Drivechains is a controversial proposal aiming to allow for the creation of sidechains containing coins meant to be pegged 1:1 to Bitcoin. Here we will analyze what drivechains actually proposes, technical flaws of drivechains, how drivechains could affect Bitcoin itself, and finally, discuss whether or not the idea is actually likely to see significant usage.
Credit goes to Matthew Haywood in particular for productive discussions about the characteristics and flaws of drivechains. He has published his own critique of drivechains here.
- Hashrate Escrows
- BIP-301: Blind Merge Mining
- The Lack of BIP-301 and BIP-300 Interaction
- Analysis: How Can Drivechains Affect Bitcoin?
- Are Drivechains Going to See Significant Usage?
- Final Thoughts
First of all, LayerTwo Labs is paying me $2500 USD for ~16 hours of work to write this blog post. $1250 up front, with the rest due upon completion. The inventor of drivechains, Paul Stzorc, is the CEO and co-founder of LayerTwo Labs. I meanwhile have been a critic of drivechains since the idea was published. In some ways even before the idea was published, as I was also a critic of Blockstream’s (since abandoned) idea of merged-mined, pegged sidechains1.
Paul Stzorc seems to be of the idea that paying me to write this blog post will change my mind:
Thus, it is a waste of time to talk to those three, since after @peterktodd admits he was mistaken those 3 will all either cave, and/or become irrelevant. So I am excited to announced that @LayerTwoLabs has agreed to hire Peter Todd to clarify and publish, in writing, the technical pros (if any) and cons of Drivechain in detail, no holds barred. PT is the original hater of merged mining, since April 2014 at least. -Paul Stzorc, Twitter, Jul 17th 2023
I can assure you, the only thing he bought was my time.
Two BIPs have been published, BIP-300 Hashrate Escrows and BIP-301 Blind Merge Mining, and Luke-Jr was recently paid to write a draft implementation against Bitcoin Core. This post references these sources for its critiques.
Due to time constraints, this blog post can not be a comprehensive analysis of all possible objections to drivechains. While the client, LayerTwo Labs, asked me to confirm that my blog post would contain “all of your criticisms and concerns on Drivechain” they declined to pay me for the time it would take to do that analysis.
To understand drivechains, let’s start with the motivation. Paul Sztorc’s own Hivemind proposal is a good example. Hivemind is meant to be a prediction market, where people can place Bitcoin-denominated bets on real-world outcomes. The specifics of Hivemind are not important here. What is important, is that Hivemind aims to accomplish two things:
- Implement rules that the Bitcoin community is unwilling to add to the Bitcoin protocol.
- Apply those rules to BTC held in escrow.
The challenge here is implementing the escrow. One way is of course to simply use a federation. Blockstream Liquid, for example, implements this by holding the escrow BTC in an 11-of-15 multisig2, with the functionaries who can sign that multisig also signing blocks in the Liquid consensus protocol.
Liquid’s federated multisig is a compromise. Prior to launching Liquid in 2018, Blockstream proposed the use of a hashrate escrow in their 2014 Pegged Sidechains1 paper. However, for reasons we will now analyze, Blockstream gave up on the idea.
Essentially all transaction outputs in the existing Bitcoin system are protected by public key cryptography. Recall that in Bitcoin, specific transaction outputs are associated with scriptPubKey’s, small computer programs whose conditions must be satisfied for an output to be spendable. Essentially all transaction outputs are protected by scriptPubKeys with a form similar to the following:
A transaction spending such a transaction output is only valid if a valid
signature is provided, which causes the
CheckSig to return true. This means
that miners can not simply steal coins from the perspective of a fully
validating node: no amount of Bitcoin hash power can fake a digital
In the pegged sidechains proposal, Blockstream proposed the idea of a hashrate escrow, treating hash power as a dynamic-membership multiparty signature (DMMS). The simplest possible example of a hashrate escrow is the scriptPubKey:
Since this script simply returns true, no cryptographic signature is needed. Such scripts are often called “anyone-can-spend” scripts. However this term is slightly misleading: it’s miners who produce blocks, so strictly speaking, only miners can spend that output.
Of course, if an output can be spent by any miner, surely it will just be stolen, right? Probably! Blockstream proposed to fix this problem with two main mitigations:
- Multisig: multiple blocks would have to approve a spend of hashrate-escrowed funds. Blockstream’s paper suggested one or two days worth of blocks; 144 to 288 blocks total.
