Digital Settlement Without Extraction

Volume III — Money, Energy, Information

THE PLATE

A woman in a cream blouse standing in a sunlit studio, light falling through the windows in long bars across the wall. — Plate III.09The Counter at Closing.Every payment you have ever received arrived through somebody's pipe, and somebody set the price of the pipe. The question is only whether you have ever seen the number.

THE LETTER

There is a number on your merchant statement that you have probably never argued with, because it arrived looking like weather.

It is somewhere between one and a half and three per cent of everything you sell. It is called the merchant discount rate, and about four-fifths of it is interchange — a fee your bank pays the customer's bank for the privilege of having been paid. It does not vary with how hard the payment was to make. It does not fall when the pipe gets cheaper. It is charged on the value of the sale, which means it rises with your prices, and it has risen with them: American merchants paid $187.20 billion in card fees in 2024 and $198.25 billion in 2025.

This chapter is about the pipe, and only the pipe. Chapter III.03 asks what a currency should be made of and whether a community can issue its own; that is a question about the unit of account, and it is a good one. This is the plumbing question underneath it, which is narrower and more immediately worth money: once two people have agreed what they owe each other, what should it cost to move it, and who should capture the difference between that cost and the price?

The answer turns out to be measurable, because four large countries have already run the experiment in public and published their figures. Brazil, India and the United Kingdom each built an instant payment rail as public infrastructure, with no extractive margin in it. We can read what it cost, what it charged, what it did to the shops, and — this is the part nobody enjoys — what it could not do, which is sell itself.

We will also take the distributed-ledger case seriously, on its own terms, with its own numbers. It deserves better than either the enthusiasm or the dismissal it usually gets, and when you run the arithmetic it turns out to be answering a question that the public rails had already answered more cheaply.

You will end this chapter able to compute what your own payment rail costs you as a share of your net profit, which is the only ratio that has ever moved a finance director.

— The Editors

DISCOVERY

What is already working

Begin where somebody has already built it and published the numbers.

Pix, in Brazil. The Banco Central do Brasil launched Pix on 16 November 2020: instant transfers between any two accounts, settling in central bank money, free to individuals, running twenty-four hours a day. In calendar 2025 it carried 79.8 billion transactions worth R$35.36 trillion. By the second half of 2025 it was carrying 54.7 per cent of Brazil's retail payment transactions — more than half of everything, five years after launch.

The figure that matters most is the one hardly anyone quotes. The Banco Central charges participating institutions R$0.01 for every ten credits settled — a tenth of a centavo per transaction. On 79.8 billion transactions that is R$79.8 million a year for the whole national settlement layer. Against the R$35.36 trillion it moved, the central bank's charge amounts to 0.000226 per cent of value.

Hold that number. We will use it.

UPI, in India. The Unified Payments Interface opened in April 2016 and by the financial year 2024–25 carried 185.8 billion transactions worth 260.6 lakh crore rupees — 83.4 per cent of the volume of India's entire payments ecosystem. Merchant discount rate on those payments, from January 2020: zero, set by statute through Section 10A of the Payment and Settlement Systems Act and Section 269SU of the Income-tax Act. A merchant with a printed QR code accepted digital money at no cost whatsoever, which is why perhaps two million of them put one up.

Faster Payments, in the United Kingdom. Live since 27 May 2008, and the oldest of the three. In 2025 it carried 5.5 billion payments worth £4.8 trillion. Pay.UK, the operator, ran a cost base of £161.1 million across all of its schemes that year — Faster Payments, Bacs, the Image Clearing System and the standards work together. Attribute the whole of it to Faster Payments alone, which overstates it substantially, and the ceiling on what one instant payment costs to exist is £0.0293 — under three pence. As a share of the value it moved: 0.00336 per cent.

And here is the detail that tells you what kind of institution this is. In March 2026 Pay.UK gave £27.5 million back to its participants, because it had collected more than it needed. Divide that by the year's volume and the error term on a public rail's pricing was half a penny a payment. A commercial rail does not return an over-collection. It books it.

iDEAL, in the Netherlands. A fourth model, and the one an executive finds easiest to copy, because no central bank is involved. Dutch banks built a shared online payment scheme in 2005 and priced it as a flat fee per transaction rather than a percentage of the basket. A flat fee has one property that a percentage fee never has: it is honest about where the cost actually is. Moving €40 and moving €4,000 take the same messages, the same fraud checks and the same storage, and a scheme that charges the same for both is telling the truth about its own economics.

