There Is No Such Thing as Spam in a Priced System
The moment you decide which paying transactions deserve a slot, you have built a permissioned network and given it a euphemism
Keywords: payment finality; permissionless systems; fee markets; transaction filtering; institutional economics; regulation; governance; rationing; information insensitivity; blockchain scaling
Thesis statement: Calling a fee-paying, rule-valid transaction “spam” is not an engineering classification. It is an admission-control decision, and admission control is governance. Any system that decides which paying users may transact — by payload type, by purpose, by “standardness”, or by escalating penalties aimed at disfavoured uses — has instituted a permission regime, has acquired regulators, and has thereby abandoned the one property that made it distinctive. The alternative is not moral policing but arithmetic: scale the capacity, price the resource, and let the payer and the miner settle it between them.
1. The word is doing work no argument is doing
Spam has a meaning. It has had one since long before anyone tried to move money over a public ledger. Spam is unsolicited bulk messaging: content pushed at a recipient who did not ask for it, in volumes the sender could not afford if the sender bore the cost, imposing a cost on someone who receives no compensation. Every element of that definition matters, and every element is doing analytical work. Unsolicited. Bulk. Uncompensated. Imposed.
Now take a transaction that is signed correctly, formatted correctly, spends outputs the signer controls, carries a fee at or above the rate the market is clearing at, and is accepted by a miner who chose to include it. Apply the definition.
It is not unsolicited: the miner published a price and the sender met it. It is not uncompensated: the sender paid, and the miner collected. It is not imposed on a non-consenting party: the only party obliged to store and serve the data is the party being paid to do so, and any other node participates voluntarily under rules it knew before it joined. It may well be bulk. Bulk is not a defect. Every payment network on earth would prefer more volume to less, and any network that would not is not a payment network but a club.
So the word does not fit. It is being used because it carries moral freight that the underlying argument cannot generate on its own. “This transaction is economically trivial and I would prefer it did not exist” is a preference. “This transaction is spam” sounds like a finding. The substitution is rhetorical, and once you notice it, the rest of the position becomes visible: it is not a claim about abuse, it is a claim about entitlement to decide.
That claim deserves to be examined on its own terms rather than smuggled in under a borrowed word. This essay examines it.
2. Premises, stated openly
Because the argument that follows is a normative and institutional one, the premises it rests on are set out here rather than left implicit. Each is either definitional or empirically checkable; none is offered as self-evident.
P1. Block space is a scarce, produced resource. Producing it consumes bandwidth, storage, validation cycles and — under proof of work — energy. This is a matter of engineering fact, not preference.
P2. A fee is a price paid for that resource, and its acceptance by a miner is a sale. Both sides act voluntarily.
P3. “Permissionless” means, at minimum, that admission to the ledger is conditioned on objective rule-validity and on payment, and not on any further judgement about who the payer is or what the payer wants the entry to mean. If it means anything weaker than that, it means nothing, because every controlled system in history has permitted the uses its controllers approved.
P4. Rules that are applied uniformly to all payers (signature validity, no double spending, a defined serialisation) are constitutive of the system. Rules that sort among rule-valid, fee-paying transactions on the basis of content or purpose are not constitutive; they are discretionary.
P5. Where a discretionary rule exists, someone wrote it, someone maintains it, and someone can change it. That someone holds power over who may transact.
P1 through P5 are the whole scaffolding. If you reject P3 — if you are content that “permissionless” merely means “no formal application form” — then we disagree about vocabulary and the rest of this essay will read as an elaborate statement of the obvious. If you accept P3, the conclusion follows with unpleasant tightness.
3. What a fee actually is
A fee is not a tax, a toll, or a donation. It is a price, and prices do something specific: they allocate a scarce resource to whoever values it most, without anyone needing to know why they value it.