- Fraud proofs: once the spend was approved, there would be an additional one to two day waiting period where proofs of fraudulent transfer could be published on the Bitcoin chain, potentially cancelling the spend before it actually happens. This idea is similar in spirit to Lightning’s justice transactions, with the very important difference that Lightning’s justice transactions are actually implemented and have been proven to work in the wild.
In my discussions with some of the authors of the pegged sidechains paper, I’ve been told that one of the main things — in their view — that halted the pegged sidechains project was that actually implementing fraud proofs turned out to be much more difficult than they had hoped.
BIP-300: The Drivechain Take on Hashrate Escrows
The first part of the drivechains proposal, BIP-300, describes a hashrate escrow scheme. As per the BIP-300 introduction:
In Bip300, txns are not signed via cryptographic key. Instead, they are “signed” by hashpower, over time. Like a big multisig, 13150-of-26300, where each block is a new “signature”.
Unlike Blockstream’s proposal, BIP-300 does not have any concept of fraud proofs. As the BIP-300 introduction suggests, BIP-300 is a strict hashrate escrow where miners unilaterally decide if and how to spend the escrowed coins, regardless of what rules the associated drivechain is meant to enforce. This includes spending coins in ways that some people may characterize as “theft”.
While Blockstream proposed relatively short, one to two day, voting and fraud proof periods, BIP-300 proposes much longer timescales of approximately 3 to 6 months. Concretely, to withdraw coins from the BIP-300 hashrate escrow, a miner must propose the hash of what is supposed to be3 a “bundle” of transaction outputs in their coinbase to pay. Following that proposal, in the next 26300 blocks — approximately 6 months — at least 13150 blocks must ACKs that specific bundle in the coinbase transaction. If the bundle is succesful, a transaction may be included in the block spending some or all of the escrowed funds to the bundled transaction outputs.
There’s some more details in the BIP itself, for instance the fact that specific drivechains must first be proposed by a supermajority of hashing power, the downvote/alarm functionality, etc. But the exact details aren’t particularly important to this critique. What is important, is to understand that miners vote to decide what to do with the escrowed coins, that the vote happens over a long time period, and that there is no mechanism to challenge a vote with a proof of fraud/theft.
Other Unimportant BIP-300 Criticism
While an unimportant criticism in the overall context of what we just discussed, I’ll also point out that the inclusion of human readable names, descriptions, tarball hashes, git commit hashes, etc. in the BIP-300 specification is very silly, amateur mistake. Obviously, any real use of a drivechain would not start with non-consensus-enforced information obtained on a blockchain. The inclusion of such fields merely serves to aid in fraudulent attempts at confusing users.
BIP-301: Blind Merge Mining
The second part of the drivechains proposal aims to provide a consensus mechanism. This BIP describes itself as a form of merge mining, with the crucial difference that rather than have miners run full nodes for the merge-mined chain, they are simply paid to blindly commit to that chain in transactions via a fee-bidding mechanism.
Merge mining itself is a very old idea, pioneered by Namecoin in 2010. Essentially a merge-mined coin has consensus rules such a valid attempt at finding a mainchain (eg Bitcoin) block can simultaneously be a valid attempt at finding a merge-mined coin block.
BIP-301 doesn’t actually specify in detail how a blind-mined chain is meant to actually form a consensus. Those specific rules would be part of the blind-mined chain’s consensus, not Bitcoin itself. However it’s notable that for the most obvious and efficient way of implementing those rules, BIP-301 has a serious cryptographic flaw which we will discuss separately.
With respect to Bitcoin, you can think of BIP-301 as a mechanism where mining a
special type of transaction requires a miner to also add a specific
output to their coinbase transaction that commits to 32-bytes.
Since only one such transaction can be included, the economically rational outcome of the mechanism is a fee-based auction: highest fee wins.
As per BIP-301, the result of an accepted Blind Merge Mined block is that an
OP_Return output in the following format is added to the coinbase:
1-byte - OP_RETURN (0x6a) 4-bytes - Message header (0xD1617368) 32-bytes - h* (obtained from Simon)
An obvious problem with this design is it’s subject to equivocation attacks.
The obvious and efficient implementation of a merge mined chain is to use a SPV
proof to the coinbase transaction to prove that a block has been
blind-merge-mined. But since the
OP_Return format doesn’t commit to which
chain is being blind-merge-mined, a miner could in fact include multiple
OP_Return transactions at the same block height. This is similar
to the “double-proof”
problem that Namecoin solved back in 2011.
An obvious solution would be to tag the
OP_Return outputs with some kind of
merge-mined chain identifier, and implement a strict rule consensus rule in the
blind-merge-mined chain where the first matching tag wins.
