Censorship Resistance, Atomic Settlement, and the Limits of Coalition Power

A reply to the two-tier critique...

A reply to the two-tier critique: why fresh-address transaction granularity is the structural answer that state-licensed analogies cannot reach.

Keywords: censorship resistance; atomic UTXO settlement; transaction granularity; fresh-address regime; mining coalitions; chain analysis; regulatory technology; structuring; propagation topology; economic versus political power

I. Introduction

A recent critique of my work on the propagation topology of miners offers two concessions and three pushbacks. The concessions are generous and correct.

The pushbacks are sharp, well-formed, and worth answering at the level they deserve. They are also, I will argue, answering a question that the protocol's native operation does not raise.

The critique grants two structural points. First, that the small-world property of the miner subgraph, combined with positive expected revenue from propagating fee-bearing transactions, implies that a sender's access problem reduces to reaching any single competent operator. The interlocutor accepts this as a real structural property of the system. Second, that the empirical map of deployed direct-submission infrastructure — ARC on the BSV implementation, comparable channels on other implementations — is useful and, if verified, narrows the two-tier concern substantially.

The pushbacks then move in three directions. The first is that my definition of censorship resistance proves too much: every two-sided market with multiple service providers satisfies it, including the postal service and the banking system. The second is that the propagation argument has a boundary condition where regulatory exposure dwarfs fee revenue and a black-market regime emerges. The third is internal: my coalition analysis, developed across the related sequence of essays, requires that miners coordinate strategically, while the propagation argument requires that they act independently as fee-maximisers; the two, the interlocutor claims, cannot both be true under all conditions.

These are serious arguments. They deserve serious replies. They also share a single load-bearing assumption that, once identified and removed, dissolves the force of all three at once. The assumption is that a Bitcoin transaction is a large, identifiable, individually consequential transfer between persistent counterparties. The protocol was not designed for that model and does not operate naturally in it. The protocol was designed for atomic settlement with single-use keypairs, and its censorship-resistance property is a property of that mode of operation, not of the legacy-payment model the critique implicitly imports.

The remainder of this piece develops the structural answer. It begins with the definitional question, proceeds to the granularity question that is the heart of the matter, examines what regulatory technology can and cannot reach at atomic scale, returns to the coalition tension in light of the granularity, and closes with the architecture implied when the system operates at the volumes its design supports.

II. What the Postal Service Analogy Actually Says

The critique's first move identifies a class of systems that satisfy my definition of censorship resistance: a transaction whose sender is willing to compensate a competent operator cannot be prevented from inclusion by any non-operator. The postal service, banks, commodities exchanges, taxi dispatch — all of these fit. The interlocutor concludes that censorship resistance, on this definition, is a property of any multi-operator competitive market and not distinctive to Bitcoin. He grants that the identity of the gatekeepers differs across these systems but treats this as a difference of less consequence than the similarity of outcome under regulatory pressure.

The first half of the observation is correct. The second half inverts the controlling variable. The identity of the gatekeepers is the categorical distinction, because the manner of admission to gatekeeper status is what determines whether the market is structurally resistant to interdiction or merely conditionally resistant to it.

Consider what postal-service censorship resistance actually rests upon. In the United States, the postal service is protected by the Private Express Statutes (18 U.S.C. §§1693–1699), which prohibit private carriage of non-urgent letters and thereby establish the state's effective monopoly over the defined service. Comparable protections exist in most jurisdictions. The censorship resistance the critique attributes to the postal system is not a property of competitive entry; it is a property of a state-granted monopoly enforced by statute against entry. The non-postal operator cannot prevent delivery because the state has made it unlawful for there to be a non-postal operator who could.

The same structure governs the banking system. Banks operate under state chartering. The clearing systems through which interbank transfers flow are state-operated or state-supervised. The non-bank cannot prevent a wire transfer because the state has defined the operator class to exclude non-banks. When the state instructs that operator class to filter — through OFAC, the Bank Secrecy Act, FinCEN, or analogous regimes — the filter operates universally across the licensed operators, because there is no unlicensed operator class to defect to.