Four rails, four jurisdictions, one shared finding. In every case the cost of settlement was discovered to be far smaller than the price that had been charged for it, and in every case the discovery was made by building the cheap thing rather than by arguing about the expensive one.

THE ARITHMETIC

What works, what does not, and where the line sits

First, what the fee actually is.

In 2024, American merchants paid $1.57 in fees for every $100 of cards accepted. That implies card volume of $11.92 trillion and it means the fee is not small: divided across 131.4 million American households, card acceptance fees came to $1,424.66 per household per year — a number that appears on no household's statement, because it is in the shelf price.

The legislated comparisons show how much of that is a price rather than a cost. The European Union caps interchange at 0.2 per cent on debit and 0.3 per cent on credit. The combined Visa and Mastercard credit interchange rate in the United States reached 2.36 per cent in 2025. On a $50 basket that is $1.18 against a European cap of $0.15 — the same networks, the same silicon, the same message format, 7.87 times the price, with the difference being nothing but which regulator was watching.

Second — and this is the cut — what the fee is for.

The intuition is that the merchant discount rate buys settlement. It does not. Set the American interchange rate beside the only published figure we have for what settlement costs a central bank to provide, and the wedge is not a margin, it is a different order of magnitude:

  Brazil, credit card merchant discount rate        2.2 %
  Brazil, average merchant cost of a Pix payment    0.22 %
  Banco Central's charge for settling that Pix      0.000226 % of value
  ------------------------------------------------------------------
  card MDR  /  central bank settlement              9,748 x
  Pix merchant cost  /  central bank settlement       975 x

Read the second ratio before the first, because it is the honest one. Pix costs a merchant roughly nine hundred and seventy-five times what the central bank charges to settle it — and that gap is real work: fraud screening, disputes, onboarding, support, the app, the refund. Nearly all of the 0.22 per cent is somebody doing something. Which is precisely what makes the first ratio indefensible. If the whole apparatus of a modern payment business fits inside 0.22 per cent, then the other 1.98 per cent is not apparatus.

So what is it? Here the arithmetic takes a turn that most people do not see coming, and it comes from the Federal Reserve Bank of Boston. Schuh, Shy and Stavins calibrated who actually bears interchange once merchants recover it in the shelf price, which they do, because almost none of them price differently by payment method. Their finding: the average cash-using household pays $149 a year to card-using households, and the average card-using household receives $1,133 — a ratio of 7.60. Sorted by income, the household earning under $20,000 pays $21 a year and the household earning over $150,000 receives $750.

The merchant discount rate is not a price for moving money. It is a transfer running uphill, and it is collected at the till of the shop where the poorest customer buys. The card rail is not expensive because settlement is hard. It is expensive because the rewards have to be paid for, and they are paid for by the people who do not get them.

Third, the distributed ledgers, assessed on their own numbers.

The promise was to remove the intermediary who collects that wedge. Take it seriously and measure it.

  rail            finality                energy per transaction
  ----------------------------------------------------------------
  Bitcoin         60 min, probabilistic           1,100 kWh
  Ethereum (PoS)  12.8 min, deterministic             35 Wh
  Visa            seconds                             1.5 Wh
  Pix / UPI / FPS seconds, irrevocable        conventional scale

Bitcoin's network draws about 138 TWh a year, which at 1,100 kWh a transaction implies 125.5 million transactions annually — 3.98 per second. Pix averages 2,530 per second, 636 times more. Run Pix's 2025 volume at Bitcoin's per-transaction energy and you would need 87,780 TWh a year, which is 2.93 times the electricity the world generates.

Now the honest handling, because the per-transaction figure is a contested one and a chapter that quotes it without saying so has not done its work. Proof-of-work energy is spent per block, not per transaction: one more payment in a block costs essentially no extra energy, so 1,100 kWh is an average and not a marginal cost, and batching or a payment channel moves it a long way. The figure without that problem is the security budget. Bitcoin issues 3.125 coins per block, 144 blocks a day: 164,250 coins a year, which at a round $100,000 is $16.43 billion a year, or $130.92 per transaction — paid by holders as dilution rather than by payers as a fee, which is exactly why it appears on nobody's receipt.