That last clause is the entire moral content of a price system. The price mechanism is indifferent to purpose. It does not ask whether you are buying steel to build a hospital or a casino. It does not ask whether your use is dignified. It asks whether you will pay, and if you will, the resource moves. This is not an incidental feature of markets that we tolerate for efficiency’s sake. It is the mechanism by which decentralised systems avoid needing a central authority competent to evaluate purposes. Remove indifference to purpose and you have not improved the price system. You have replaced it with something else and kept the invoice.
When a miner accepts a transaction at the going rate, two parties have expressed consent. The sender has demonstrated that the entry is worth at least the fee. The miner has demonstrated that the fee is worth at least the marginal cost of inclusion. Nothing further is required for the transaction to be legitimate, because nothing further is required for any other market transaction to be legitimate. The fact that a third party finds the entry silly is not a defect in the transaction. It is a fact about the third party.
The counter-move at this point is usually to say that block space is different because inclusion imposes costs on other nodes, who are not paid. This is a real argument and it deserves a real answer, which comes in Section 9. But note what it is: an externality argument. Externality arguments licence pricing, metering, and cost-recovery. They do not licence content classification. If the problem is that a resource is under-priced, the remedy is to price it. Deciding which payers are worthy is a different act with a different justification, and the externality argument does not supply that justification.
4. The three-part test, applied
Set out the test formally, because the informal version keeps getting waved through.
For an activity to be spam in any sense that carries moral weight, three conditions must hold jointly:-
Non-solicitation. The recipient did not offer to receive it.
-
Non-compensation. The recipient bears a cost and receives nothing in return.
-
Non-consent of the burdened party. The party who bears the cost had no opportunity to decline.
A fee-paying transaction accepted by a miner fails all three. The miner solicited it by publishing a fee schedule and running a mempool. The miner is compensated, by construction. And the miner consented, transaction by transaction, by building it into a template it was free not to build.
What about the non-mining node that must store and relay the data? It fails condition three as well, and this is the part people skip. A node operator chose to run software implementing a published ruleset, knowing what that ruleset permitted. If the ruleset permits data-carrying outputs, then data-carrying outputs are part of what the operator agreed to carry. Discovering afterwards that you dislike some of the traffic you volunteered to relay does not retroactively convert the traffic into an imposition. It converts you into someone with a grievance about your own configuration, which you are free to change.
The honest version of the complaint is: I want the network to be cheap for me to run, and volume makes it expensive. That is a legitimate interest. It is also an interest in a smaller network, held by a party who would prefer the costs of smallness to fall on the excluded rather than on themselves. It should be argued in those terms and weighed against the interests of the people being excluded. Instead it is argued as sanitation.
5. Two gates, and only one of them is a rule
Here is the structural point on which everything turns.
A transaction passes through two conceptually distinct gates before it reaches a block. The first gate is consensus validity: does the signature verify, are the inputs unspent, is the serialisation well-formed. This gate is objective, verifiable by anyone with the same data, and symmetric — it treats every payer identically because it does not look at who the payer is or what the payer intends. It is a rule in the strictest sense.
The second gate is policy. Is this payload of an approved type? Is this output above the threshold we have decided is dignified? Is this use one we consider a proper use of the ledger? This gate is not objective, is not verifiable in the same way, and is not symmetric. It sorts among transactions that all passed gate one and all paid.
Figure 1. Gate 1 is a rule. Gate 2 is a judgement about the payer. Everything that lives in Gate 2 is governance, whatever the software calls it.
The reason to draw the picture is that the two gates are routinely presented as one thing — “the protocol” — when they are two things with entirely different legitimating logics. Gate one needs no legitimation beyond consistency: it is the thing everyone agreed to. Gate two needs the whole apparatus of legitimation that any exercise of authority needs, and it has none of it.
