Proof-of-Approval: A Better Blockchain Consensus Protocol

[shunsaitakahashi]

Hello Fresheneesz,

Am I right that the goals of this protocol are primarily to reduce the finality time in comparison to Bitcoin as well as reduce the cost of running the system?

Yes, the goals are to
1. Improve security by achieving near 50% adversarial stake tolerance. Near 50% tolerance would offer deterrence even more than the corresponding mining power. Only Cardano achieves that (not counting TON because I have no knowledge of their protocol).
2. Improve finality. The fact that Ethereum is trying to get Casper FFG shows how important finality is (although Casper FFG only improves tolerance to long-range attacks). Cardano's finality degrades continuously with rising adversarial stake.
3. Reduce cost. AFAIK, Bitcoin's breakeven (cost of mining vs its price) is around $8050. That's a lot of expense for the node owners.

The image of how blocks are set up (https://cdn-images-1.medium.com/max/1000/1*fPsB4-t6J_1u1UpF37dYHw.png) reminds me strongly of the way blocks are chained in the "two-hop blockchain" (https://eprint.iacr.org/2016/716.pdf). I would like to see more in-depth comparisons to other consensus protocols, especially DPoS which seems particularly close in mechanism to Proof-of-Approval.

Thanks for pointing it out - I had not seen that before. I will definitely offer comparison even if they are simply superior

Your comparison table claims Proof-of-Approval is fully finalized after a single block,  however, you also talk about how there may be multiple top-level blocks where the "preferred" chain is unresolved. Doesn't this imply that your transaction being inside a valid block doesn't mean its been fully finalized? How do you reconcile that?

In order to be fully robust, the fork selection has to work on all conditions that may be produce in normal operation. It is not a detriment that the fork selection actually works even beyond those conditions. It's just a good design practice to design for more than needed (within cost constraints of course).

Since any node can produce a candidate block, this seems like it would potentially cause unbounded network traffic with everyone and their mother suggesting blocks to vote on. How is this resolved in the case of, say, an adversarial ddos attack?

Traffic can be a problem but only the headers (which are few hundred bytes) are being transmitted. Still, a node would need to validate the blocks of headers it receives (after downloading full blocks). Having a smaller time-slot (as opposed to larger) will reduce the number of headers received by a node (assuming the software's low level code simply drops headers received in mismatched time-slots). An adversarial ddos attack may look like network fragmentation and may prevent blocks from being added to chain temporarily.

But I have the same question as monsterer2, why would anyone with stake approve the block of anyone without stake? In fact, why would anyone with stake approve *anyone* else's blocks? It seems there's no incentive to mine on top of someone else's blocks. This could manifest in more subtle attacks where you have stakeholders who refuse to mint on top of a competitor. It could also manifest in deadlock as nodes try to maximize their probability of winning the minted block.

Yes, and there may be a need to reward block approvals. I will work on adding block approval rewards. They can be added to approval-block since the content-block has already been created.

I have a hard time seeing the purpose of having nodes pay attention to timestamps and comparing them to their receive-time. At its core, the mechanism is just voting, right? I don't see why the protocol doesn't just let nodes vote for whatever blocks they want. I see the appeal of a simple stake-voting process for block creation (similar to DPoS), but the Proof of Approval method seems like rational profit-seeking minters would make things devolve into the top 50% stakeholders passing around permission to mine the blocks, completely excluding the other 50%.

Nodes are allowed to rank in any order they choose. The time is just the default order built into the software.

Also, could you provide a glossary of variables you use in this paper? Its hard for me to keep track of which variables mean what and find their definition in the prose when I need to refer to what they mean later.

Good point, I will add glossary.

Thanks,
Shunsai

[shunsaitakahashi]

Hello Aliashraf,

Just read the article, nice work but, as one of the guys have mentioned earlier, the 'proof of something' discourse is exhaustively saturated and it is hard to attract enough attention from the community these days for a new one.

I choose Proof-of-Approval name simply because it implies to have something to do with approvals. I could have chosen "Takahashi Protocol" name (similar to Ripple Protocol or Ouroborous) but that name wouldn't have told the reader anything other than the creator is named Takahashi.

I'm not satisfied yet with the protocol itself as it seems not to be secure against the bootstrap poisoning attack...