The categorical distinction is therefore between two species of censorship resistance.

The first is granted censorship resistance. It exists as a positive policy of the state, which has established and protected an operator class capable of fulfilling service requests. It is real, but it is conditional on the state's continued willingness to grant it. The state that grants it can withdraw it, restrict it, or instruct the operator class to apply selective filters. The operator class cannot resist these instructions because its existence depends on the state.

The second is structural censorship resistance. It arises from open entry into the operator class — entry that does not require permission, licensing, or charter, and that the state cannot prevent without coordinated global jurisdiction over the relevant technology. The state cannot instruct this operator class to apply filters because there is no comprehensive list of who is in it. Members can defect. New members can enter. Entry from outside the state's jurisdiction is structurally available.

The relevant work of Ayn Rand offers the right vocabulary for this distinction. Rand drew a sharp line between economic power, which is the offer of value in voluntary exchange, and political power, which is the initiation of physical force — including the force employed to define who may legally offer a service. The postal service exists in the domain of political power; its monopoly is enforced by criminal sanction against unauthorised entry. The protocol's mining market exists in the domain of economic power; participation is voluntary, entry is permissionless, and the exchange between sender and operator is consensual on both sides. The two systems satisfy the same surface definition for entirely different metaphysical reasons.

This is not a rhetorical reframing. It is the controlling distinction. Once one sees that the postal service's censorship resistance is a function of who the state has granted operator status to, the apparent equivalence between the cases collapses. The relevant question for the protocol is not whether the definition is satisfied — many systems satisfy it — but whether the operator class is open or closed. In Bitcoin, by design, it is open.

III. The Transaction Granularity the Critique Assumes

The second pushback identifies a regime where propagation incentives invert: transactions for which regulatory exposure dwarfs fee revenue. The critique points to OFAC-compliant mining pools, sanctioned addresses, and the coordinated filtering of transactions that touch flagged counterparties.

This argument is well-formed and would be devastating if its implicit transaction model were the protocol's actual operating mode. It is not. The argument assumes — without naming the assumption — that transactions are large, that addresses persist across transactions, and that the unit of economic transfer is the unit of regulatory identification. None of these is a feature of the protocol's native operation.

Consider the transaction model the critique requires. A sanctioned address must exist as a persistent identifier, used across multiple transactions, listed on the OFAC SDN inventory, and matchable at mempool ingestion. A miner faces regulatory risk that dwarfs any fee because the transaction is identifiable as connected to a listed entity, large enough to attract regulatory attention if included, and consequential enough to expose the miner to liability. The compliance department the critique invokes is performing per-transaction screening against a list of identified counterparties.

This is the legacy payment model in everything but name. It is the model in which transactions are large, counterparties are identified, identity persists across exchanges, and a screening regime that operates on listed identities can be applied efficiently to each economic event. It is the model that built the banking system's regulatory apparatus, and it is the model the critique imports without notice when it constructs its boundary case.

The protocol's design is the opposite model. Section 10 of the original Bitcoin whitepaper recommends that a new key pair should be used for each transaction to keep them from being linked to a common owner. Atomic settlement on the UTXO graph treats each transfer as a discrete event between freshly generated keypairs. There is no protocol-level concept of a persistent counterparty. There is no protocol-level concept of identity. There is no protocol-level concept of a sanctioned address, because addresses are not subjects with histories; they are ephemeral keys generated for single use and discarded.

Consider what the critique's boundary case looks like under this model. A user wishes to transfer one million dollars of value. In the legacy framing this is a single transaction from a persistent address, identifiable, subject to screening, and within reach of any filtering regime that operates on the unit of transfer. In the protocol's native framing this is one million single-dollar transactions, each from a freshly generated keypair, each below any conceivable regulatory reporting threshold, each individually a transfer of trivial economic consequence between counterparties that have no prior existence in any database.

The OFAC SDN list, the FATF travel-rule thresholds, the Bank Secrecy Act Currency Transaction Report obligation, the FinCEN Suspicious Activity Report regime — all of these technologies are built to identify and act upon transactions of a certain shape: a transfer above a defined threshold, from or to an identified counterparty, processed by a regulated intermediary. The atomic transaction below the threshold, from a fresh address, with no regulated intermediary in the chain, is a shape these technologies were not designed to see.