Ethereum's Merge is the strongest single data point in the whole field: a 99.988 per cent cut in consumption, from an industrial load to 2,600 MWh a year, with no change to what the ledger does. It is proof that the energy was never the function. And at 35 Wh a transaction Ethereum is now within striking distance of conventional rails — 31,429 times better than Bitcoin, and 23.33 times worse than Visa.

The honest conclusion for this movement is unflattering to the thing we would like to be true. Of everything a distributed ledger promises a payer — instant, cheap, final, open to anybody with an account, no single company setting the price — a public instant-payment rail delivers all of it, faster (seconds against 12.8 minutes at best), cheaper (0.000226 per cent of value at the settlement layer), and at conventional energy. Where the ledgers remain genuinely superior is the case where you do not trust the settlement authority — cross-border, sanctioned, censored, or simply a country whose central bank you have good reason to doubt. That is a real and important case. It is not the corner shop.

And now the negative that governs the whole chapter. A rail with no extractive margin has no salesforce. Nobody's commission depends on your merchant signing up, so nobody knocks. Look at what each country had to pay to overcome that:

  rail    launched   measured   years   share   mechanism
  ---------------------------------------------------------------------
  FPS       2008       2025      17     11.1 %  neither mandate nor commission
  UPI       2016      FY2025      9     83.4 %  zero MDR plus state subsidy
  Pix       2020       2025       5     54.7 %  mandatory participation

Share gained per year: the United Kingdom 0.65 points, India 9.27, Brazil 10.94 — fourteen to seventeen times faster where somebody bought the acceptance. (The three shares sit on different denominators, so read this as orders of magnitude of adoption speed and not as a league table.)

India's bill for buying it is public: 1,389 crore rupees in the year to March 2022, then 2,210, then 3,631, and an outlay of 1,500 in the year to March 2025 — 8,730 crore, about $992 million, and even in its largest year that covered only 18.2 per cent of the industry's estimated 20,000 crore annual cost of running UPI. On 15 October 2026 India ends zero MDR: 0.4 per cent on merchant payments above ₹2,000, capped at ₹300. Still 5.90 times cheaper than American credit interchange. But no longer free, and the reason is the one this movement began with. Somebody has to be paid to build acceptance, and if it is not a commission it has to be a mandate or a budget line.

DREAM

What becomes ordinary

In the economy where this has already happened, the payment is not a product.

The shopkeeper's statement has a line for the rail and the line is small enough that she stops modelling it. It is priced per transaction, not per unit of value, because everyone now understands that moving R$40 and moving R$4,000 cost the same to move. Her prices fall by a little less than the fee she stopped paying and her margin rises by the rest, and neither movement is dramatic, and both are permanent.

The fee that used to sit inside the shelf price is gone from it, which means the customer paying with a card no longer has the customer paying with anything else buying her rewards. Nobody experiences this as a redistribution. They experience it as the shop being slightly cheaper, which is what a redistribution feels like when it is unwound at the right place.

The settlement layer is boring in the way that water is boring. It is a regulated utility with a published tariff, it recovers its costs and returns what it over-collects, and its annual report is read by twelve people. Its uptime is a national statistic. Its price is a line in a schedule that anybody can look up, and when the price changes there is a consultation, because it is infrastructure and everybody knows it.

Around it, competition is fierce — and it has moved to where it belongs. Banks and fintechs compete on the app, the fraud engine, the dispute experience, the credit product sitting beside the payment, the speed of onboarding a merchant on a Sunday afternoon. They do not compete on access to settlement, because settlement is not a thing anybody owns. The margin has moved from the pipe to the service, which is the only place a margin has ever been honestly earned.

The distributed ledgers are still here and they are doing the work they are actually good at: moving value across a border between two parties who share no authority, settling a claim where the counterparty will not take your court's word for anything, holding a registry that must survive a government that wishes it would not. They are no longer proposed as the way to buy coffee, and the field is far stronger for having stopped saying so.

And the most ordinary thing of all: a person who has never been asked for a credit check can be paid. Not through a wallet that requires a card underneath it, not through an account tier — paid, in final money, into an account that costs nothing to hold.

DESIGN

The structure that gets there

Four components. They have to arrive roughly in this order, because each one makes the next one cheap.

One: settlement at cost, in the safest money available.