Now decompose gate two. Karen Yeung’s analysis of algorithmic regulation gives the cleanest available test. She treats regulation as a cybernetic process with three components: ways of setting standards, ways of gathering information, and ways of enforcing those standards to modify behaviour when deviation is identified. Her taxonomy of algorithmic regulation is built by varying each component; and she notes that the simple forms — fixed, reprogrammable behavioural standards enforced automatically — are continuous with a long tradition of control exercised through architecture and design rather than through law (Yeung 2018).
Apply the three components to a transaction filter.
Standard-setting. Someone must define what counts as an acceptable transaction. Not valid — acceptable. That definition is written by people, encoded, and shipped.
Information-gathering. Every candidate transaction must be inspected against the definition. The system must look at the payload, not merely at the signature.
Enforcement and behaviour modification. Non-conforming transactions are dropped, delayed, or surcharged, with the explicit purpose of changing what senders do.
All three components are present. On Yeung’s own definition, a transaction filter is not “policy” in some deflationary technical sense. It is a regulatory system implemented in code, with fixed standards and automatic enforcement — precisely the archetype she identifies as the most basic form of algorithmic intervention.
Julia Black’s definition closes the loop from the other direction. She defines regulation as sustained and focused attempts to change the behaviour of others to address a collective problem or attain identified ends, usually through a combination of rules or norms together with means of implementation and enforcement, which may be legal or non-legal; and she stresses that regulatory functions may be exercised by one actor or dispersed among many, with the degree of dispersal determining how polycentric the regime is (Black 2008). Dispersal is not absence. A regime with no central locus of authority is still a regime. Black’s entire subject is how such regimes construct and contest legitimacy precisely because they cannot claim it from a constitutional settlement.
That is the situation of a filtered ledger exactly. Sustained, focused attempts to change the behaviour of senders. A collective problem, as defined by the rule-writers. Rules plus enforcement. Dispersed across implementers, relay operators and pool operators. It satisfies the definition in every particular. The only thing missing is the acknowledgement.
6. What kind of thing has just been created
Call it what it is in institutional terms, because the terminology exists and it is precise.
John Searle’s account is the sharpest. An institution, on his analysis, is a system of constitutive rules of the form X counts as Y in C, where the Y term assigns a status function; and the essential point of such institutions is not to constrain people but to create power relationships of a particular kind — the deontic powers marked by rights, duties, obligations, authorisations, permissions, empowerments, requirements and certifications (Searle 2005).
Read a standardness rule against that template. This payload, of this type, within these limits, counts as a permissible transaction on this network. That is X counts as Y in C with no paraphrase required. The rule assigns a status function. The status function carries deontology: some senders are authorised, others are not. Permission is not a metaphor here; it is the technical term for the thing that has been created. A network with a policy filter is, on the most influential philosophical account of institutions in circulation, a permission-granting institution.
Frank Hindriks and Francesco Guala’s rules-in-equilibrium framework sharpens it further. They argue that rule-based and equilibrium-based conceptions of institutions are each inadequate alone but jointly satisfactory, and — this is the part that bites — that constitutive rules are regulative rules augmented by the introduction of new theoretical terms (Hindriks and Guala 2015). “Standard transaction” is exactly such a theoretical term. It does not name anything in the consensus rules. It is introduced by a regulative rule and then treated as though it described a natural kind. Once the term is in circulation, arguments can be conducted in it — is this transaction standard? — and the prior question of who was entitled to introduce the term disappears from view. That disappearance is the mechanism by which governance becomes invisible to the governed.
Sinclair Davidson, Primavera De Filippi and Jason Potts make the same point from the economics side, and they make it approvingly. Their argument is that blockchains are best understood not as a general-purpose technology but as an institutional technology: a new mode of economic coordination and governance, standing alongside markets, firms, relational contracting, clubs, commons and states, and expanding the scope of economic governance rather than merely making existing processes cheaper (Davidson, De Filippi and Potts 2018).