This is not an evasion technique. It is the operation the protocol was built to perform. The legacy regulatory apparatus has matured into its present form against the backdrop of a payment infrastructure that produces large, identified, intermediated transactions. The protocol produces atomic, peer-to-peer transfers between fresh keypairs. The mismatch between the regulatory technology and the protocol's native output is not a contingent gap that can be closed by adapting enforcement; it is a structural feature of the protocol's design.

IV. Why Structuring Law Does Not Reach the Native Mode

A natural objection at this point is that the regulatory regime already anticipates the breaking-up of large transactions into small ones. The relevant statute in the United States is 31 U.S.C. §5324, the structuring offence, which prohibits the act of evading reporting requirements by breaking a transaction into multiple smaller transactions each below the reporting threshold. Comparable provisions exist in other jurisdictions. If the protocol's native mode is to perform a single economic transfer as a stream of small transactions, does this not simply fall within the existing structuring prohibition?

The objection has surface appeal and dissolves on examination. The structuring offence has three operative elements. First, there must be a regulated reporting obligation that the small transactions would, in aggregate, otherwise trigger. Second, there must be a regulated intermediary on whom the reporting obligation falls. Third, there must be intent to evade reporting — the offence is not the small transaction itself, but the act of breaking a larger transaction to avoid triggering a reporting requirement that would otherwise apply.

None of these elements is naturally present in the protocol's atomic operation. There is no regulated reporting obligation that attaches to a peer-to-peer transfer through the protocol; the reporting obligations of the Bank Secrecy Act attach to financial institutions defined as money service businesses, banks, or money transmitters under the statute, and they apply to transactions those institutions process. The protocol's miners are not, in the operational sense the statute contemplates, the originators or the senders' bankers; they are operators offering inclusion services to whichever transactions are presented to them. There is no regulated intermediary in the sense the structuring statute contemplates, because the transaction flow is end-to-end between sender and recipient, with the miner functioning as a settlement provider rather than as a custodial bank.

The intent element fails most plainly. The protocol was designed for atomic transactions with fresh keypairs. This is its native mode of operation. A user who sends a stream of small transactions because the protocol naturally produces small transactions is not breaking a single transaction to evade reporting; he is using the protocol as it was designed to be used. The intent the statute targets — the specific intent to evade a reporting obligation — is absent when the relevant reporting obligation does not attach in the first place.

Attempts to extend the regulatory regime to cover this mode have been made and have largely failed at the protocol level for reasons that are structural rather than contingent. FinCEN's 2013 guidance on virtual currencies defined money transmitter broadly enough to capture some intermediaries operating in the cryptocurrency space, but the guidance and its successors apply to identified intermediaries — exchanges, custodial wallet providers, mixers — and do not, in operation, reach the miners or the protocol itself. The reason is not regulatory restraint; it is that the regulation requires a regulable entity, and the protocol's native operation does not provide one at the inclusion layer. A miner offering inclusion services to whichever transactions are broadcast on the network does not occupy the same regulable position as a money transmitter who custodies funds and processes identified transfers between named parties.

Structuring law, in short, applies where the regulated intermediary exists, where the reporting obligation exists, and where the intent to evade can be demonstrated. The protocol's native operation does not produce the first or the second, and absent the first and second, the third has nothing to attach to.

V. Chain Analysis at Atomic Scale

The natural rejoinder is that chain-analysis tools — Chainalysis, TRM Labs, Elliptic, and their competitors — exist precisely to address this gap. These tools cluster addresses by behavioural heuristics, tag clusters with identified entities, and produce forensic reports that allow regulators and law enforcement to pierce the apparent anonymity of fresh-address transactions. The critique's two-tier picture implicitly assumes that this technology is available to mining pools at the ingestion layer and that filtering policies can be implemented against the cluster identities rather than against individual addresses.

