Hi Sergej,

Thanks for the insightful posting, especially highlighting the FX risk which was far from being evident on my side!

I don't know in details the security architecture of Bitrefill zeroconf acceptance system, though from what I suppose there is at least a set of full-nodes well-connected across the p2p network, on top of which some mempools reconciliation is exercised
and zeroconf candidate sanitize against. While I believe this is a far-more robust deployment against double-spend attempts, there is still the ability for a sophisticated attacker to "taint" miner mempools, and from then partition judiciously the transaction-relay network to game such distributed mempool monitoring system. There is also the possibility of an attacker using some "divide-and-conquer" transaction broadcast algorithm to map Bitrefill monitoring point, though as far as I'm aware such algorithm has not been discussed. I agree with all of that, easier said than done.

(Which let me think that such distributed mempool monitoring system should be provide some enhanced security even in a full-rbf world, that they would require far more resources than the average node from the p2p network as a whole might be a counter-argument for their social acceptance, however I'm also thinking that a robust Lightning infrastructure of the future might require multiple mempool/transaction-relay endpoints, at least to reduce  cross-layer mapping links, though conversation for another day...).

About the FX risk itself, this is far from being isolated from 0conf, as Lightning payments themselves might still have a time lapse between the issuance of invoices and the settlement of the HTLC at the payee endpoint. In fact this volatility concern is endured by anyone using Bitcoin regularly in interface with the fiats worlds, i.e everyone excepted the long-term store of wealth crowd. From a merchant perspective, effectively, the options to cover themselves against this risk are simple. One could take positions directly in traditional financial derivatives, like doing participants in international trades, though it would require an educated manpower on the merchant side. Or leveraging some stablecoins derivatives system, coming with its own technical complexity and social trust hazards. Another direction would be to clearly define the responsibility between merchants or users, on whom is the FX risk. If it's on users, they should be the one RBFing/CPFPing to increase the merchant address output, beyond the fact "dynamic pricing" would be a weird UX, it would require liveliness from the wallets until block confirmation (introducing here many requirements of a LN wallet). If it's on the merchants, they could be the ones CPFPing thanks to package relay, though it would come again with some engineering complexity and overhead blockspace cost (and the first version of package relay likely won't enable CPFP batching for concerns of potential bandwidth/CPU DoS).

On the efficacy of RBF, I understand the current approach of assuming "manual" RBFing by power users ill UX thinking. I hope in the future to have automatic fee-bumping implemented by user wallets, where a fee-bumping budget and a confirmation preference are pre-defined for all payments, and the fee-bumping logic "simply" enforcing the user policy, ideally based on historical mempool data. True fact: we don't have such logic in consumer wallets today. Or at least only rudimentary in the backend of LN implementations, and only for time-sensitive on-chain claims for now (or at least speaking for LDK). If we take the history of browsers as a comparison, while we might be out of the Lynx-style phase of wallets, we might still be more in the late Netscape kind of thing than something like Chrome today. In other words, there are many directions for improvements for users' wallets.

All that said, I learn to converge that as a community we would be better off to weigh deeper the risks/costs between 0confs applications and contracting protocols in light of full-rbf.

Best,
Antoine

Le mer. 19 oct. 2022 à 10:33, Sergej Kotliar via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> a écrit :
Hi all,

Chiming in on this thread as I feel like the real dangers of RBF as default policy aren't sufficiently elaborated here. It's not only about the zero-conf (I'll get to that) but there is an even bigger danger called the american call option, which risks endangering the entirety of BIP21 "Scan this QR code with your wallet to buy this product" model that I believe we've all come to appreciate. Specifically, in a scenario with high volatility and many transactions in the mempools (which is where RBF would come in handy), a user can make a low-fee transaction and then wait for hours, days or even longer, and see whether BTCUSD moves. If BTCUSD moves up, user can cancel his transaction and make a new - cheaper one. The biggest risk in accepting bitcoin payments is in fact not zeroconf risk (it's actually quite easily managed), it's FX risk as the merchant must commit to a certain BTCUSD rate ahead of time for a purchase. Over time some transactions lose money to FX and others earn money - that evens out in the end. But if there is an _easily accessible in the wallet_ feature to "cancel transaction" that means it will eventually get systematically abused. A risk of X% loss on many payments that's easy to systematically abuse is more scary than a rare risk of losing 100% of one occasional payment. It's already possible to execute this form of abuse with opt-in RBF, which may lead to us at some point refusing those payments (even with confirmation) or cumbersome UX to work around it, such as crediting the bitcoin to a custodial account.