I have no quarrel with that framing; it is correct. But notice what follows from it. If the thing is an institutional technology, then design choices within it are institutional choices, and institutional choices are political in the ordinary sense — they distribute rights, duties and permissions among people. A decision about which transactions may enter the ledger is not analogous to a decision about buffer sizes. It is analogous to a decision about who may open an account. De Filippi and Wright put the general form of this plainly: design choices in blockchain protocols reflect discrete choices, and where rules are both defined and enforced by autonomous code, people may be left with little or no recourse against an unfair application of those rules (De Filippi and Wright 2018).
Little or no recourse. Hold that phrase; it returns in Section 11.
7. Coase, and the network that quietly became a firm
There is an older and more devastating way to state the point, and it is worth stating because it converts a political objection into an economic one.
Ronald Coase’s question in 1937 was why firms exist at all in an economy where the price mechanism is supposed to allocate resources. His answer began with a definition: the distinguishing mark of the firm is the supersession of the price mechanism. Outside the firm, price movements direct production, coordinated through exchange transactions in the market. Inside the firm, those market transactions are eliminated and replaced by an entrepreneur-coordinator who directs production. When a worker moves from department Y to department X, Coase observed, they do not move because relative prices changed; they move because they were told to. He drove the point home by noting that in law, too, it is the fact of direction that is the essence of the employment relation — the right to control what work is done and how (Coase 1937).
Now consider what a content-based filter does. Before the filter, block space is allocated by price: pay the rate, get the slot, and the reason for wanting the slot is nobody’s business. After the filter, block space is allocated by price within a set of uses that an authority has approved. Direction has been substituted for price at exactly the margin where the substitution matters. The allocation of a scarce resource among competing uses is now, in part, an administrative decision.
By Coase’s own definition, that is the boundary of a firm. The network has not become more decentralised by filtering; it has taken a step from the market side of the boundary to the hierarchy side. It has acquired an entrepreneur-coordinator — diffuse, unelected, and unpaid for the role, but performing the function. Coase was careful to note the one important asymmetry between planning imposed on an industry and the planning that occurs within a firm: firms arise voluntarily because they are a more efficient way to organise production, whereas economic planning is imposed. A filter shipped as the default behaviour of the reference implementation of a supposedly neutral protocol is much closer to the second case than the first, because the people subject to it did not choose a firm. They chose a market and were handed a hierarchy.
This is why the “miners can just choose” reply does not work, and it needs to be dealt with head-on. Of course an individual miner may decline any transaction; that is ordinary commercial freedom and nobody disputes it. The governance question is not about the individual decision. It arises when a normative rule about which payers are legitimate is embedded in widely deployed default software, propagated through relay policy so that non-conforming transactions cannot even reach the miners who would accept them, and defended as protocol rather than as preference. At that point the discretion is no longer distributed. It has been centralised into whoever maintains the default, and then re-described as emergent.
8. A cap does not solve the allocation problem; it relocates it
Now the economics, which is where the position collapses of its own accord.
Capacity in these systems is either capped or elastic. The choice is not between a cap and chaos; it is between two different allocation regimes with different institutional requirements.
Figure 2. With a hard cap, growth in demand shows up as a fee spike plus a mass of unserved demand that somebody must allocate. With elastic, priced capacity, growth shows up mostly as volume. Panel A creates a rationing problem; Panel B does not.
Panel A is the capped case. Supply is a vertical line. When demand shifts out, quantity cannot respond, so the entire adjustment falls on price. Two things follow, and both are consequential.
First, the fee spike is borne by every user, including the ones the cap was ostensibly protecting. A cap does not shield small payers from congestion pricing; it guarantees that they experience the maximum possible congestion pricing, because they have no other adjustment margin. The rhetoric of protecting ordinary users by restricting capacity has the sign wrong.
Second, and more importantly, at any price the community is willing to tolerate, there is a wedge of demand that will not be served. That wedge has to be allocated somehow. If it is not allocated by price, it will be allocated by something else — by queue position, by relationship, by conformity to whatever the rule-writers approve of. This is the mechanism by which “we need a cap” becomes “we need to decide whose transactions matter”. The second sentence is not a moral failing on the part of the people who say it. It is the logical consequence of the first sentence. Once you fix quantity below the quantity demanded, you have created a rationing problem, and rationing problems are solved by rationers.