This assumption mistakes what chain analysis is for. Chain analysis is a post-hoc forensic technology. It runs against historical transaction data, identifies probable connections between addresses through statistical heuristics — the common-input-ownership heuristic, change-address detection, behavioural clustering, temporal correlation — and produces reports of varying confidence about which addresses are likely controlled by which entities. It is used by exchanges to flag suspicious deposits days or weeks after they occur, by law enforcement to construct cases against identified suspects, and by compliance teams to satisfy retrospective reporting obligations. It is not used, and cannot feasibly be used, to make per-transaction inclusion decisions at the speed and cost a mining operation requires.

Three structural features prevent chain analysis from operating as a pre-ingestion filter. First, the cluster inference is probabilistic, not deterministic. The common-input-ownership heuristic assigns a likelihood that multiple addresses are controlled by the same entity; it does not produce a certain answer. At the inclusion decision point, the miner faces a binary choice — include or exclude — based on probabilistic input. Errors are inevitable, and the miner who excludes a transaction that turns out to be benign has forgone fee revenue for no compensating benefit.

Second, the computational cost is asymmetric to the transaction value. Running a chain-analysis classifier against a fresh-address transaction in the mempool requires non-trivial compute and data access. The cost is approximately constant per transaction regardless of the transaction's value. For a single large transfer, the screening cost is small relative to the regulatory exposure; for a stream of single-dollar transfers, the screening cost per dollar processed becomes enormous, and the regulatory exposure per transfer is negligible. The economics of compliance break.

Third, and most decisively, the freshness of the addresses defeats the clustering heuristics that chain analysis depends on. The common-input-ownership heuristic produces signal when a transaction has multiple inputs and one can therefore infer that the spender controls all of them. A transaction with a single fresh-address input and a single fresh-address output produces no such signal. Behavioural clustering produces signal across repeated address use; addresses used once and discarded produce no behavioural pattern to cluster. Temporal correlation produces signal when transactions follow recognisable patterns; transactions distributed across a stream of single-use addresses without identifying metadata produce statistical noise rather than identifiable structure.

Chain analysis, then, is well-suited to the legacy use pattern in which addresses are reused, transactions are large and infrequent, and economic flows leave persistent traces. It is poorly suited, and at sufficient granularity is structurally inapplicable, to the protocol's native pattern of fresh-address atomic transactions at high volume. The forensic technology is built for one regime; the protocol natively operates in another.

The implication for the critique's two-tier picture is direct. The picture requires that compliance filtering can be implemented at the inclusion layer against transactions connected to flagged counterparties. The technology that would implement that filtering exists for legacy patterns and does not exist for the protocol's native mode. The two-tier picture is a description of what happens to a specific transaction shape under specific filtering technology. It is not a description of what happens to atomic-mode operation, because the filtering technology cannot identify the object of its filter.

VI. The Coalition Cannot Coordinate on a Classifier

The third pushback is the deepest and the one most worth engaging at length. The critique identifies an internal tension: my coalition analysis treats miners as strategic actors who coordinate around shared incentives, while my propagation argument treats them as independent profit-maximisers who forward every fee-bearing transaction they see. The interlocutor suggests both are true under different conditions, with the coalition mechanism activating under regulatory or political stress and producing coordinated filtering that the small-world propagation argument cannot account for.

The reconciliation is structural and was implicit in the original framing but warrants explicit development. The coalition mechanism operates strongly on protocol-rule coordination and weakly on transaction-level coordination. The asymmetry is not stipulated; it follows from the cost structure facing miners in each context.

Protocol-rule coordination has the properties that sustain Folk-Theorem-type equilibria. The cooperative surplus is large — a stable rule set is enormously more valuable than a fractured one. Defection is observable but gradual; signalling occurs through public debate, client adoption, and hash-rate trends, long before any commitment is made. The cost of defection — forking off, losing network effects — is concentrated and durable. And, crucially, the coalition retains the ability to punish defection across multiple subsequent rounds because the defector remains a recurring participant in the protocol whose future cooperation matters.

