7 Theses on a next step for BIP-119
Martin Luther had 99, I'm going to give you 7Tweet
Warning: this post assumes a great deal of context on CTV is already understood by the reader. If you are not familiar, you may wish to start with utxos.org, the advent calendar, and the BIP-119 text and reference implementation, as those may provide much needed context about what CTV is and why a next step is being discussed. If you only have a little time, minimally I would advise these two from the advent calendar as the minimum required context: Contracting Primitives and Upgrades to Bitcoin and RoadMap or Load o’ Crap?.
This post starts with a conclusion:
Within a week from today, you’ll find software builds for a CTV Bitcoin Client for all platforms linked here:
- Mac OSX TODO:
- Windows TODO:
- Linux TODO:
These will be built using GUIX, which are reproducible for verification. The intended code to be built will be https://github.com/JeremyRubin/bitcoin/tree/checktemplateverify-v23.0rc5-paramsv0 which is based on Bitcoin Core v23.0 release candidate 5, with commit hash dd9a4e0ea8a109d1607ca1ec16119b1bc952d8b0. You can begin testing this immediately, and even producing your own GUIX builds as well.
Signatures for the builds will be available below:
- TODO: … .asc
The source tarball:
- TODO: … .tar.gz
The client has a Speedy Trial release similar to Taproots with parameters proposed to be:
- Signal Start MTP: 1651708800 (May 5th, 2022, 00:00 UTC)
- Signal Timeout MTP: 1660262400 (August 12th, 2022, 00:00 UTC)
- Activation Height: 762048 (Approximately Nov 9th)
See the appendix to verify these parameters.
This ensures 6 signalling periods to activate CTV. The Start and Timeout are targeting mid-period (if hashrate stays steady) times to ensure that it is unlikely we would have more or fewer periods.
The week delay between this post and builds is to provide time for review on the selection of parameters as well as ability to rebase onto a final v23.0 release, should it become ready within the week. Backports are in the works for v22.0, but release builds may not be made available as Bitcoin’s release build processes have changed since v22.0 to use GUIX. The branch for backports is available here: https://github.com/JeremyRubin/bitcoin/tree/checktemplateverify-v22.0 with current commit hash 4d2c39314834a28cd46da943a12300cca8ffcb10, if you would like to help with testing.
Why this, why now?
I’ve just returned from the Bitcoin Miami “Bacchanal”. Personally, I had a couple different goals for being there1. One of my primary focuses was on talking to as many people as possible about BIP-119 and the future road to take.
While consensus has to happen among a much broader set of people than can fit in a conference in Miami, the reality is that there were more than 20,000 Bitcoiners at this event and a good representation across industry, developers, journalists, podcasters, plebs, whales, pool operators, miners, venture capitalists, and more. To say it was a representative sample wouldn’t be fair, but it certainly was not a homogeneous crowd. And I spoke to as many people as I could.
There were a couple common threads across the feedback I received:
- Agree or disagree with CTV in particular, folks generally liked how/that I was driving a conversation forward, and respected the hustle required to do so.
- A lot of people felt that CTV would help them in a tangible way, and more than a few times, individuals approached me with a concrete use case they needed but had not fully written it up yet.
- A lot of people wanted to know what the next step was and what I was planning to do to get it activated and when.
Some people had some suggestions on what I should do as a next step:
- Some folks said I should just do a UASF and rally the users.
- Some said I needed to organize a summit for developers to explore covenants2.
- Some said I didn’t need to do a UASF, nor advocate for it, but I did need to decide on exact release parameters and distribute a reproducible binary so that it was clear what should be run and when so that end-to-end activation testing could proceed.
It’s (un?)remarkably difficult to integrate all feedback on a complex topic like a Bitcoin upgrade coherently. But, having thought it through, I decided that the approach above was the correct next step. Below, you’ll find some reasoning on why I believe this to be proper and not out-of-line with how soft-fork development should go.
However, if I’m wrong in your view, consider me a mere messenger and please don’t shoot the messenger. You just need to communicate clearly to the community why they should not run and signal for CTV and I’m confident that the wisdom of the consensus set will decide in it’s best interests.
So why ship a binary and release parameters?