Reorgs Invalidating Coins
The blind-merge-mine request has the following format:
3-bytes - Message header (0x00bf00) 32-bytes - h* (side:MerkleRoot) 1-byte - nSidechain (sidechain ID number) 4-bytes - prevMainHeaderBytes (the last four bytes of the previous main:block)
The last field,
prevMainHeaderBytes is designed to commit to the previous
block header, ensuring that a blind-merge-mined request transaction is only
valid in a specific block. This directly conflicts with an important principle
of Bitcoin consensus design: reorgs should not invalidate transactions. For
example, this is why coinbase outputs are only spendable after 100 blocks, and
why we have not and will not implement consensus-level transaction expiry.
A way to fix this would be to limit request transactions to containing exactly
one output, an
OP_Return containing the request message.
The Lack of BIP-301 and BIP-300 Interaction
A very notable omission in BIP-301 is that BIP-301 does not interact with BIP-300 Hashrate Escrows. There is no mechanism to pay miners to ACK or NACK a specific BIP-300 hashrate escrow withdrawal proposal. The only way that coins can be withdrawn from a drivechain is by the direct participation of at least a majority of hash power over the 3 to 6 month withdrawal period.
BIP-301 clearly states in the Motivation section that it is intended to fix drawbacks of traditional merge mining, including the fact that in traditional merge mining:
Miners must run a full node of the other chain(s). (Thus, they must run “non-Bitcoin” software which may be buggy.)
But this obviously conflicts with the requirement that miners either run drivechain full nodes, or trust others to run them for them, to safely allow withdrawals to happen. If miners are not doing that, either coins become stuck, or they become subject to theft.
Frankly, when I first learned about drivechains I had assumed that the blind-merge-mine mechanism also allowed for blind withdrawal proposals of some kind given the decentralization claims around drivechains. I was shocked when I learned otherwise. The proposal simply punts on this obvious issue, with no solution.
We’ll discuss the implications of this omission further.
Analysis: How Can Drivechains Affect Bitcoin?
Now that we’ve covered how the drivechain proposal works, along with some more obvious flaws, we can analyze how drivechains can affect Bitcoin. Let’s start with our goals. We want Bitcoin to be:
- Useful: Bitcoin must continue to allow people to provide a store of value and medium of exchange, which collectively comprise the vast majority of what people use Bitcoin for.
- Secure: Bitcoin must continue to be useful even in the face of attack.
- Decentralized: in achieving goal #2, we do not trust any single entity or group of entities; we assume any single entity or group of entities may attack Bitcoin.
The last goal has generated the bulk of the discussion around drivechains: do they encourage centralization of Bitcoin mining?
Bitcoin mining comprises of both hashing, and block verification and production. The economics of Bitcoin mining have both economies and diseconomies of scale. These economies and diseconomies of scale act as centralizing and decentralizing pressures.
The primary diseconomy of scale is the fact that Bitcoin hashing consumes energy and produces heat. Opportunities to obtain cheap energy, and to dispose of or make use of waste heat, have inherent diseconomies of scale due to the fact that the cheapest energy production is inherently spread out across the globe. For example, flare gas and stranded/surplus renewable energy are some of the cheapest sources of energy available. Obviously, opportunities to collect flare gas and renewable energy are inherently spread across the globe. This is a diseconomy of scale because the maximum size of these operations is limited by the available energy at any one location. A oil well that generates 100KW of free waste flare gas in a given location simply can’t generate more free waste gas at any reasonable price.
Similarly, some hashing operations make use of their waste heat, eg to heat buildings. Obviously, the amount of waste heat that any one location can make use of is inherently limited. Again, this acts as a diseconomy of scale: a house that can be sufficiently heated with 50KW of hash power simply has no use for more hash power in that location.
Grid stabilization is another example of a diseconomy of scale. Some Bitcoin hashing operations have been able to get credits on their electricity bills for providing the ability to quickly reduce their power usage. Obviously, the maximum amount of grid stabilization needed in any one geographic area is limited, creating a diseconomy of scale.
The primary economies of scale are in mining pools. They exist because of variance: with just 144 Bitcoin blocks generated per day, the average Bitcoin hashing operation mining blocks directly for themselves (solo-mining) would need to wait months or even years between payments4. Secondly, the operating costs of a mining pool are mostly fixed: the bandwidth and compute power needed per hashing operation is almost zero. If the overhead costs of operating the necessary nodes is non-trivial, there is a significant savings for a large pool compared to a small pool as that overhead is spread over more clients.