Protocols that use "older" stakes (stakes from many blocks ago) are susceptible to bootstrap poisoning attack. In this protocol, since the stake used is at the current block, I have not been able to come up with a successful bootstrap poisoning attack scenario.

...I have not found strong enough game theory and incentive analysis (why people should approve  each other's blocks) and I do not like the idea of zero cost block generation..

Multiple people have mentioned it and I am analyzing giving rewards for approvals.

The way I see the situation in crypto world, we will be stuck in wasting resources and jeopardizing performance and scalability  as long as we are seeking a solution for this problem: consensus.

The reason the crypto world is focused on consensus is because in absence of a good protocol, all other blockchain properties are compromised. Bitcoin is said to have breakeven of $8050 due to its resource consumption requirement just to achieve consensus. Therefore, blockchains need to be solve consensus to actually achieve that we all know they can.

Yes, I'm crazy enough to don't pay attention to anything other than game theory and pure mathematics and to this point...

Game theory and pure mathematics - just the way I like it

So, my advice, just forget about consensus protocols, they either do not exist or they are bad protocols.

Here I'd differ from you a bit - I do feel that a good consensus protocol (that works and doesn't consume resources) is needed for blockchain success - not just as price of token but their actual impact on the world.

Thanks for your comments,
Shunsai

[shunsaitakahashi]

Hello fresheneesz,

How would other nodes detect that they're intentionally not approving blocks in order to blacklist them? Isn't anyone allowed to not approve blocks if they don't want to?

Node A is allowed to not approve another node's (say B's) blocks. That also means that B is allowed to not approve A's blocks. It works both ways. To detect if it is intentional or just network, point-to-point ping and bandwidth can be measured to guess number of blocks that should have reached within the time-slot.

But isn't only including your block as the only approved block the best way of gaining an advantage? This is known as the "bullet-voting" problem in approval voting, ranked voting, and range voting circles. https://en.wikipedia.org/wiki/Bullet_voting . While in normal voting, this shouldn't usually be much of a problem, it seems like in Proof-of-Approval there is a big incentive to bullet vote.

Yes, block approval should be incentivized. I am working on it.

Thanks,
Shunsai

[aliashraf]

Node A is allowed to not approve another node's (say B's) blocks. That also means that B is allowed to not approve A's blocks. It works both ways. To detect if it is intentional or just network, point-to-point ping and bandwidth can be measured to guess number of blocks that should have reached within the time-slot.

Very weak strategy to detect a node's intentional avoidance of approving peer's blocks. Nodes can simply stop supporting ICMP or manipulate it to delay response packets a few milliseconds.

Yes, block approval should be incentivized. I am working on it.

Block generation has zero cost and starting a node is very cheap, a typical player can run thousands of nodes that their trivially generated blocks will be approved by a masternode which in turn will be rewarded the bonus for its fake cooperative gesture.

I don't see this 'approval rewarding' strategy a good one and I have not an alternative in hand to suggest, in this moment.

[shunsaitakahashi]

Hello Fresheneesz,

The image of how blocks are set up (https://cdn-images-1.medium.com/max/1000/1*fPsB4-t6J_1u1UpF37dYHw.png) reminds me strongly of the way blocks are chained in the "two-hop blockchain" (https://eprint.iacr.org/2016/716.pdf). I would like to see more in-depth comparisons to other consensus protocols, especially DPoS which seems particularly close in mechanism to Proof-of-Approval.

Comparing Proof-of-Approval with 2-hop Blockchain (2-hop)
1. 2-hop uses PoW consuming resources (in addition to PoS), Proof-of-Approval has no PoW.
2. 2-hop's alternate blocks are different, one is from PoW and the next is from PoS. All Proof-of-Approval blocks are the same (but each block has 2 parts to it).
3. In 2-hop, during PoS round, the network choose one stakeholder for PoS block. In Proof-of-Approval any (or all) stakeholders approve blocks.
4. In 2-hop, fork selection is based on the longest PoW chain. In Proof-of-Approval fork selection is based on a better approval score.
5. Proof-of-Approval has 1 block finality. 2-hop's finality properties seem to be similar to PoW chains. The analysis here is more difficult since neither PoS nor PoW can be analyzed independent of each other. The PoS part of the analysis matches Cardano's Ouroboros (https://eprint.iacr.org/2016/889.pdf) static case. Since the static case isn't a reality (stakes will change during chain's lifetime), it is unclear if the PoS does add any additional security to the chain.