1) CTV passes a basic pre-flight checklist.
This discussion has to start anchored in a “pre-flight checklist” for CTV. These are fundamental questions that we should be able to tick boxes for for any proposed upgrade… Sadly, the community at large doesn’t have a codified checklist3, but personally I tick off the following boxes:
- No material changes to the BIP/Spec/Ref Impl necessary for ~2 years (beyond rebases).
- A reasonably well reviewed and tested PR-implementation exists.
- ~5 Months of a 5.5 BTC Bounty for bugs in CTV
- I socialized a similar roadmap 5 months ago, which received a reasonably warm response so there are ‘few surprises’ here against previously communicated intent.
- A community of supporters: breakdown, 16 supporting orgs, 109 individuals, 3 mining pools (totalling about 15-18% depending on when you look).
- Only 3 individual NACKs + 1 org (owned by one of the individuals). You should read them yourself, but I think the NACKs are summarizable as “it’s too soon” and “there should be a different process” rather than “I have identified a flaw in the proposal”. See section 4 for more on this. The NACKs are linked below:
- Ample time to have produced a nack with a technical basis.
- 7 regular meetings over ~16 weeks to discuss the upgrade.
- There exists software from multiple different parties for using CTV to accomplish various tasks. None of these users have uncovered issue or difficulty with CTV.
- Many in the community are arguing for more functionality than what CTV offers, rather than that the functionality of CTV might be unsafe. CTV ends up being close to a subset of functionality offered by these upgrades.
- CTV does not impose a substantial validation burden and is designed carefully to not introduce any Denial of Service vectors.
- There exists a Signet with CTV active where people have been able to experiment.
- Backports for current release and prior release (will be) available.
2) Unforced errors
In tennis, an unforced error is a type of lost point where a player loses because of their own mistake. For example, suppose your opponent shoots a shot that’s a lob high in the air and slow. But it looks like it’s going out, so you do a little victory dance only to uncover that… the ball lands in. You had enough time to get to the ball, but you chose not to because you didn’t think it would go in. Contrast this to a forced error – your opponent hits a shot so hard and fast across the court no human could reach it let alone return it4.
What’s this got to do with Bitcoin?
Community consensus is the ball, and we don’t know if by August it will be in or out.
Getting to where the ball might land is preparing a software artifact that can activate.
If an artifact isn’t prepared that does this, even if community consensus is ready by then, it’s an unforced error that it wasn’t ready which precludes us from being live in August.
When should you avoid an unforced error like this? When the cost of getting to the ball is sufficiently small.
I’m already maintaining a backportable to Bitcoin Core 23 and 22 patchset for CTV. It’s not much work to set parameters and do a release.
Which brings us to…
3) Product Management is not “my Job” – it’s yours.
Devs don’t swing the racquet, we get to the ball. It’s the community’s job to decide to swing or not. One might rebutt this point – the community isn’t well informed to make that call, but fortunately devs and other tuned in individuals can serve as “coaches” to the public and advise on if the swing should happen.
Producing the code, the tools, the reviews, the builds, the dates, these are all getting to the ball activities.
It’s possible that as a developer, one could say that we should not get to the ball unless we know the community wants to swing.
But that’s false. What if the community wants to take a swing at it but developers haven’t gotten to the ball? What if developers refuse to get to the ball because they don’t want the community to take that shot? Well, tennis can be a game of doubles (I’m really sticking with this metaphor), and maybe your teammate – the community itself – strives to go for it and sprints cross court to make up and take a shot. Maybe that shot looks like a UASF, maybe it looks like a hard fork, maybe it’s lower quality since there was less time to make the right shot placement (worse code if usual devs don’t review). But ultimately, the economic majority is what backstops this, the devs just have the opportunity to help it along, perhaps, a little smoother.
Largely, the formal critiques of CTV (the 3 NACKs) are based on topics of whether or not to swing the racquet, not if we should be at the ball. There are other critiques as well, about the generality of the upgrade, but we’ll discuss those later in this post.
I’ll excerpt some quotes from the NACKs below:
I also think attempting regular soft forks with single features is a disturbing pattern to get into having just activated Taproot. As I say in the linked post it means less and less community scrutiny of consensus changes, it becomes a full time job to monitor the regular soft forks being activated (or attempting to be activated) and only a tiny minority of the community can dedicate the time to do that.