Transaction-level coordination has none of these properties. The cooperative surplus is small per transaction. Defection is observable and instantaneous — the block containing the censored transaction is broadcast at the moment of mining, and every coalition member sees that the included transaction was the one the coalition was supposed to filter. The reward to defection is collected immediately in the form of the block reward and the included transaction's fee. The defector cannot be punished in future rounds in any meaningful sense, because mining is permissionless and exit from a punishing coalition has no consequence beyond losing the coalition's coordination benefits, which are minimal at the transaction level to begin with.

The atomic-transaction model adds a second, deeper reason that coalition coordination fails at the inclusion layer. To coordinate on filtering identified counterparties, coalition members need a shared list of objects to filter. The OFAC SDN list is such an object. A coalition that announces a policy of refusing to include transactions from addresses on this list can in principle implement that policy by reference to the list. But this requires the addresses to be on the list. In the fresh-address regime, there is no list to share, because the addresses do not exist until the transaction is broadcast. The coalition would need to coordinate not on a list of identified objects but on a classifier: a statistical procedure that produces a per-transaction judgment about whether the transaction is, probabilistically, connected to a flagged flow.

Coordinating on a classifier is qualitatively harder than coordinating on a list. Different members will deploy different classifiers, with different false-positive and false-negative rates. The same transaction may be classified differently by different members. The coalition's filtering decision becomes ambiguous. Defection becomes ambiguous: a miner who includes a transaction the coalition's consensus classifier would have flagged may have been defecting, or may simply have been running a classifier that returned a different result. The Folk Theorem requires that defection be unambiguously identifiable so that punishment can be applied to the right party. Statistical classification with member-specific error rates breaks this condition.

A coalition that attempts to enforce a sufficiently strict classifier — one that errs on the side of exclusion to suppress ambiguous transactions — faces a separate failure mode. The classifier produces false positives. The included rate of legitimately fee-paying transactions falls. Hash rate within the coalition that disagrees with the classifier's calls observes that profitable transactions are being excluded and exits to a non-classifying pool. The exit is permissionless. The coalition loses hash rate. The classifier's effectiveness, measured by share of network applying it, declines.

The coalition mechanism, in short, can sustain coordination on protocol rules because the relevant object is shared, identifiable, and consequential over many rounds. It cannot sustain coordination on transaction-level filtering in the fresh-address regime, because the relevant object is statistical, contested across members, and consequential within a single round in which the defector has already collected the reward.

VII. Propagation, Pools, and the Direct-Submission Layer

The previous arguments establish that filtering cannot be coordinated effectively at the inclusion layer in the atomic-transaction regime. The original propagation argument made a stronger claim: that the sender's access problem reduces to reaching any single competent operator. This claim is independently supported by deployed infrastructure that the critique correctly identifies as worth taking seriously.

Direct-submission infrastructure such as ARC on the BSV implementation, and analogous channels on other implementations, allows transactions to be submitted to miners without traversing the gossip propagation layer. The sender presents the transaction directly to a mining operation; the mining operation includes it in a candidate block if profitable. The intervening propagation network is not a precondition of inclusion; it is one method of access among several.

This bears on the critique's two-tier framing in two ways. First, it confirms that the access problem is structurally solvable without requiring universal propagation. A sender whose transaction would not naturally propagate retains the option of direct submission. The cost of direct submission is a small operational overhead in establishing the submission channel and is not a function of the transaction's content. The channel does not screen transactions; it conveys them.

Second, the existence of direct-submission infrastructure changes the unit of the reach-one-equals-reach-all argument. The original claim was about propagation across the miner subgraph. With direct submission, the claim becomes simpler: reach any miner you can address directly. If the addressable hash rate is substantial — and on current evidence, on multiple implementations, it is — the access problem is solved without reliance on the propagation topology at all.

In the atomic-transaction regime the case for direct submission is weaker per transaction, since each transaction is below the threshold at which special handling is worthwhile. But the propagation incentive is correspondingly stronger. A miner who receives a fresh-address one-dollar transaction in the mempool faces no cost reason to suppress it. The transaction is below any plausible regulatory threshold, attached to no identified counterparty, and carries a small but positive fee. The marginal incentive is to include and to propagate. The aggregate effect across millions of such transactions is that the propagation layer operates as designed: every fee-bearing transaction reaches the operators capable of including it, because no individual transaction is worth the cost of suppressing.