To compare zeroconf risk with FX risk: I think we've had one incident in 8 years of operation where a user successfully fooled our server to accept a payment that in the end didn't confirm. To successfully fool (non-RBF) zeroconf one needs to have access to mining infrastructure and probability of success is the % of hash rate controlled. This is simply due to the fact that the network currently won't propagage the replacement transaction to the miner, which is what's being discussed here. American call option risk would however be available to 100% of all users, needs nothing beyond the wallet app, and has no cost to the user - only upside.

Bitrefill currently processes 1500-2000 onchain payments every day. For us, a world where bitcoin becomes de facto RBF by default, means that we would likely turn off the BIP21 model for onchain payments, instruct Bitcoin users to use Lightning or deposit onchain BTC to a custodial account that we have. 
This option is however not available for your typical BTCPayServer/CoinGate/Bitpay/IBEX/OpenNode et al. Would be great to hear from other merchants or payment providers how they see this new behavior and how they would counteract it.

Currently Lightning is somewhere around 15% of our total bitcoin payments. This is very much not nothing, and all of us here want Lightning to grow, but I think it warrants a serious discussion on whether we want Lightning adoption to go to 100% by means of disabling on-chain commerce. For me personally it would be an easier discussion to have when Lightning is at 80%+ of all bitcoin transactions. Currently far too many bitcoin users simply don't have access to Lightning, and of those that do and hold their own keys Muun is the biggest wallet per our data, not least due to their ease-of-use which is under threat per the OP. It's hard to assess how many users would switch to Lightning in such a scenario, the communication around it would be hard. My intuition says that the majority of the current 85% of bitcoin users that pay onchain would just not use bitcoin anymore, probably shift to an alt. The benefits of Lightning are many and obvious, we don't need to limit onchain to make Lightning more appealing. As an anecdote, we did experiment with defaulting to bech32 addresses some years back. The result was that simply users of the wallets that weren't able to pay to bech32 didn't complete the purchase, no support ticket or anything, just "it didn't work 🤷‍♂️" and user moved on. We rolled it back, and later implemented a wallet selector to allow modern wallets to pay to bech32 while other wallets can pay to P2SH. This type of thing  is clunky, and requires a certain level of scale to be able to do, we certainly wouldn't have had the manpower for that when we were starting out. This why I'm cautious about introducing more such clunkiness vectors as they are centralizing factors.

I'm well aware of the reason for this policy being suggested and the potential pinning attack vector for LN and other smart contracts, but I think these two risks/costs need to be weighed against eachother first and thoroughly discussed because the costs are non-trivial on both sides.

Sidenote: On the efficacy of RBF to "unstuck" stuck transactions
After interacting with users during high-fee periods I've come to not appreciate RBF as a solution to that issue. Most users (80% or so) simply don't have access to that functionality, because their wallet doesn't support it, or they use a custodial (exchange) wallet etc. Of those that have the feature - only the power users understand how RBF works, and explaining how to do RBF to a non-power-user is just too complex, for the same reason why it's complex for wallets to make sensible non-power-user UI around it. Current equilibrium is that mostly only power users have access to RBF and they know how to handle it, so things are somewhat working. But rolling this out to the broad market is something else and would likely cause more confusion. 
CPFP is somewhat more viable but also not perfect as it would require lots of edge case code to handle abuse vectors: What if users abuse a generous CPFP policy to unstuck past transactions or consolidate large wallets. Best is for CPFP to be done on the wallet side, not the merchant side, but there too are the same UX issues as with RBF. 
In the end a risk-based approach to decide on which payments are non-trivial to reverse is the easiest, taking account user experience and such. Remember that in the fiat world card payments have up to 5% chargebacks, whereas we in zero-conf bitcoin land we deal with "fewer than 1 in a million" accepted transactions successfully reversed. These days we have very few support issues related to bitcoin payments. The few that do come in are due to accidental RBF users venting frustration about waiting for their tx to confirm.
"In theory, theory and practice are the same. In practice, they are not"

All the best, 
Sergej Kotliar
CEO Bitrefill.com


--

Sergej Kotliar

CEO


Twitter: @ziggamon 


www.bitrefill.com

Twitter | Blog | Angellist



--

Sergej Kotliar

CEO


Twitter: @ziggamon 


www.bitrefill.com

Twitter | Blog | Angellist

_______________________________________________
bitcoin-dev mailing list
bitcoin-dev@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev