implement, which is hard or impossible to do on mobile/tablet browsers. URI registration is simple, universally supported and does not impose any browser requirements.

Yes, I saw that it mentioned it. I thought it was an inadequate reference without the link. Thanks for adding one.

I can't say that I'm a fan of the new name you've given this but I'm not the person to ask. It'll be interesting to see what other people think.

The consecutive block requirement is a bit ugly, as it requires N transactions instead of one or two and the provided fees will need to be somewhat high to get an acceptably gapped run with high confidence.

What advantage do you see in using a consecutive block run over the two transaction announce-commit sequence promoted by Peter Todd?  What he proposes would have less pre and post pruning blockchain data, would result in passports 1/3rd the size, and I believe it would be less likely to fail for reasons beyond the user's control.

The dummy output could be completely eliminated by instead authoring the transaction as an anyone can spend and putting part of the fee there.

If you fail to provide a mechanism for transferring ownership of passports it means that there is no economically rational motivation to not violate trust using an identity that you're done with (e.g. as you rotate them out for anonymity purposes). If the passports can be resold while in good standing the problem is removed so long as a liquid market exists.  If they are only weakly transferrable (E.g. I give you a couple private keys that I continue to own) then the only place to sell a used bond is to blackhats.

I notice you seem to be describing this using protobuff. If you hash a representation without a single canonical encoding then users could escape blacklisting by changing their encoding.

More thoughts later— I'm glad to see other people looking at this.

This makes me wonder. Would this naturally evolve over time into its own crypto currency? You would just sign a public statement with a private key making a prediction about the way in which bitcoins will be destroyed. If your prediction was the first and accurate than it would be considered by the network to be legitimate and stored in a distributed database. (because there is no reason not to allow people to store copies of the database in a distributed manner) Hopefully the end result would be a product that felt a lot like bitcoin. Where you could transfer the value you had assigned to private keys (in this case by "destroying" bitcoins) just like you do with bitcoin. But you would also have the option to send "insurancecoins" into, what would be functionally equivalent to, a joint account to be held in escrow by the distributed network and its protocol until such a time that both parties agree to the release of the funds.

the core of what we are really addressing is the fact that the community lacks a decentralized and anonymous form of escrow.

The problem that I have with fee values is that the purchasing power with Bitcoin is not fixed - a passport created 3 years ago might have needed to burn 100's or 1000's of BTC to claim legitimacy; requiring any application to have some sort of exchange rate lookup as part of the passport application.

I would think it would be better to have some adjusted-purchasing-power token that you can purchase, and then those could be traded on an alt-chain explicitly (kind of like in Anon136's comment above.) Since it lives outside of Bitcoin there is much more flexibility to the rules that can be enforced explicitly to implement contracts and the like with real power, instead of a web-of-trust method of servers checking hashes or polling for fraud notices that live inside the Bitcoin block chain. It is transferable so you don't have to worry (too much) about key compromise destroying the whole reputation.

I posted a thread on this system a few days ago in Repcoin: a decentralized reputation currency.

Unfortunately I think it's not really the right audience in Alternative Cryptocurrencies, and I haven't really gotten any good feedback, even a 'you suck and this system sucks' would be valuable though ...

I'm not really happy with it either, but I think it's better than fidelity bonds: my issue with this name is that both "fidelity" and "bond" (as a financial concept) are not terms commonly used in English. Fidelity tends to have connotations to sexual loyalty even though it can be used more generally, so it could be confusing to have it crop up in another context. Also whilst anyone who is interested in finance knows all about bonds, most people have never actually experienced one themselves.

Passports are an everyday concept, and everyone immediately understands that you present them to get into places, they are revokable, they are hard to obtain and forge, etc. The main sticking point is that "anonymous passport" is something of a contradiction, but I feel it can be workable.

Still, I'm sure better names are possible.


Seeing as the transactions are all-fee anyway, miners have no incentive to not include them (unless Bitcoin turns into a network for high value transactions only of course). I'd think you could set a threshold, say 4 of 6 blocks must include them, or something like that. The right gap thresholds could be determined empirically.

I think the main advantage of Peters earlier suggestion for semi-consecutive blocks is simplicity - there's no need for new logic specific to this app in the miners, though auto-pruning OP_FALSE transactions would be a neat optimization for nodes that upgrade. I'm not a big fan of having app-specific stuff in core Bitcoin ... the IsStandard() stuff being a necessary evil in my eye rather than something to strive for. Besides, embedding a transaction inside another transaction seems superfluous (the second tx could just be broadcast and sit in the memory pool for a while).

Post-pruning size with OP_FALSE outputs is zero, is it not?


Ah yes, so you did. I must have forgotten the consecutive tx trick was originally from you as well.