Thus we see that while Bitcoin hashing is quite decentralized — with a very large number of individual hashing operations in existence — the number of Bitcoin mining pools is much more limited. At the time of writing, the majority of hash power is spread over just two pools:
While undesirable, this situation isn’t fatal as, in theory, bad behavior can cause hashing operations to quickly switch from one pool to another. Additionally, there are technologies like StratumV2, p2pool, and braidpool that decentralize block production (mining).
While we don’t have space to go into details about these technologies, the important thing in the context of drivechains is that they all require hashing operations to produce blocks themselves, requiring them to run full nodes.
Do Miners Need to Run Drivechain Nodes?
The reason why is simple: if drivechains are to work correctly, coin withdrawal needs to work. Drivechains requires that a majority of hash power directly vote on withdrawal proposals, and the only way to validate a withdrawal proposal for a given drivechain is to run a node for that drivechain.
Since there could be dozens, or even hundreds, of drivechains this is an obvious economy of scale that will significantly centralize mining: every one of those drivechains will have associated software that needs installation, maintenance, and upgrades. Those drivechain nodes will require bandwidth, CPU time, and storage space.
This is why the fact that blind merge mining and coin withdrawal votes are unconnected is such a glaring omission: while blind merge mining claims to fix the mining decentralization problem, the fact that withdrawal votes require direct majority miner involvement negates this claim.
I suspect the reason for this omission is that the authors of drivechains attempted to make blind withdrawal voting, and failed due to it being incompatible with rational game theory: if you can pay money to steal coins, obviously someone will do exactly that. Requiring a majority of hash power to vote in favor of a coin withdrawal obscures this problem.
The fact that miners need to run drivechain nodes is particularly problematic to the claims that drivechains will scale Bitcoin. Obviously, if drivechains supporting hundreds or thousands of transactions per second catch on, miners will need to invest in thousands or tens of thousands of dollars worth of computing hardware and bandwidth to mine. This is completely incompatible with technologies that decentralize mining such as the aformentioned StratumV2 and p2pool/braidpool.
The “Phone a Friend” Model
To avoid the necessity of miners running drivechain nodes, it has been suggested that miners could instead just “phone a friend” and just copy withdrawal approval hashes from “trusted sources” like social media. There’s a lot of problems with this model:
- Why should miners do anything at all? There’s no direct incentive to approving withdrawal requests, so miners have no reason to actually go through these steps. This is especially true in the case of decentralized mining tech such as StratumV2 and p2pool/brainpool. I personally used to mine with p2pool, and I never touched my configuration after setting it up.
- Workload: drivechain proponents imagine a future with dozens or even hundreds of drivechains. With a 3 to 6 month withdrawal period, that would require miners to research and input NACKs and ACKs for drivechains every week or so. Even without actually running drivechain software that’s an enormous increase in workload compared to the status quo for running a full node, where you might want to change software every few years for new soft forks.
- Drama and legal risks due to failures: given how broken this system is, it’s inevitable that some drivechains will fail via funds getting stuck or stolen, and pressure will be put on miners to fix the problem.
The last is a particularly serious problem in light of Bitcoin’s high degree of mining centralization: a 51% majority of hash power can salvage a drivechain failure by 51% attacking other miners to overrule their votes, or lack of votes. We do not want there to be more incentives to 51% attack!
A key thing enabling miner decentralization is the fact that at the moment it is still reasonably possible to form a new pool, and to use decentralized mining pool tech. The rules to create a new, valid, block are clear, and provided you follow those rules, you are not risking enormous financial loss by using a small pool or decentralized pool. If these rules become unclear because drivechains funds are being frozen or looted, that’s an enormous centralization pressure inducing hashing power to move towards the largest pools.
Legal Risk: Do Drivechains Make Miners Custodians?
At the moment miners are not in a position where they can steal arbitrary funds. As long as people run full-nodes, even a 51% attack can’t steal arbitrary coins since no amount of hashing power can fake a digital signature.
Drivechains fundamentally change this. In the drivechain model, miners absolutely can steal funds. Does this make miners custodians? Maybe! It certainly invites lawsuits and court orders against miners to recover or seize funds. I personally am being sued by Craig Wright in a lawsuit aiming to do just that. I’m glad that I can confidently tell a court in my defense that I simply do not have the ability to do that. With drivechains, miners won’t have that defense.
This also means that voting “incorrectly” could open miners to legal action. This is particularly problematic in the “phone a friend” model of operation where the miner isn’t making a genuine attempt to actually validate the drivechain. It’s easy to imagine courts deciding that operating in this way is negligence.