Panel B is the elastic case. Supply slopes upward: more capacity is available, at a cost that rises with quantity because bandwidth, storage and validation are not free. Demand growth is absorbed mostly as volume and partly as price. Nobody needs to hold an opinion about which transactions deserve a slot, because the question never arises. The market clears.
Note what Panel B does not claim. It does not claim capacity is free, unbounded, or costless to supply. The supply curve slopes up precisely because it is not. The claim is narrower and stronger: that the costs of capacity should be recovered through the price of capacity, and that once they are, the allocation problem dissolves. Whether a given fee schedule actually recovers the full marginal cost — including the long-run cost of unspent-output set growth, which is the one genuinely hard case — is an engineering and pricing question that deserves a serious answer rather than a slogan. I am not pretending otherwise. What I am denying is that failure to answer that pricing question licences a content-classification regime instead. Under-pricing calls for better pricing.
The strongest counter-argument, stated fairly
There is a serious formal argument that deliberately setting a price below the market-clearing level can raise welfare, and intellectual honesty requires putting it on the table rather than pretending the field is one-sided.
Yingyi Qian’s model of shortage does exactly this. He shows that a benevolent government maximising social surplus may rationally decline to raise a price to the market-clearing level. The mechanism is a screening one: where firms operate under a soft budget constraint — the state bank cannot credibly commit not to refinance bad projects once costs are sunk — and where the seller cannot separate firms from households competing for the same good, a market-clearing price would exclude households without deterring the inefficient projects, because those projects are not disciplined by price. Rationing then acts as a screening device: a bad project may obtain finance but not inputs, which deters the manager from applying in the first place. Qian is explicit that shortage carries real costs — misallocation, delay, queuing, hoarding, rent-seeking — and equally explicit that his result is peculiar to the soft-budget environment: under a hard budget constraint, bad projects are screened out directly by the difficulty of refinancing, so there is no social benefit to holding prices below clearing (Qian 1994).
That is the counter-argument in its strongest form, and it does not support a transaction filter. It supports the opposite conclusion, for the reason Qian himself supplies. The screening rationale requires that price fail to discipline the demand you want to deter. In a fee market, price disciplines everyone, because every payer bears the full cost of their own transaction out of their own funds. There is no state bank refinancing the sender. Budget constraints are as hard as they get. Under Qian’s own conditions, the case for below-clearing rationing evaporates and the residual — misallocation, queuing, hoarding, rent-seeking — is all that is left.
I state this as an argument from non-satisfaction of Qian’s conditions, not as a proof that no rationing rationale could ever exist. If someone wishes to construct one for block space, the burden is theirs, and it must identify the specific market failure that price cannot address and show that a content filter addresses it better. No such argument has been offered. What has been offered is a vocabulary.
9. The externality reply, and why it does not reach content
Return to the argument deferred in Section 3, because it is the only technically respectable version of the case for restriction.
The claim is that a transaction imposes costs on parties who do not receive the fee: every node that must receive, validate, store and serve the data forever. If so, the fee under-prices the resource, and under-priced resources are over-consumed. This is correct as far as it goes, and it identifies a real design problem.
But observe carefully what it licences and what it does not.
It licences metering — charging by the resource actually consumed rather than by a crude proxy. It licences pricing state growth separately from throughput, since a transaction that permanently enlarges the set of unspent outputs imposes a different cost profile from one that does not. It licences charging for relay, or letting relay be a paid service rather than a volunteered subsidy. It licences requiring that the marginal node’s costs be covered by the marginal transaction’s fee. Every one of these remedies operates on the quantity of resource consumed, is neutral as to purpose, and requires no one to form a view about the payer.
