[bitcoin-dev] Documenting the lifetime of a transaction during mempool congestion from the perspective of a rational user

[Jeremy]

Devs,

This email is primarily about existing wallet behaviors and user
preferences, and not about CTV. However, towards the end I will describe
the relevance of CTV, but the email is worth reading even if you have no
interest in CTV as the problems described exist today.

One point of confusion I've seen while discussing CTV based congestion
control is that it requires a bunch of new wallet software.

Most of the software requirements that would make CTV work well are things
that either already exist in Bitcoin Core, or are 'bugs' (where bug is
defined as deviation from rational utility maximizing behavior) that should
be fixed *whether or not CTV exists.*

In this post, I am going to walk through what I expect rational behavior to
be for a not unlikely circumstance.

First, let's define what rational behavior for a wallet is. A rational
wallet should have a few goals:

1) Maximize 'fully trusted' balance (fully confirmed + unconfirmed change
outputs from our own txns)
2) Process payments requested by the owner within the "urgency budget"
requested by the user.
3) Maximize "privacy" (this is a vague goal, so we'll largely ignore it
here.).

Rational wallet behavior may not be possible without metadata. For example,
a rational wallet might prompt the user for things like "how much do you
trust the sender of this payment to not RBF this transaction?", or "how
much do you trust this sender to not double spend you?". For example, a
self-transfer from cold wallet to hot wallet could have a trust score of 1,
whereas a payment from an anonymous source would have a trust score of 0.
Exchanges where you have a legal agreement to not RBF might sit somewhere
in between. Other pieces of exogenous information that could inform wallet
behavior include "has hashrate decreased recently, making longer reorgs
likely".

In the model above, a user does not request transactions, they request
payments. The rational wallet serves as an agent to assist the user in
completing these payments. For example, if I have a wallet with a single
unconfirmed output, and I spend from it to pay Alice, if the unconfirmed
gets replaced, my wallet should track that it was replaced and prompt me to
re-sign a new transaction. Rational wallets that maximize balance should be
careful to ensure that replaced payments are exclusive, guaranteed either
through sufficient confirmations or 'impossibility proofs' by reusing an
input (preventing double-send behavior).

-----------------------------

Now that we've sketched out a basic framework for what a rational wallet
should be doing, we can describe what the process of receiving a payment is.

Suppose I have a wallet with a bevy of fully confirmed coins such that for
my future payments I am sufficiently funded.

Then, I receive a payment from a highly trusted source (e.g., self
transfer) that is unconfirmed.

I then seek to make an outgoing payment. I should have no preference
towards or against spending the unconfirmed transfer, I should simply
account for it's cost in coin selection of CPFP-ing the parent transaction.
If fees are presently historically low, I may have a preference to spend it
so as to not have a higher fee later (consolidation).

Later, I receive payment from an untrusted source (e.g., an anonymous
donation to me). I have no reason to trust this won't be double spent.
Perhaps I can even observe that this output has been RBF'd many times
previously. I do not count on this money arriving. The feerate on the
transaction suggests it won't be confirmed immediately.

In order to maximize balance, I should prioritize spending from this output
(even if I don't have a payment I desire to make) in order to CPFP it to
the top of the mempool and guarantee I am paid. This is inherently "free"
since my cost to CPFP would be checked to be lower than the funds I am
receiving, and I have no expected value to receive the payment if it is not
confirmed. If I do have a transaction I desire to do, I should prioritize
spending this output at that time. If not, I would do a CPFP just in favor
of balance maximizing. Perhaps I can piggyback something useful, like
speculatively opening a lightning channel.

If I just self-spend to CPFP, it is very simple since the only party set up
for disappointment is myself (note: I patched the behavior in this case to
accurately *not* count this as a trusted balance in
https://github.com/bitcoin/bitcoin/pull/16766, since a parent could disrupt
this). However, if I try to make a payment, my wallet agent must somehow
prompt me to re-sign or automatically sign an alternative payment once it
is proven (e.g. 6 blocks) I won't receive the output, or to re-sign on a
mutually exclusive output (e.g., fee bumping RBF) such that issuing two
payments will not causes a double-send error. This adds complexity to both
the user story and logic, but is still rational.