So this seems to be a point largely about product management – we should only take a shot when we can line up more than once, due to the cost of swinging the racquet. Not really my job, it’s the communities.
Hence I’d like to register a “No” soft signal as I fundamentally disagree that a CTV soft fork should be attempted in the near future and my concerns over premature activation outweigh my enthusiasm for digging into the speculated real world use cases of CTV.
If you disagree that it should be attempted, that’s fine. Your time to voice your concerns is in making the swing.
Generally, I do not think Bitcoin is ready for any new soft forked features at all in the short term. Taproot just arrived and there is already so much work to be done to adopt and utilize it.
This is a product management point. “Work on feature X should block all work on other features”. It’s not a point on if CTV is a feature of independent merit.
Further, it’s a layer violation. Wallet progress is wholly independent of consensus upgrades, and generally, we don’t operate on a waterfall development model.
Any improvements CTV-eltoo may bring to LN are not significant enough to claim they are urgent for adoption or R&D for LN progress and adoption.
Again, a product management point. How do we measure what is important to the Lightning Community? Oh, on utxos.org, there are multiple Lightning Network companies and individuals (Lightning Labs, Muun Wallet, Roasbeef, ZmnSCPxj, LN Markets, Breez, fiatjaf, and more). So if that represents the community, then it seems like a greenlight.
Since I am not qualified, nor are 99% of Bitcoiners, to evaluate this deeply, I feel much more time and scrutiny is required to accept CTV and the related projects Jeremy has prepared with it.
If you’re not qualified, remind me why we’re listening?
Sorry, I couldn’t help the snark. Graciously, let’s accept the framing for a second – who are the stakeholders who need to sign off? What is this process like, concretely?
Is this a process that happens before or after we ‘get to the ball’?
It can definitely be after we get to the ball, and the decision to swing or not is a bit too product-management-y for how a dev should engage.
Also, related projects are a bit like the mix-ins at Cold Stone Creamery™. You are free, of course, to just get ice cream! That there are a myriad of uses doesn’t mean you need to accept all of them, it’s sufficient to just consider the one or two you care about.
I am currently happy with what we have in Bitcoin, and would prefer Core development prioritize optimizations and cleanup work over more and more features that have no urgent need or chance of adoption anytime soon.
This belies a basic misunderstanding of FLOSS:
- People work on what they want to
- CTV is already ‘finished’
A non-dev’s preference to spend time on cleanup or optimization doesn’t make any dev write that code or shift focus. Most core devs don’t have a boss, and if they do, it’s probably not you! It’s structurally impossible to direct the attention of developers.
And it so happens that I am not happy with what we have in Bitcoin, so I did something about it. With respect to adoption, people will likely be using CTV pretty soon after it’s available, since it is a big step up for a number of critical uses like vaults. These applications are already being built. They can be used on signet which can be deployed to mainnet immediately5. The implementation details for basic custody contracts are pretty simple and don’t require the level of coordination for support that other contracts like lightning or DLCs, so adoption can be at the individual level.
The path around prioritization remains a product management question, and not something devs can be compelled to follow.
4) There are other things to work on.
I can get to the ball for this shot, but then I’d like to work on getting in position for the next shot on time.
There are other important technologies to work on, keeping covenants in limbo ties up a lot of human capital in trying to solve for getting something, vs. having something and being able to work on building solutions using it plus designing new technologies that make Bitcoin even better in different or complimentary ways.
What’s the right amount of rumination (chewing) to swallowing? Eventually, the mouthful you have is stopping you from taking the next bite.
5) Consensus is memoryless
A memoryless process is something that “never makes progress”. For example, consider a board game, where you need to roll a 6 to win. You expect to need 6 rolls to win. You roll a 5. How many more rolls do you need? It’s not 5. It’s 6 – the process is memoryless.
Clearly consensus isn’t entirely memoryless. Something that is a concept only obviously has to be turned into a hard artifact.