Hope this comparison helps.
Regards,
Shunsai

[monsterer2]

Hello Monsterer2,

In the following example, say I had a majority of stake at block A. Then at B0, C0 all the way up to head block H0, I sell off my majority of stake, understanding that there is an upper limit on stake transfer.

Code:

A - B0 - C0 - D0 - H0
  \ B1 - C1 - D1 - H1

Then, I create a double spend of my stake using the historical private keys, which I still own, thus sending it to myself in a fork continuing fork B1 - C1 - D1 that I create and approve myself (majority stakeholder).

This is not a 50% attack, because I don't own the value held in the private keys anymore, the cost for doing this is nothing.

Why doesn't the network just accept this fork as canon, given that its the same length (or longer if I chose) than the genuine fork?

Any single person owning (or colluding with) a quorum (majority) stake is a problem because that person (or colluders) can rewrite the chain as they choose. This situation is essentially the same as an adversary owning a quorum stake which is the limit of the protocol.

Hello Shunsai,

This is not how I would categorise the presented scenario. This is the NaS problem in it's essence. The attacker, as a historical majority stakeholder, can still take over the blockchain at any point in the future after he has sold his stake. So, at the current block height, said attacker owns 0 stake, yet he can still rewrite the entire chain due to this NaS problem.

Cheers, Paul.

[Bonzenboss]

Hey shunsia,
I read your technical paper and from what I understood it seems to be quite a good idea but of course as all protocols it first needs to be tested in action to see how good it actually works.

Could you maybe explain what permission, adversary and adversary tolerance means in the comparison chart? (I’m not a native English speaker as you can tell and translator Programm give me weird results so..)

Thanks in advance

-Bonzenboss

[shunsaitakahashi]

Hello Paul,

In the following example, say I had a majority of stake at block A. Then at B0, C0 all the way up to head block H0, I sell off my majority of stake, understanding that there is an upper limit on stake transfer.

Code:

A - B0 - C0 - D0 - H0
  \ B1 - C1 - D1 - H1

Then, I create a double spend of my stake using the historical private keys, which I still own, thus sending it to myself in a fork continuing fork B1 - C1 - D1 that I create and approve myself (majority stakeholder).

This is not a 50% attack, because I don't own the value held in the private keys anymore, the cost for doing this is nothing.

Why doesn't the network just accept this fork as canon, given that its the same length (or longer if I chose) than the genuine fork?

Any single person owning (or colluding with) a quorum (majority) stake is a problem because that person (or colluders) can rewrite the chain as they choose. This situation is essentially the same as an adversary owning a quorum stake which is the limit of the protocol.

This is not how I would categorise the presented scenario. This is the NaS problem in it's essence. The attacker, as a historical majority stakeholder, can still take over the blockchain at any point in the future after he has sold his stake. So, at the current block height, said attacker owns 0 stake, yet he can still rewrite the entire chain due to this NaS problem.

You are correct in pointing out an issue here which must be solved. I'd describe the issue slightly differently even without any single party holding a quorum stake.

An adversary buys empty private keys (with zero current stake) that once upon a time did hold a quorum stake in the blockchain. That adversary can start rewriting blocks from the time the keys did hold the quorum stake to the present - completely rewriting all the transactions.

I do need to think about this problem. Thanks for pointing it out.

Regards,
Shunsai

[shunsaitakahashi]

Hello Bonzenboss,

I read your technical paper and from what I understood it seems to be quite a good idea but of course as all protocols it first needs to be tested in action to see how good it actually works.

You are 100% correct that the testing with implementation is required. It turns out that even that isn't sufficient since an adversary will keep his/her attack vectors hidden till the blockchain actually succeeds when the attacks can earn him/her maximum profit. I think a peer review (what's being conducted here) can bring out many issues that a testing without serious adversarial attacks may not find.

Could you maybe explain what permission, adversary and adversary tolerance means in the comparison chart?

Permission - A party (node) needs some kind of authorization before it can join the protocol execution. In permissionless environment (e.g. Bitcoin), any party can download code and start a computer to join the protocol execution.

Adversary - A party or a group of parties that are trying to take unfair advantage of the blockchain e.g. spending the same coins multiple times. Such adversarial actions are called "attacks" and blockchains should be able to deter them.