It does not licence deciding which payloads are legitimate. A data-carrying output and a payment output that consume identical bandwidth and identical storage impose identical costs. If the cost is the problem, they should be priced identically. Treating them differently is not cost-recovery; it is preference expressed as pricing. And the moment a fee schedule discriminates by purpose rather than by resource, the schedule has stopped being a price and become a fine.
This distinction — between metering a resource and judging a use — is the whole ballgame, and it is the distinction that the word “spam” is deployed to blur.
10. What filtering costs you: money stops being money
There is a further cost, and it is the one least discussed, because it operates on a property of the system that is invisible when it is working.
Gary Gorton and Guillermo Ordoñez’s account of collateral crises rests on a micro-foundation with a clean statement. Short-term collateralised debt — private money — is efficient precisely when agents are willing to lend without producing costly information about the collateral backing it. Such debt is designed to be information-insensitive: it is not worth anyone’s while to investigate what stands behind it, and so trade proceeds without fear of adverse selection. When a shock makes information production profitable again, the debt becomes information-sensitive, agents start checking, and the machinery of exchange seizes (Gorton and Ordoñez 2014).
That is a statement about money in general, and it is the deepest thing anyone has said about why bearer instruments work. Cash is accepted because accepting it requires no inquiry. The moment acceptance requires inquiry, the instrument’s liquidity is impaired, because inquiry is costly and its very possibility introduces the fear that the party who did inquire knows something you do not.
Now consider what a content filter does to a ledger entry. It makes admission conditional on an inspection of the entry’s contents against a standard. It creates a class of transactions that are valid and paid but may not be included. It therefore makes the question will this be accepted? a question that senders must investigate, and investigation is exactly the cost that information-insensitivity exists to avoid. A filtered ledger is an information-sensitive payment system by construction. It has traded the property that makes an instrument money-like for the ability to express disapproval of some of its users.
Honesty requires reporting the qualification that Gorton and Ordoñez themselves attach, because it cuts against a naive reading. They show that a social planner would produce more information than private agents do, and would not always want to eliminate fragility, because information acquired today has option value tomorrow that private agents — who do not internalise the future — ignore. So “never produce information” is not an unconditional social optimum in their model. The claim licensed here is narrower and is the one I make: on the static exchange margin, information-insensitivity is what makes a bearer instrument liquid, and a filter destroys it. Whether some information production is dynamically optimal for the system as a whole is a separate question, and the answer to it is not “let the maintainers of a relay policy decide which payloads offend them”.
11. The reply that permissionlessness was never the point
Some will say that permissionlessness was always a slogan, that every system has rules, and that objecting to filtering rules is objecting to rules as such.
This confuses P4 with P3, and the confusion is worth dismantling because it is the most common evasion in the debate.
Every system has rules. The question is what kind. A rule that says a signature must verify applies to every payer identically, is checkable by every participant against the same data, and does not require anyone to hold an opinion about anyone. A rule that says this class of use is not welcome applies selectively, requires interpretation, and requires an interpreter. The first kind constitutes the system. The second kind governs the people in it. Searle’s distinction between constitutive and regulative rules maps onto this exactly, and Hindriks and Guala’s demonstration that constitutive rules can be reconstructed as regulative rules plus theoretical terms tells you how the second kind disguises itself as the first: by introducing a term like “standard” and then reasoning in it.
There is a legitimate position available to anyone who wants a filtered network, and it is this: we are running a governed system with an admissions policy, we think that is correct, and here is our answer to the governance questions. That position is intellectually respectable. It is also the position nobody wants to occupy, because occupying it means answering the questions.
The questions are not rhetorical. Black’s work on polycentric regimes is precisely an account of how bodies exercising regulatory power without a constitutional settlement must construct legitimacy and are then held to it, through accountability relationships that are dialectical and contested rather than granted (Black 2008). A network with an admissions policy has entered that world, and the entry fee is:-
Who sets the standard? Name the body. Not the process by which patches merge, the body that decides which uses are legitimate.