Fees and Drivechain Blocksize Limit
It’s been claimed that drivechains will benefit Bitcoin miners through higher fees generated via blind merge mining.
However, drivechains have no mechanism to limit total blocksize, and indeed, they’re being promoted as a way to dramatically scale Bitcoin. Without a limit, it’s obvious that the supply of drivechain blockspace is unlimited, and thus there is no reason to expect economic supply and demand to converge to anything more than trivial fees.
Note how this particular claim is reminiscent of big blocker claims from the blocksize wars.
Time Value of Money
If drivechains do somehow generate non-trivial fee revenue, that can actually centralize blind merge mining by greatly increasing the capital costs requird to blind merge mine! The problem is that the creators of blind merge mined blocks have to pay transaction fees now, denominated in Bitcoin. But they receive the transaction fees in the drivechain coins. In theory those coins are supposed to be pegged 1:1 to BTC. But obviously, with a 3-6 month withdrawal period that is certainly not guaranteed to succeed, paying those fees now requires large amounts of liquidity and risk. Shinobi has a good article explaining this problem.
Sztorc has argued that swaps fix this issue. But as Shinobi explains, “the reality is that [swaps] just shove the liquidity requirements onto yet another party, assuming they will provide massive amounts of liquidity for almost nothing in return”.
Note that this isn’t directly a problem for Bitcoin users at first glance. But drivechains introduces new incentives to Bitcoin mining, and it would not be surprising if this blind merge mining centralization problem also spilled over to Bitcoin mining.
Are Drivechains Going to See Significant Usage?
In any discussion of a Bitcoin soft-fork, we have to weigh costs and benefits. Any soft-fork has significant costs, including technical risks, maintenance burdens, etc.; drivechains are not a trivial feature to implement due to the state required.
The history of Bitcoin and altcoins has shown that the overwhelming, long-term, non-speculative, market demand is for the straightforward use-cases of secure store-of-value and cheap and convenient medium-of-exchange, along with a smaller niche occupied by Monero’s enforced privacy5 use-case. Of those three, drivechains clearly do not meet market demand for a secure store-of-value: it is obviously risky to leave coins on drivechains. Drivechains are also not needed for enforced privacy.
That leaves the cheap and convenient medium-of-exchange use-case. Unfortunately, drivechains inherently pose significant medium-of-exchange problems by adding to the exchange rate volatility problem. It is already problematic enough for the medium-of-exchange use-case that Bitcoin has a floating exchange rate relative to other currencies. While drivechains are meant to be 1:1 pegged to Bitcoin, due to the long peg-out time windows and inherent risks of holding drivechain coins, drivechain coins will have a floating exchange rate that is likely to slip by significant amounts. Conversions between drivechain coins and Bitcoin will also not be practical without using swaps, and thus incurring transaction fees on top of exchange rate volatility costs.
This is an obvious UI/UX problem for drivechain wallets! We already see people often preferring custodial Lightning wallets over sophisticated non-custodial wallets like Phoenix due to their ability to avoid on-chain transaction fees. The transaction fees and exchange rate volatility drivechain coins will experience is an even worse version of this problem.
Writing this blog post has proven to be surprisingly difficult for me. Usually when I’m asked to review an idea, there’s a reasonable amount of “substance” to the idea. Concrete technical concepts and game theory that can be distilled down to logic and math, and critiqued with logic and math. For example, my 2016 post on Block Publication Incentives for Miners.
Drivechains is not like that. Drivechains is a vapid, handwavey, idea that replaces the careful incentive design we see in other Bitcoin protocols — and Bitcoin itself — with blind trust in miners. Of course, blind trust could very well work in practice! But if you’re that willing to rely on trust, why are we going to the trouble of using Bitcoin in the first place?
Enabling Blockchain Innovations with Pegged Sidechains, Adam Back et al, 2014-10-22 ↩ ↩2
Specifically an 11-of-15 multisig with a second set of emergency backup keys that become active after a 28 day expiry period. ↩
Homework question: what happens if the bundle hash is invalid? ↩
A good real world example of the cost of variance is how Riot switched to a larger pool due to liquidity problems. ↩
While Bitcoin can achieve good privacy with proper use of technologies like coinjoin and Lightning, the design of Bitcoin does not mandate the use of those technologies; many users in practice use Bitcoin in ways that have very poor privacy. Conversely Monero’s consensus algorithm and widely used wallets make it quite difficult to bypass Monero’s coinjoin-like privacy tech, enforcing privacy. Thus entities like darknet markets who want to avoid their customers getting doxxed have good reasons to accept Monero. ↩