Adversary tolerance -  Power of the adversary's attacks increases as he/she accumulates more and more of the resource critical to the blockchain. For Bitcoin, this resource is mining power. For Proof-of-Approval, this resource is the stake in the network. Adversary tolerance is the maximum amount of adversarial power a blockchain can deter.

Regards,
Shunsai

[kohi]

No such thing as "better blockchain consensus protocol" until your so called better blockchain consensus protocol has been through hell and back, namely, attacked by all angles on the technical aspect and attacked by all angles in the social engineering aspect (fork attempts by governments, social media attacks such as the twitter handle taken over by a fork, exchanges against you...) this is what Bitcoin has survived.

So far there is no PoW replacement. PoS has been proven to fail with NXT in the past, it wasn't pretty. We've seen IOTA fail big time as well, so DAG doesn't cut it. Hybrid attempts such as Decred.. no one hasn't even bothered to attack it for real yet.

I don't see realistic reasons for anyone to sell their BTC to get anywhere else.

PoW was also attacked and actually hacked in the early days of Bitcoin.

There's no reason these new consensus algorithms can't become as secure as PoW over time.

So far, the PoS and DPOS algorithm have been operating just fine for years now with no recent hacks. I don't doubt all the blockchains using these consensus algorithms are being attacked at all angles.

[monsterer2]

You are correct in pointing out an issue here which must be solved. I'd describe the issue slightly differently even without any single party holding a quorum stake.

An adversary buys empty private keys (with zero current stake) that once upon a time did hold a quorum stake in the blockchain. That adversary can start rewriting blocks from the time the keys did hold the quorum stake to the present - completely rewriting all the transactions.

I do need to think about this problem. Thanks for pointing it out.

Regards,
Shunsai

Hi Shunsai,

The standard response that most PoS developers have given in the past is that only bootstrapping nodes would be vulnerable to such an attack because all online nodes would reject the fork because they have knowledge and timestamps for the canon chain.

However, this concession is a loss of objectivity compared to PoW, and a loss of security because of that. I see no real way to avoid Nothing at Stake in a system with no PoW - you could try something like punishment of locked up funds (ala Etherum), but that's a real mess to analyse and may be less efficient overall than PoW as well as being inelegant.

Cheers, Paul.

[Slava79]

Hi Shunsai,

Very interesting article.

Did you read Tendermint white paper ? I scanned article and the core of it looks quite similar to Tendermint's PBFT consensus. Could you please provide a comparison ?

 

[aliashraf]

Hello Paul,

In the following example, say I had a majority of stake at block A. Then at B0, C0 all the way up to head block H0, I sell off my majority of stake, understanding that there is an upper limit on stake transfer.

Code:

A - B0 - C0 - D0 - H0
  \ B1 - C1 - D1 - H1

Then, I create a double spend of my stake using the historical private keys, which I still own, thus sending it to myself in a fork continuing fork B1 - C1 - D1 that I create and approve myself (majority stakeholder).

This is not a 50% attack, because I don't own the value held in the private keys anymore, the cost for doing this is nothing.

Why doesn't the network just accept this fork as canon, given that its the same length (or longer if I chose) than the genuine fork?

Any single person owning (or colluding with) a quorum (majority) stake is a problem because that person (or colluders) can rewrite the chain as they choose. This situation is essentially the same as an adversary owning a quorum stake which is the limit of the protocol.

This is not how I would categorise the presented scenario. This is the NaS problem in it's essence. The attacker, as a historical majority stakeholder, can still take over the blockchain at any point in the future after he has sold his stake. So, at the current block height, said attacker owns 0 stake, yet he can still rewrite the entire chain due to this NaS problem.

You are correct in pointing out an issue here which must be solved. I'd describe the issue slightly differently even without any single party holding a quorum stake.

An adversary buys empty private keys (with zero current stake) that once upon a time did hold a quorum stake in the blockchain. That adversary can start rewriting blocks from the time the keys did hold the quorum stake to the present - completely rewriting all the transactions.

I do need to think about this problem. Thanks for pointing it out.

Regards,
Shunsai

Shunsai,

I appreciate your hard work and feel deep sympathy with you, a person who knows something is wrong and current situation with crypto is not good enough because of the classical paradox between security and decentralization on one side and scalability on the other side.