The settlement layer is the natural monopoly and it is the piece that must not carry a margin. Brazil's tariff — R$0.01 per ten credits — is the working example of what "at cost" looks like when somebody publishes it. The design rule is that the layer recovers its costs, holds a stated amount of risk capital, and returns the remainder. Pay.UK's £27.5 million rebate is that rule operating, and it is the single most persuasive fact about public rails: a utility that over-collects gives it back, and the giving back is what proves there was no margin to begin with.

Two: a price per message, never a price per unit of value.

The percentage fee is the entire disease. It has no engineering basis — the payment does not get harder as the number attached to it grows — and it silently indexes the payment industry's revenue to inflation and to the price of housing, fuel and food. Charge per transaction. Cap any percentage that survives. India's new schedule does both at once: 0.4 per cent above ₹2,000, capped at ₹300, which means the cap bites at ₹75,000 and above that the rail is effectively priced per message. That is a percentage fee with a conscience, and it is a reasonable place to land if the politics will not carry a flat fee.

Three: acceptance, bought deliberately, in the open.

This is the part everybody skips, and it is the part the arithmetic says decides the outcome. There are exactly four ways to get a rail onto a shop counter and you must choose one and pay for it.

Mandate participation. Brazil required institutions above a size threshold to join. Cost to the state: near zero. Cost to the banks: real, and they paid it. Fastest of the four.
Subsidise the merchant side. India paid 0.15 per cent of value on small merchant transactions, which is roughly a seventh of what the card rail would have charged, and it bought nine points of share a year. Honest, expensive, and it ends — as it did on 15 October 2026.
Let the banks own it and price it flat. The Dutch route. No public money, slower than a mandate, and it works because the owners are the same institutions that would otherwise be paying interchange to each other.
Do none of the above. The British route, and the control group. Seventeen years, 11.1 per cent, an excellent rail that most people have still never knowingly used to buy anything in a shop.

Four: competition above the rail, protected by rule.

Open the rail to non-bank participants on published criteria, mandate that every account-holding institution can be reached, and forbid the settlement operator from competing with its own participants. Then leave the app layer entirely alone. The whole point of taking the margin out of the pipe is that it shows up somewhere a customer can choose to pay it.

DESTINY

How it holds when nobody is pushing

A public rail is durable in a specific way and fragile in a specific way, and it is worth being precise about both.

What makes it hold. The volume is the moat. Once a rail carries half a country's retail transactions it is not removable by anything short of legislation, and legislation about a thing 150 million people use daily is legislation nobody wants to sponsor. The economics compound in the right direction: the cost per transaction falls as volume rises, so the tariff can be cut, which raises volume. And the rebate mechanism is quietly the strongest governance device in the whole design, because a participant who receives money back has evidence that the operator is not a rentier, and that evidence is worth more than any amount of assurance.

Where it fails, named plainly.

It fails at acceptance, always, and for the reason we priced: there is no commission, so there is no salesforce. Britain is the honest control group here. Seventeen years of an excellent rail, and it reached 11.1 per cent of payments because nobody was ever paid to put it in front of a shopper.

It fails when the subsidy ends before the habit forms. India spent 8,730 crore and then reintroduced a merchant fee — and it worked only because by then UPI carried 83.4 per cent of ecosystem volume and the habit was set. Had the money run out in year four, the same decision would have read as a broken promise and the rail would have stalled.

It fails when the operator is also a competitor. A settlement utility owned by the largest participants will underinvest in reaching the smallest ones, quietly and without anybody having to decide to.

And it fails on the security budget nobody costed. A rail carrying 2,530 transactions a second is a national-scale fraud target, and fraud losses on instant, irrevocable payments land on the payer in a way that card chargebacks never did. The card rail's expensive 2.2 per cent buys a dispute process, and the cheap rail has to fund an equivalent out of something. Any design that does not name where that money comes from has moved the cost rather than removed it.

DELIGHT

What it feels like

The pleasure is a small one and it arrives on a Tuesday.

A shopkeeper looks at the second page of her statement — the page nobody reads — and finds that the line for payment acceptance has become too small to be worth her attention. She stops checking it. Some weeks later she notices that she has stopped doing the small arithmetic she used to do at the counter, the one where a customer wanting to pay by card for something under ten euros made her hesitate for half a second before saying yes. The hesitation is gone. She did not decide to stop; it simply became untrue.