I'm not worried about the size. As I said, all-fee transactions should be completely prunable away. You don't need a whole lot of blocks, just a few would be sufficient, so gaps are OK. I'm not worried about miners who manage to mine the majority of all blocks within the consecutive-blocks period because if that could happen we'd have a 51%+ attacker, and then we'd all have bigger problems than this.

I'm not sure minimum fees will be much larger in future. That's certainly not a consensus anyway. I see fees being similar, lower or even being zero for most casual users, that could happen in a world where miners are funded through assurance contracts of participants who need the security (merchants, exchanges, traders etc).

Being able to re-sell the identities (without changing them) seems important to me. People don't like waiting. Any proposed scheme has a long lead-time to the manufacture of new passports. Having dedicated sellers that keep some in stock means you'd be able to get started right away instead of making them yourself. Yes, it means you have to trust the seller. I like your suggestion of making passports smart property, that could be a useful extension indeed. Having each sacrifice tx have a control output that is then spent by the next sacrifice in the chain seems like the way to go, then you can just use the last control output to sell the whole thing.


The app that checks the passports needs the block headers anyway. It can always record spot rates along side its storage of the headers. There's no trust-free way to bootstrap that, but exchange rates are a matter of public record and would be hard to forge in any significant way without someone noticing that it's historically inconsistent.


Chris, thanks, but as you can see from this thread most of the thinking here is done by retep (Peter Todd). I wanted to flesh out more precise specs so we can put it on the wiki, like I've done for other protocols, that way it's possible that someone who is more of a coder than designer has good documentation they can work from. When I did this for smart property and distributed bonds, it did indeed encourage someone (jgarzik) to go ahead and implement it.

I posted my first draft of something approaching a proper tech spec for this fidelity bond/contract stuff: https://github.com/petertodd/trustbits/blob/master/fidelitybond.md


Keep in mind that the term "fidelity bond" isn't just a finance term, it also crops up in stuff like security guard jobs. That's actually the first place I heard it, talking to someone explaining how they were now a bonded guard. That said, more specifically for what you are talking about re: wiki and similar services, I think "passport" is reasonable enough. It's also good that it has the real-world connotation that acceptance of a passport is not absolute. Similarly what is blacklistable behavior is ultimately a human criteria, and you'll find that rough consensus emerges among groups of mutually trusted services, but a passport revoked by Wikipedia may not be considered revoked by 4chan.

Fidelity bonded banking has the advantage that the contracts can be made machine readable. Of course, equally, the disadvantage is that the contracts are machine readable... One little mistake...

As for the term "sacrifice", I think this is a really good one for describing the nitty gritty of any of these protocols; I was racking my brain trying to come up with a decent term and I think sacrifice solves that problem.


It's not just about incentive, what happens when you get a near orphan by pure luck? We already see transactions getting passed up by miners who missed them for whatever reason all the time. Yet as gaps increase, it just makes it easier for a relatively low hash rate miner to just wait until they get lucky, while only risking the small chance that their block is orphaned and another miner picks up the tx for the fee.


How do you prove that a tx existed in the memory pool after the fact?


I'm pretty sure any script ending in OP_FALSE is guaranteed unspendable. Even the OP_IF stuff has a check (via vfExec) to ensure that an IF block is terminated; there are no op-codes that mark a script as valid and terminate execution. OP_RETURN at the end of the script would work too I think.


It's not just blockchain size, it's proof size. You have to be able to prove to a SPV client that the transactions exist, so you have to hand them the transactions and the merkle path to the block header. In addition a malicious miner can currently pad transactions to a maximum of 10KiB scriptSigs.


Again, you don't have to have 51%, you just need to get lucky once every few months.


Well, you're assuming the block size will be lifted which I'm dead set against.


Being able to sell part of the cost of an identity, as you suggest, fixes the problem of fee values just fine: sell off the part of the fee that you don't need. It's a lot nicer than the huge pain of trying to record spot rates, although having said that you will need some estimate of the liquidity and supply and demand for the current market for fidelity bonds for any given application.



Well, I think this touches on one of the reasons why I'm 100% against raising the 1MiB block limit, especially with the sort of floating limit that Gavin and others have proposed: you can't have really trustworthy fidelity bonded transaction processors, and unbounded block sizes at the same time. It's in miner's incentives to do what they can to undermine the off-chain transactions to try to get as many transactions as they can for themselves, and just by increasing the block-size they achieve that goal by further centralizing who can verify the utxo set.