CTV has been a ‘hard artifact’ for 2 years. 2 years ago I took a poll of 40 or so developers who attended my utxos.org workshop in Feb 2020. An average sentiment was that maybe we try to do CTV in a year or so, and that we could definitely do it maybe in 2 years.
I hear the same today from people who advocate a slower process. Maybe a year from now, we could definitely do it in maybe 2 years.
In 2 years, if we wait, won’t we hear the same?
Here’s a few reason why we might hear the same complaints in the future:
In 2 years, suppose Bitcoin is ~2x the price.
Shouldn’t we acknowledge twice as much at risk and do twice as much work to security audit things going into it? What if it’s not just El Salvador with Bitcoin as a national currency, but now Guatamala too?
Suppose we want to get to a point where 50% of the community has reviewed a change. In 2 years, what if the community is 2x the size? Then even if we hit 50% of today’s community, we only have 25% of the community read up.
Hopefully we keep innovating. And if we do keep innovating, we’ll come up with new things. If we come up with a new thing that’s 2x as good as CTV in 2 years, but it takes another 2 years to implement it concretely, should we wait till we finish that? But what happens when we come up with something 2x better than that? Wait another 2 years? Now we’re 4 years out from when the swing to get CTV over the net was doable, and we’d have 0 in hand solutions for the problems CTV tries to solve.
All this points to the nature of the memorylessness of trying to get consensus in an open network.
The best I can do is to get to the ball and let the community decide to take the shot.
Concretely – what is the cost to having CTV in bitcoin during this time while the “better” alternative is being worked on, if we do decide to activate knowing we might one day obsolete CTV? What are the benefits to having 3-5 years of basic covenants in the meantime? On the balance it seems, to me, net positive. It also seems to be a decision that the community at large can judge if the costs are worth the benefits.
6) You can fight against it.
A criticism of the soft fork process is it’s not safe with Speedy Trial (ST) and ST is bad so we shouldn’t do it. This is not a strong criticism: with Taproot, our most important upgrade in years, went smoothly even though there was sharp disagreement over the safety of the mechanism at the time.
Here’s a breakdown of why Speedy Trial is OK from the perspective of different participants:
You want CTV and won’t take no for an answer.
Start with a ST. After 3 months, it either goes or it doesn’t. At least we were at the ball. Now, let’s do a UASF with a LOT=true setting. Because ST is a fail fast, it’s in theory using time where otherwise we’d have to spend coordinating for the harsher realities of a LOT=true effort, so it’s a happy-path optimization.
If you’re a miner, you should signal during the period.
You’d like CTV and might take no.
ST is for you. Can always follow up with another ST if the first fails, or another option if your opinion changes.
If you’re a miner, you should signal during the period.
You do not want CTV, but if others do want it, whatever.
ST is for you – it only goes through if others signal for it.
If you’re a miner, you can signal no during the period, and you can write a blogpost on why you don’t.
You do not want CTV, and will not take yes for an answer.
If you’re a miner, you signal no during this period. You may want to optimize the forkd code for never building on the chain you don’t like.
7) Bitcoin Core is not Bitcoin
Bitcoin Core is a ‘reference client’ for the Bitcoin network, but it is not the job of any of the maintainers of Bitcoin Core to decide what happens with respect to consensus upgrades.
When I’ve previously asked maintainers for clarity on what ‘merge rubric’ they might apply to the CTV pull request, I’ve been effectively stonewalled with no criterion (even for things that were historically merged) and claims that soft-fork discussion is outside the purview of maintainership. To be clear, I’m not asking maintainers to merge, merely when they do make the decision to, what they are evaluating. The reticence to make clear guidelines around this was surprising to me, at first.
But then I understood: it isn’t the maintainer’s fault that they cannot give me any guidance, it’s how it must be.
The idea that Bitcoin Core might serve as the deciding body for consensus upgrades puts developers of Bitcoin Core into a dangerous position in the future, whereby various agents might wrongfully attempt to compel Core developers (e.g., using the legal system) to release a soft-fork client for whatever nefarious goal. Making it clear that soft-forks are released by independent efforts and adopted by the community at large is the only process we can take that keeps Bitcoin Core apolitical and unexposed.