And then the better pleasure, which belongs to the person paying. The money leaves and arrives while you are still standing there. Not pending, not authorised, not clearing — arrived, in the other person's account, final. You can see it on their screen. There is something specific and good about watching a stranger's face register that they have actually been paid, and realising that this is what all the machinery was always supposed to do and that it took twelve seconds.

A payment that costs almost nothing stops being an event. That is the whole delight: a thing that used to be a transaction becomes a gesture.

OPERATIONALIZE THIS

At the level of finance

The instrument: a merchant acceptance mutual, funded by a levy set at one twentieth of the fee it displaces.

The honest negative said that a rail with no margin has no salesforce. Here is the structure that hires one without reintroducing the margin, and it is a structure a treasurer will recognise immediately, because it is a purchasing cooperative with a sunset.

The problem in one line. Acceptance costs real money to build — terminals, QR estate, integration, disputes, support, training — and on a rail priced at cost there is nobody whose revenue pays for it. So the beneficiaries pay for it collectively, at a tenth of what they were paying a network to do it badly.

The mechanics.

Members. Merchants with payment volume above a stated floor, plus their trade association or chamber. Anyone may join on the same terms; the terms are published. That is what keeps it a mutual rather than a buyers' club.
The levy. 0.10 per cent of migrated volume, collected by the acquiring participant and remitted to the mutual. On a pooled R$1.00 billion of migrated volume that is R$1.00 million a year of funding.
What it buys. Acceptance: the estate, the integration work, the shared dispute desk, a small field force whose job is to sign up the next thousand merchants, and the fraud engine that the cheap rail has to fund from somewhere.
Term and sunset. Five years, with the levy stepping down to zero if penetration targets are met early. Write the sunset into the constituting document on day one. A levy without a sunset becomes interchange within a decade, and it will be defended by people who were hired using it.
Governance. One member, one vote, not one vote per unit of volume. The large merchants have the volume; the small ones have the acceptance problem, and the mutual exists to solve theirs.

The balance-sheet treatment. The levy is an operating cost within cost of sales, directly offsetting the merchant discount rate line it replaces — which means it is margin-accretive from month one and needs no capital approval. The mutual's own capitalised estate sits on the mutual's balance sheet, not the members'. Members hold a membership interest at nominal value. Ask your auditors about it early and ask them about it as a purchasing cooperative, because that is what it is and they have seen one before.

The counterparty. Start with your trade association, which already has the membership list, the legal personality and the collection mechanism. Second choice is your acquirer, who will do it because the alternative is losing the volume entirely. Do not start with the central bank; approach it in year two, with two years of penetration data, because then you are showing a mechanism rather than asking for one.

The arithmetic that decides it. One inequality, on the front page:

   ( MDR_old  −  MDR_new  −  levy )  ×  migrated volume   >   0

   ( 2.2 %    −   0.22 %  −  0.10 % )  =  1.88 % of migrated volume

On R$1.00 billion of pooled volume that is R$18.80 million a year retained by merchants rather than paid away, against R$1.00 million of levy — a coverage ratio of 19.8 times. Over the five-year term: R$5.0 million of levy against R$99.0 million of fees avoided.

And the number that decides it for one member, which is the one to put in the board paper. Not the fee saving. The fee saving as a share of net profit:

      ( MDR_old − MDR_new ) × revenue × share migrated
      ------------------------------------------------
                       net profit

A merchant on R$1,000,000 of card-borne revenue at a 4.0 per cent net margin earns R$40,000. Moving the whole of it from a 2.2 per cent card fee to a 0.22 per cent rail saves R$19,800 — 49.5 per cent of net profit. At half migration, 24.8 per cent. At a 3 per cent margin and full migration, 66.0 per cent. And the threshold worth carrying: at a net margin below 1.98 per cent, the payment fee is larger than the entire profit of the business. That is not a rhetorical flourish. It is a division, and for a great many food retailers it is the correct side of the line.

The first ninety days.