Keep in mind it's not even just a total blockchain size issue, it's a bandwidth issue. 1MiB/10minutes is just 2KiB/s, small enough to keep up with even on highly bandwidth constrained nodes stuck behind restrictive firewalls in unfriendly countries. A node operating under these conditions may never need to actually make an on-chain transaction, but they will be able to easily keep track of fund movements for the fidelity-bonded transaction processors they are depending on to move their Bitcoins, and they'll be able to keep track of those movements with a much higher degree of certainty than with any of the utxo proposals, especially because that one single node, which might be the only connection to the blockchain for a big area, can redistribute everything every other node in that area needs to know. And again, in an unbounded blocksize world miners have every reason to sabotage UTXO mechanisms if they'll stop off-chain transaction processors from operating successfully.

I'll write more on this later, but increasing the block size is madness. I thought people were still taught how ugly O(n) and O(n^2) scaling was in school...

The thing is that you missed one obvious choice: paying the bitcoins to the forum/email site/whatever itself. That may actually be the best option of all, as there is an argument that the reason why sites are so eager to infringe upon our privacy is that advertising revenue is the only revenue they have, and so if we can convince users to pay a 15 cent fee to the website for setting up an account we may end up fixing not just the spam problem, but also the privacy problem at the same time.

If we do not want to do that, I would prefer if there could be some way to dual-use your mining spending - that is, make your Bitcoin passport out of a transaction fee that you are spending anyway to make a given transaction go faster. Then, the rate could be increased significantly, and the only users that would have to actually pay anything extra are those creating multiple passports, with normal users spending nothing extra at all.

This is a really interesting idea.

Ha. Might be a geographic thing; I'm in Canada. Of course, security is a niche field ultimately...


Someone else should chime in with the math to work it out explicitly - I know way too little about probability math - but can you see how the probability aspect of it makes hash power much less important than the two-step publish commit?


Right, but if that node trying to verify the sacrifice didn't see it in the mempool, there is no way to prove it ever was in the mempool after the fact.

I sat down and (half) wrote a god-damn paper this weekend analyzing this stuff properly. I'll finish it up and make it public if anyone is actually interested, but the process absolutely convinced me that the two-step tx-in-a-tx solution is the only thing close to feasible that will keep a stable sacrifice market value. (without just sending coins to unspendable txouts of course)

First of all you have to assume that services will pop up to create these sacrifices for you. Thus you need to assume the sacrifice is being created by the entity with the largest single concentration of hashing power, and that's likely to be in the range of 10% to 30%.

Secondly because of that, for any mining-fee-based sequence, the first transaction is always nearly free. You can always mine it yourself by waiting a bit, and the only cost is the approximately 1% or 2% probability that the block gets orphaned and another miner gets the tx fee instead.

This means that asking for "two blocks n apart" is at best 50% efficient, because you have to assume the first block was self-mined. Secondly mining is a random process, so asking for exactly 1 block apart or 100 makes absolutely no difference.


Sure, but this is a supply and demand problem. Let ω be the probability that the block a transaction is in is orphaned, and the transaction is mined by someone else. If we attempt to mine a transaction containing a fee of value V ourselves, the cost to us is then V*ω. Assuming that miners always build on the best known block and don't deliberately attempt to orphane blocks to collect the transactions within ω is probably approximately equal to the orphan rate itself, which appears to be around 1% or so. Also note that for a large miner ω is less than for a small miner.

Now if I control q of the total hashing power my expected number of blocks before I get n consecutive is ((1-q)^(-n)-1)/(1-q) For 10% I need 1 million attempts for n=6, however the number of attempts drops extremely quickly as q and n decrease. Basically that means if there exists q and n such that ((1-q)^(-n)-1)/(1-q)*ω < v/V, where v is the actual value of the sacrifice and V is the face value, you are better off attempting to mine the sacrifice yourself rather than buying it fairly. For q=25% and n=3 this is true for v=V. (remember that you can always finish the other blocks in a sacrifice by the conventional way) Allowing for n-of-m blocks just makes the problem worse, because it effectively increases the apparently hashing power available to the miner. Yet at the same time griefers can do a lot of damage by deliberately excluding your sacrifice transactions if n-of-m is not allowed, and additionally orphans quickly push up the cost of sacrifices. (about 10% extra for n=6 and 1% orphans)

Is this really an issue? Well, I think the the question really is why would you use a sacrifice method where the value is so sensitive to so many variables? In particularly, a method where the actual cost of the sacrifice is inversely and exponentially dependent on the size of the largest miner.

My later tx-in-a-tx proposal is far easier to reason about because the cost is simply bounded by the largest miners hashing power and has no dependence on difficult to measure values like orphan rates. It's just a much better idea, which might explain why it took me 6 more months to think of it...


But the tx-in-a-tx sacrifice isn't valid until the second tx, the one that actually sacrifices coins, is spent so the tx being in the mempool is a non-issue. Anyway sacrifices need to be provable to SPV nodes who don't care about the mempool.