The thing is POS, even your 2nd phase commit version, is not proved to be an answer because of problems like Nothing at Stake mentioned by Paul here and a lot more. It is not just a simple 'issue' to 'think about' it has been discussed over and over and no answer available yet, other than complicating everything to make it harder for both adversaries and innocent participants and/or leading to a sophisticated protocol vulnerable to implementation bugs and a series of other limitations. Just ugly, far from elegant solutions too complicated, too fragile.

I have gone through this before and have chosen another approach: forgetting about the answers and re-thinking questions.

For instance, let's take a look at your 'permissionless' index (whether participating in a protocol needs permission from an authority or not): What if one needs permission but not from a centralized authority but a decentralized entity like a curated list which is maintained in a classic PoW base blockchain tp participate in another protocol that carries the much burden of the transaction load, being secured by a trivial, low cost algorithm ?

See? A lot of possibilities out there and nothing is 'obvious' and, as I'm used to say, you shouldn't stick with common sense.

[Slava79]

In the following example, say I had a majority of stake at block A. Then at B0, C0 all the way up to head block H0, I sell off my majority of stake, understanding that there is an upper limit on stake transfer.

Code:

A - B0 - C0 - D0 - H0
  \ B1 - C1 - D1 - H1

Then, I create a double spend of my stake using the historical private keys, which I still own, thus sending it to myself in a fork continuing fork B1 - C1 - D1 that I create and approve myself (majority stakeholder).

Indeed, interesting, very valid point. I suppose you mean the security rule named "validator's bond" or "security deposit", so that after "depositing" (required to become a validator) some amount of tokens, it is not possible to "withdraw" immediately.  It is not mentioned in the document, but should be easy to add in order to address the issue you mentioned.

[shunsaitakahashi]

Hello Slava79,

Did you read Tendermint white paper ? I scanned article and the core of it looks quite similar to Tendermint's PBFT consensus. Could you please provide a comparison ?

Comparing Proof-of-Approval with Tendermint

1. Tendermint is a PBFT (http://pmg.csail.mit.edu/papers/osdi99.pdf) like algorithm where all nodes together agree on transactions and their order that would be put in a block. Therefore, all nodes together come up with a single block to be inserted in the blockchain. In Proof-of-Approval, just like Bitcoin, individual nodes bundle transactions on their own and the network chooses one of those bundles.

2. The "validators" in Tendermint volunteer to put some of their money as collateral that will earn them rewards but could also be confiscated. Validators are similar to those in Casper the Finality Gadget of Ethereum (https://arxiv.org/pdf/1710.09437.pdf). This article compares the two https://blog.cosmos.network/consensus-compare-casper-vs-tendermint-6df154ad56ae.

3. Tendermint a single "proposer" who proposes the blocks. The proposer is selected by some process from the validators in round robin.

4. Tendermint transactions achieve instant finality while Proof-of-Approval takes 1 block to finality.

5. Tendermint is likely to have less "liveness" than a block creating protocol like Bitcoin or Proof-of-Approval.

6. The amount need to takeover Tendermint is 1/3 of the collateral amount while the amount to takeover Proof-of-Approval blockchain would be almost half (just under half) of the value of the blockchain.

Regards,
Shunsai

[DevilOper]

Thank you sir for the enlightening, but NO! I don't get it, you know why? Because it costs too much and hell it is not scalable and I can't imagine a crypto dominated globe with millions of machines all agreed upon every single transaction.

Well, I would say more: any or almost any idea of decentralisation based on the Proof-of-Anything is vulnerable at the level of it's concept.
Because it strongly tends to concentration (i.e. centralisation) of abovementioned Anything, and thus all the consensus might be overriden; in fact, all the history might be overwritten by one who takes a control over majority of that Anything.

[shunsaitakahashi]

Hello Aliashraf,

I believe I have a good solution to this History attack (aka Costless Simulation attack) issue. While most PoS systems are relying on the so called "Weak Subjectivity" solution, this solution is purely objective. I am writing it up now and will post it here soon for everyone's review. I can't overstate the value of the feedback provided in this forum. Thanks everyone!

Regards,
Shunsai

Hello Paul,

In the following example, say I had a majority of stake at block A. Then at B0, C0 all the way up to head block H0, I sell off my majority of stake, understanding that there is an upper limit on stake transfer.