We’ve seen in other communities what it looks like when lead devs exert too much influence over the protocol and reference clients directly. Not good. We do not want to have a similar precedent for Bitcoin.
While previous soft-forks have generally been released by Core, I have no qualms with leading by example for how future soft-fork development should be. And if Core wants to merge CTV and do a release with compatible parameters, they are welcome to, without such a release being driven by the project maintainers directly, but rather to maintain compatibility with the will of the community.
Thus, Alea Iacta Est.
Bonus: What do I do now?
I believe the community’s next steps are:
- Evaluate the software proposed above and find any bugs (claim 5.5 BTC Bounties?)
- Discuss vociferously through the next few months if BIP-119 should be activated or not (that means you should e.g. post publicly if you/your org endorses this particular path, cover it in your news org, etc).
- Before the end of July, Miners should signal if the speedy trial should succeed
- Before November, if Speedy Trial passes, then all users should ensure they upgrade to validate CTV
- If Speedy Trial fails, at least we were at the ball, and we can either try again next year, meaning CTV would be availble for use in at minimum 1.5 years, or we can re-evaluate the design of CTV against alternatives that would take more time to prepare engineering wise (e.g., more general covenants, small tweaks to CTV).
What is Jeremy Rubin going to do?
Well, at this point I am unemotional about any outcome. Judica, my startup, is focused on Bitcoin infrastructure that can be used to great impact with or without CTV, so I’ve explicitly positioned myself to be personally indifferent to outcome. I personally think that CTV is ready to go, and will deliver immense benefits to the network, so I’ll advocate for signalling for it.
However, in my advocacy, I’ll be careful to note that it’s not a must. All actors must decide if it’s in their own rational self-interest to have the soft-fork proceed.
But what about UASF?
Regrettably, no one has produced the ST compatible UASF code since last year, for various reasons. I understand the motives and tradeoffs of a UASF out the gate, but I still personally believe a UASF is best done as a follow-up to a ST, as I detailed in my mailing list post on the subject here.
Appendix: Parameter Check!
$ gdate -d@1651708800 -u Thu May 5 00:00:00 UTC 2022 $ gdate -d@1660262400 -u Fri Aug 12 00:00:00 UTC 2022
The below script simulates the passage of time and confirms that we are beginning at an expected mid-period time, and we are also ending near a mid-period time, given the assumed number of SIGNAL_PERIODS. This technique should guarantee with high certainty at least SIGNAL_PERIODS - 1, and repeating the simulation with up-to-date numbers as the signalling window progresses will produce more accurate forecasts.
import datetime SIGNAL_PERIODS = 7 current_time = 1650301349 height_now = 732450 minutes_till_may_5th = 23457 height = int(height_now + minutes_till_may_5th/10.0) print("Expected Height", height) start_height = height + (2016 - (height % 2016)) print("Expected Start Height", start_height) print("Start is mid period: ", (start_height-height)/2016.0) minutes_from_now = 10*(start_height - height_now) print("In This many Minutes", minutes_from_now) print("In This many days", minutes_from_now/60.0/24.0) stop_height = start_height + 2016*SIGNAL_PERIODS print("Stopping at height", stop_height) total_blocks = stop_height - height_now end_time = (total_blocks - 2016/2)*10*60 + current_time print("End of signalling at expected time", datetime.datetime.fromtimestamp(end_time)) active_height = 762048 secs_till_active = (active_height - height_now)*10*60 print("Active at", datetime.datetime.fromtimestamp(current_time + secs_till_active))
Expected Height 734795 Expected Start Height 735840 Start is mid period: 0.5183531746031746 In This many Minutes 33900 In This many days 23.541666666666668 Stopping at height 749952 End of signalling at expected time 2022-08-10 23:02:29 Active at 2022-11-09 23:02:29
including roller skating along the beach of course… ↩
in theory, pleb.fi was intended to be part of this… ↩
I think defining such a process more formally would be great, but it’d be controversial. ↩
technically, if you don’t touch it, this is a winner. A forced error, as Mike West notes, is where you still got your racquet to the ball. However, winners and forced errors are pretty similar in contrast to unforced errors. ↩
probably following a thorough security review ↩
sigh, if you really think I’m not working in good faith it’s a lost cause… ↩