Pools, in this picture, are aggregation devices for variance reduction in mining rewards. They have no special handle on the transaction set their constituent hash rate processes. A pool that announces a filtering policy in the fresh-address regime is announcing a classifier; the classifier produces errors; the errors cost hash rate via exit to non-classifying pools. Pool-level coordination on filtering is unstable for the same Folk-Theorem reasons that coalition-level coordination is, with the additional fragility that pool affiliation is more easily reassigned than coalition membership.

VIII. Billions of Transactions Per Second

The picture sharpens further when the system operates at the volumes its design supports. The protocol scales, in implementations that have removed artificial block-size limits and built the supporting infrastructure, to transaction throughputs that can be reckoned in millions, and at architectural limit, in billions per second. The specific numbers depend on hardware deployment and network topology; the order of magnitude is determined by the protocol's combinatorial structure rather than by any operational constant of present mining hardware.

At these volumes, every claim made above hardens. A regulator attempting to filter transactions at the inclusion layer would face the task of screening billions of fresh-address atomic transactions per second. A chain-analysis classifier would need to run, in real time, against each. The false-positive cost would be the suppression of legitimate fee-paying transactions and the consequent exit of hash rate from the screening regime. The false-negative cost would be the inclusion of transactions the regulator wished excluded. The screening cost would scale linearly with transaction count; the regulatory benefit, against a transaction set dominated by trivial economic transfers, would not.

The legacy regulatory apparatus is built for the throughput of legacy payment systems: thousands of transactions per second in aggregate, with screening focused on identified flows above defined thresholds. The protocol at scale presents a transaction set whose aggregate throughput exceeds the legacy apparatus by several orders of magnitude and whose individual transactions are below every threshold the apparatus is built to apply. The mismatch is not a quantitative gap that can be closed by adding compute; it is a structural mismatch between two architectures of value transfer.

This is the architecture the protocol's original design implies. A peer-to-peer cash system, in the whitepaper's framing, is one in which transfers are atomic, addresses are fresh, and the volume of transactions is bounded by the technology of settlement rather than by the technology of intermediation. The legacy apparatus is intermediation technology. The protocol at scale is settlement technology. Censorship resistance under settlement technology is a property of the inability of intermediation-based regulatory tools to reach the units the settlement layer naturally produces.

IX. Conclusion

The critique's three pushbacks share a single load-bearing assumption: that Bitcoin transactions are large, identifiable, and bound to persistent counterparties. Under this assumption, the postal-service analogy works, the OFAC boundary case has bite, and the coalition mechanism can in principle activate against flagged flows. Each of these conclusions follows from the assumption, rather than from any feature of the protocol.

The protocol's native operation does not satisfy the assumption. Transactions are atomic. Addresses are fresh. The unit of economic transfer is decoupled from the unit of regulatory identification. The screening technology that would implement the critique's two-tier picture is designed for a transaction shape the protocol does not natively produce, and is structurally inapplicable at the granularity the protocol natively operates.

The deeper distinction, in Rand's vocabulary, is between economic and political power. Political power is the power to define an operator class and to compel that operator class to apply filters. Economic power is the power to offer value in voluntary exchange. The legacy systems the critique cites as analogues operate within political power: their operator classes are state-defined and state-compelled. The protocol operates within economic power: its operator class is open-entry, its exchanges are voluntary on both sides, and its output is a stream of atomic transfers that produce no objects of coercion for the legacy apparatus to act upon.

Censorship resistance, then, is not a property the protocol shares with every two-sided market. It is a property of a specific architecture in which the operator class is permissionless, the transaction granularity is atomic, and the identifying machinery of the legacy regulatory apparatus has no native handle on the units the architecture produces. The market is not a cage. The market is also not a cartel. The market is the natural form of value transfer when the technology of settlement has rendered the technology of intermediation unnecessary.