-
By what procedure? Is there notice? Comment? A record of reasons?
-
On what evidence? Is there any measurement showing that the excluded traffic imposes costs its fees do not cover, or is exclusion driven by aesthetics?
-
With what appeal? De Filippi and Wright’s warning is that code-enforced rules can leave people with little or no recourse against unfair application. What is the recourse here?
-
Accountable to whom? Users? Node operators? Miners? Holders? These groups have divergent interests and the answer determines whose interests the filter serves.
-
Removable how? A power that cannot be withdrawn is not a policy. It is a constitution, adopted without a convention.
Any regime that exercises this power and cannot answer these six questions is not neutral infrastructure. It is unaccountable authority with good branding. That is a harsher description than most participants would accept, and it follows directly from the definitions they themselves would endorse if the definitions were applied to anyone else.
12. The evidence that cuts the other way
The rules of this exercise require reporting evidence that cuts against the position, not merely the evidence that supports it. There is such evidence, and it is substantial.
Eric Budish’s analysis in the Quarterly Journal of Economics is the most serious economic critique of permissionless consensus in the literature, and it does not support techno-optimism of any variety, mine included. His argument is three equations. A zero-profit condition on the suppliers of trust support implies that the recurring payment to them is fully dissipated in the cost of supplying it. An incentive-compatibility condition on security against majority attack implies that the cost of attack must exceed the gain. Linking them yields the constraint that the recurring flow cost of maintaining the system must be large relative to the one-off stock value of attacking it — with the consequence that the cost of securing permissionless consensus scales linearly with the value secured, becoming, in his phrase, preposterous at global scale. He is explicit that this critique applies to permissionless consensus in the sense of Nakamoto (2008) generally, across proof of work, proof of stake, and any other mechanism, and that it is a critique of the pure form without implicit support from the rule of law (Budish 2025).
That result should be reckoned with rather than waved away, and it is not answered by anything in this essay. Two observations, offered as inference from his framework and marked as mine rather than attributed to him.
First, his constraint bears on the value secured, not on the number of transactions processed. Throughput and value at stake are distinct quantities. A system processing many small transactions and a system processing few large ones can face very different positions in his inequality.
Second, and more pointedly, in his zero-profit condition the recurring payment to trust support is what funds security. Fees are a component of that payment. Restricting transaction volume therefore restricts the fee base that must, on his own logic, be large relative to attack value. Whatever else a capacity cap does, it does not relax Budish’s constraint; it tightens the funding side of it. If one takes his analysis seriously — and it should be taken seriously — the case for maximising the fee base through volume is stronger, not weaker.
Neither observation rescues permissionless consensus from his critique. I record it as an unresolved challenge, because the alternative is to cite only what flatters the argument, and that is the intellectual vice this essay is complaining about in the first place.
13. What “scale it and price it” actually requires
It would be cheap to end on the slogan. The position carries obligations, and they should be stated.
Capacity must actually scale. Asserting that a system can process arbitrary volume is not the same as demonstrating it. Throughput claims require measurement under adversarial conditions, at sustained load, with honest reporting of where the ceilings are.
Pricing must cover marginal cost, including state growth. The hardest case is a transaction that permanently enlarges the set of unspent outputs, imposing a cost that recurs for every future validator. If a fee schedule does not price that, it is not a complete answer to the externality objection, and saying so is not a concession — it is a specification of work to be done.
Relay should be a service, not a subsidy. Much of the “who pays for this?” complaint arises because a class of participants performs costly work for free and is then surprised to discover that free things get used. Pricing relay dissolves the grievance at its source.
Node operators must be able to choose their own cost level. The choice between running a full archival node and running something lighter is a legitimate one, and the answer to “archival storage is expensive” is architectural, not prohibitionist.