Day	Action	Artifact
1–15	Pull twelve months of merchant statements; separate interchange, scheme fees and acquirer margin	The three-line fee decomposition
16–30	Compute fee saving as a share of net profit at three migration rates	The one ratio, on one page
31–45	Convene four peer merchants; size the pooled volume	Pooled volume estimate
46–60	Draft the mutual: levy, sunset, one-member-one-vote	The constituting document
61–75	Sign the first cohort; instrument the migration rate weekly	Migration dashboard
76–90	First verified quarter; take the ratio to the board	The board paper, one number

APPRECIATIVE QUESTIONS

Twelve, for a room

Discovery — what is already working

Where in this business does a customer already pay us by a route that costs us almost nothing — and what made that route possible?
Who here has successfully moved a customer from an expensive payment method to a cheaper one, and what did they actually say to the customer?
Which of our payment costs have fallen in the last five years without us negotiating them, and what caused that fall?

Dream — what becomes possible

If acceptance cost us a flat fee per transaction instead of a share of the sale, what would we price differently tomorrow?
Imagine the payments line on our P&L shrinks to a tenth of what it is. Where would we want that money to land — price, margin, wages, or the estate?
If every one of our suppliers were paid the instant we authorised it, what would change about what they will do for us?

Design — what we build

What is the smallest group of peer merchants we could convene who share our acceptance problem, and who would chair it?
What would have to be true for us to offer a customer a genuinely better experience on the cheap rail than on the expensive one?
Where in our own contracts are we charging a percentage for something whose cost does not scale with value — and what would a flat fee do to that relationship?

Destiny — how it holds

What would keep a levy we created from becoming the very thing we built it to replace?
Who outside this business would notice first if our migration stalled, and how would we make sure they told us?
If everyone in this room moved on, what would keep the cheap rail on the counter — a habit, a contract, or a person?

WORKS CITED

Banco Central do Brasil. Pix: estatísticas e tarifas do Sistema de Pagamentos Instantâneos (SPI). Successive releases, including the SPI tariff schedule and the retail payments and cards statistics series.

Banco Central do Brasil. Estatísticas de Pagamentos de Varejo e de Cartões no Brasil. Successive editions.

Bank for International Settlements (2021). Central banks, the monetary system and public payment infrastructures. BIS Bulletin No 52.

Bank for International Settlements (2025). Competition in retail digital payments. BIS Bulletin No 127.

Bank for International Settlements (2025). Pricing in fast payments. BIS Working Papers No 1295.

Bank for International Settlements (2024). Faster digital payments: global and regional perspectives. BIS Papers No 152.

Cambridge Centre for Alternative Finance (2025). Cambridge Digital Mining Industry Report; and the Cambridge Bitcoin Electricity Consumption Index methodology.

Digiconomist. Bitcoin Energy Consumption Index. Continuously updated.

Efficiency Valuation Organization. International Performance Measurement and Verification Protocol (IPMVP), Core Concepts. Successive editions.

Ethereum Foundation. Ethereum Energy Consumption, ethereum.org; and the consensus specification (twelve-second slots, thirty-two-slot epochs, finality at two epochs).

European Union. Regulation (EU) 2015/751 on interchange fees for card-based payment transactions.

Government of India, Press Information Bureau. Incentive Scheme for Promotion of Low-Value BHIM-UPI Transactions (P2M). Successive Cabinet releases.

National Payments Corporation of India. UPI Product Statistics; and the circular introducing a merchant discount rate on person-to-merchant transactions above ₹2,000 with effect from 15 October 2026.

Nilson Report. Merchant Processing Fees in the United States, 2024 and 2025 editions.

Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press.

Pay.UK. Annual Report and Financial Statements 2025; and Annual Summary of Payment Statistics 2025.

Reserve Bank of India. Annual Report 2024–25, payment system statistics.

Rochet, J.-C. and Tirole, J. (2006). "Two-Sided Markets: A Progress Report." RAND Journal of Economics, 37(3), 645–667.

Schuh, S., Shy, O. and Stavins, J. (2010). Who Gains and Who Loses from Credit Card Payments? Theory and Calibrations. Federal Reserve Bank of Boston Public Policy Discussion Paper No. 10-3.

UK Finance (2025). UK Payment Markets 2025.

United States, Board of Governors of the Federal Reserve System. Regulation II (Debit Card Interchange Fees and Routing), 12 CFR Part 235.

Note on figures. Every number in this chapter is computed in lib/verify/III_09.py and printed there with its inputs, its units and its source. Exchange rates, the bitcoin price, and the merchant margin used in the worked model are assumptions, and each is printed as an assumption. The per-transaction energy figure for proof-of-work is an average rather than a marginal cost, and the security-budget figure is given beside it for that reason.