Code:

A - B0 - C0 - D0 - H0
  \ B1 - C1 - D1 - H1

Then, I create a double spend of my stake using the historical private keys, which I still own, thus sending it to myself in a fork continuing fork B1 - C1 - D1 that I create and approve myself (majority stakeholder).

This is not a 50% attack, because I don't own the value held in the private keys anymore, the cost for doing this is nothing.

Why doesn't the network just accept this fork as canon, given that its the same length (or longer if I chose) than the genuine fork?

Any single person owning (or colluding with) a quorum (majority) stake is a problem because that person (or colluders) can rewrite the chain as they choose. This situation is essentially the same as an adversary owning a quorum stake which is the limit of the protocol.

This is not how I would categorise the presented scenario. This is the NaS problem in it's essence. The attacker, as a historical majority stakeholder, can still take over the blockchain at any point in the future after he has sold his stake. So, at the current block height, said attacker owns 0 stake, yet he can still rewrite the entire chain due to this NaS problem.

You are correct in pointing out an issue here which must be solved. I'd describe the issue slightly differently even without any single party holding a quorum stake.

An adversary buys empty private keys (with zero current stake) that once upon a time did hold a quorum stake in the blockchain. That adversary can start rewriting blocks from the time the keys did hold the quorum stake to the present - completely rewriting all the transactions.

I do need to think about this problem. Thanks for pointing it out.

Regards,
Shunsai

Shunsai,

I appreciate your hard work and feel deep sympathy with you, a person who knows something is wrong and current situation with crypto is not good enough because of the classical paradox between security and decentralization on one side and scalability on the other side.

The thing is POS, even your 2nd phase commit version, is not proved to be an answer because of problems like Nothing at Stake mentioned by Paul here and a lot more. It is not just a simple 'issue' to 'think about' it has been discussed over and over and no answer available yet, other than complicating everything to make it harder for both adversaries and innocent participants and/or leading to a sophisticated protocol vulnerable to implementation bugs and a series of other limitations. Just ugly, far from elegant solutions too complicated, too fragile.

I have gone through this before and have chosen another approach: forgetting about the answers and re-thinking questions.

For instance, let's take a look at your 'permissionless' index (whether participating in a protocol needs permission from an authority or not): What if one needs permission but not from a centralized authority but a decentralized entity like a curated list which is maintained in a classic PoW base blockchain tp participate in another protocol that carries the much burden of the transaction load, being secured by a trivial, low cost algorithm ?

See? A lot of possibilities out there and nothing is 'obvious' and, as I'm used to say, you shouldn't stick with common sense.

[monsterer2]

Hello Aliashraf,

I believe I have a good solution to this History attack (aka Costless Simulation attack) issue. While most PoS systems are relying on the so called "Weak Subjectivity" solution, this solution is purely objective. I am writing it up now and will post it here soon for everyone's review. I can't overstate the value of the feedback provided in this forum. Thanks everyone!

Regards,
Shunsai

I believe this to be impossible, so I await your update with the utmost enthusiasm!

[shunsaitakahashi]

Created new thread for the updated paper since it may fix some previous problems but may have created some new ones.
Thread https://bitcointalk.org/index.php?topic=3913439.0

Regards,
Shunsai

[d5000]

Interesting proposal with some nice properties, although I mostly agree with monsterer that the long-range attack problem isn't fully solved. I'll look into the updated paper this week to comment on it.

Only a little observation (and that's why I comment here in the old thread): The attack monsterer described (old majority stakeholder(s) taking over the chain with a double spend) is essentially the same attack than a profitable 50+% PoS attack. In a profitable 50+% attack, you will wait for a few blocks selling your stake first (most likely in an altcoin exchange as they confirm the "reception" of the coins relatively fast) and only then try to "take over" the new chain with a prepared attack chain. Otherwise, very likely your stake would decrease in value. So, also in this attack, in the moment you take over, you don't own stake anymore (or at least, you aren't forced to own any stake).

The difference between "long-range attack" and "profitable 50+% attack" be the time between the actual double spend and the "takeover" (when the attack chain replaces the "honest" chain) but the attack mechanism is the same. But monsterer is absolutely right and this is the reason why PoS protocols are to be considered - until this is solved - weaker than PoW protocols. I consider this attack unpractical in a chain with many stakeholders and reorg limits or "rolling checkpoints", and it's likely to be very expensive, but it may be considerably cheaper than a PoW 50+% attack.