None of these is trivial. All of them are engineering and pricing problems with engineering and pricing answers. Not one of them requires anybody to decide whose money is welcome.
14. Conclusion
State it plainly.
A transaction submitted under published rules, carrying the required fee, and accepted by a miner is neither unsolicited nor free. The sender has purchased block space. The miner has sold it. That is a completed market transaction between consenting parties, and a third party’s distaste for it is not a finding of fact.
Classify it plainly.
The moment a system sorts among rule-valid, fee-paying transactions on the basis of what they are for, it has instituted standard-setting, inspection and enforcement — the three components of any regulatory system. It has created status functions and the deontic powers that accompany them: some payers authorised, others not. It has substituted direction for price at the margin where the substitution defines the boundary of a firm. It has become an information-sensitive payment system, which is to say a worse one. And it has acquired regulators who cannot say who they are, by what authority they act, or how they may be removed.
Whether that regime is desirable is a question on which reasonable people may differ. Whether it is permissionless is not. It is not. The word has a meaning, and the meaning is incompatible with a body that decides which paying users deserve a slot.
Exponential penalties are not required. Moral categories are not required. What is required is capacity that scales, prices that cover costs, and the discipline to let a market clear without appointing anyone to supervise its taste.
State. Classify. Done.
References
Each work below was read in full. Links are to the publisher’s or archive’s landing page as surfaced and verified; no link has been constructed from an identifier.
Black, J. (2008). Constructing and contesting legitimacy and accountability in polycentric regulatory regimes. Regulation & Governance, 2(2), 137–164. https://onlinelibrary.wiley.com/doi/full/10.1111/j.1748-5991.2008.00034.x
Budish, E. (2025). Trust at scale: the economic limits of cryptocurrencies and blockchains. The Quarterly Journal of Economics, 140(1), 1–62. https://academic.oup.com/qje/article/140/1/1/7824430
Coase, R. H. (1937). The nature of the firm. Economica, 4(16), 386–405. https://onlinelibrary.wiley.com/doi/10.1111/j.1468-0335.1937.tb00002.x
Davidson, S., De Filippi, P., & Potts, J. (2018). Blockchains and the economic institutions of capitalism. Journal of Institutional Economics, 14(4), 639–658. https://www.cambridge.org/core/journals/journal-of-institutional-economics/article/abs/blockchains-and-the-economic-institutions-of-capitalism/7DFE353FEB8FA25559E6C09FE77E8291
De Filippi, P., & Wright, A. (2018). Blockchain and the Law: The Rule of Code. Cambridge, MA: Harvard University Press. https://www.jstor.org/stable/j.ctv2867sp
Gorton, G., & Ordoñez, G. (2014). Collateral crises. American Economic Review, 104(2), 343–378. https://www.aeaweb.org/articles?id=10.1257%2Faer.104.2.343
Hindriks, F., & Guala, F. (2015). Institutions, rules, and equilibria: a unified theory. Journal of Institutional Economics, 11(3), 459–480. https://www.cambridge.org/core/journals/journal-of-institutional-economics/article/institutions-rules-and-equilibria-a-unified-theory/CD04E89BC01304500EFF2B7845E17BD7
Qian, Y. (1994). A theory of shortage in socialist economies based on the “soft budget constraint”. American Economic Review, 84(1), 145–156. Issue landing page: https://www.jstor.org/stable/i337076 (an article-level stable URL for this item was not surfaced by search; the issue page listing the article is given rather than a constructed link)
Searle, J. R. (2005). What is an institution? Journal of Institutional Economics, 1(1), 1–22. https://www.cambridge.org/core/journals/journal-of-institutional-economics/article/abs/what-is-an-institution/3675101CE15BE2A7681CD5783C01F6D0
Yeung, K. (2018). Algorithmic regulation: a critical interrogation. Regulation & Governance, 12(4), 505–523. https://onlinelibrary.wiley.com/doi/10.1111/rego.12158