Now, suppose that I receive a new payment from  a **trusted** source that
is a part of a "long chain". A long chain is a series of transactions that
are all unconfirmed in the mempool. This long-chain is in the bottom of the
mempool, and is not expected to confirm any time soon.

My wallet should be configured such that it saves not only all ancestors of
the transaction paying me, but also all descendants of the root unconfirmed
transaction paying me. If I do not save all ancestor transactions, then it
would be impossible for me to claim this payment at a future date, and
would violate balance maximization. But why save descendants, if they do
not concern me? Descendant transactions are critical for balance
maximization. Someone else's spend of an output is a "free" CPFP subsidy
for driving my transaction to completion (perhaps "descendants that
increase the feerate of any parent" is the correct thing to save).
Therefore if I want to maximize balance, I would rather keep these
transactions around should I ever need to rebroadcast the transactions as
it should be cheaper than having to CPFP by myself.

Now, suppose that I receive a similar payment in a longchain from a series
of untrusted sources. The same arguments apply, but now I may have even
higher incentive to prioritize spending this coin since, if sender's trust
scores are independent, my total trust in that payment is decomposed
worst-case geometrically. It may not be a good assumption that trust scores
are independent, since a long chain might be generated as e.g. a series of
batch payments from one service provider.

Briefly mentioned above is rebroadcasting. This is sort of an orthogonal
behavior to the above, but it is "simple" to explain. Wallet agents should
retransmit txns to the network if they believe other nodes are not aware of
them and they are likely to go into a block. This encapsulates personal
transactions as well as arbitrary transactions from third parties. There
are many privacy implications of rebroadcasting that are out of scope for
this post.

-----------------

All of the behaviors documented above are things that should happen today
if you would like to have a rational wallet that maximizes your balance and
makes payments.

The reasons we don't do some of these things are, as far as I can tell:

1) Engineering Complexity
2) UX Complexity (simpler to make unconfirmed outputs "unspendable" to
minimize transaction reissuing)
3) Mempool backlog is low, things are confirmed quickly if a sender pays a
relatively low fee

Certain wallets already have parts of this functionality baked in to an
extent. For example, in Lightning Channels, you will drive payments to
completion to prevent HTLC timeouts during contested closes (HTLCs == low
trust score payments).

Should Bitcoin see development of a more robust fee market, it is highly
likely the rational behaviors described above would be emergent among all
bitcoin wallets (who would want to use a Bitcoin wallet that gets you less
money over time?). This email is not just a "Bitcoin Core" thing, hence not
being an issue on Bitcoin Core where there are currently deviations from
the above behaviors.

-----------------

What's CTV got to do with it?

A common critique of congestion control using CTV is that it complicates
wallet behavior because congestion control is designed to be useful in the
circumstances above. CTV and congestion control do not cause these
conditions. These conditions already exist whether or not we have
congestion control.

Where congestion control helps is that, in a world with a full mempool, you
have fewer payments that are *actually* unconfirmed because exchanges that
batch can fully confirm a constant sized root transaction and the sub-trees
of transactions locked in by CTV can be treated as fully trusted. This
helps reduce the need for the (rational) behavior of CPFP bumping your own
payments on receipt from lower trust senders. Further, the expansion of the
transaction tree can be done by other users receiving, so you have an
incentive to wait for funds to mature as someone else might pay for
expansion. These two factors mean that CTV congestion control can exert a
dramatic back pressure on transaction urgency by unbundling the blockspace
demand for spending and receiving coins. There are other synergies -- such
as non-interactive channel opens -- that further improve the amount of
reduction of time-preference in full on-chain resolution.

I hope this email helps clarify why CTV Congestion Control isn't
particularly creating a wallet architecture problem, it's helping to solve
an existing one.

Best,

Jeremy



--
@JeremyRubin <https://twitter.com/JeremyRubin>
<https://twitter.com/JeremyRubin>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20220113/0cad9c8d/attachment-0001.html>