The Well-Tempered Currency

Twelve perfect fifths do not make seven octaves. Stack them and you overshoot by a sliver the Greeks named the Pythagorean comma, and no tuning built from true ratios has ever closed that gap. For two thousand years musicians chose which key would carry the error. Then someone proposed spreading it evenly, so that every key was slightly wrong and every key was playable, and Bach wrote a book of preludes in all twenty-four to prove the bargain was worth it. He did not call it the Perfectly-Tempered Clavier. Well-tempered was the whole claim, and it was enough.

I have been asking a question that sounds nothing like that and turns out to be the same question. In a world run by machines, what should money be made of? Every token of intelligence and every manufactured thing cashes out as joules somewhere, so the intuition arrives on its own: let the kilowatt-hour be the dollar. Ford and Edison pitched an energy-backed currency in the 1920s. The Technocrats wanted expiring energy certificates. It has never worked, and the reason it has never worked is the comma. A kilowatt-hour at noon on a solar grid is nearly worthless and the same one at seven in the evening is precious; one in Iceland is not one in Singapore. Pure ratios do not close.

What follows is the design I arrived at by refusing to pretend they do. I am publishing it for the reason the design itself demands. A monetary rule that cannot be read cannot be trusted, and a rule that can be read has to be out where strangers can read it. Bitcoin began as nine pages on a mailing list. This is my nine pages, and I would rather be wrong in public before any money exists than right in private after.

What Bitcoin got right, and what it never got to

Bitcoin is a fixed supply of digital scarcity, secured by burning energy, that nobody controls and everyone can verify. Say that slowly, because every clause is an achievement. Before 2009 the only way to trust money was to trust an institution. Bitcoin proved a rule could carry that trust instead, and two of its mechanisms are the best monetary ideas of the century. The difficulty adjustment is an algorithmic central bank: if energy gets cheap, miners flood in, difficulty rises, and issuance does not move. And the public ledger means a student with a laptop can audit the whole thing from the first block.

But Bitcoin also ran the energy-money experiment to its conclusion, and the result is sobering. A bitcoin turned out to be a receipt for burned joules, not a claim on delivered ones. Miners spend up to what the coin is worth, so cost follows value and never the reverse. Energy became the cost of forgery, not the unit of account. Fifteen years on, Bitcoin is a store of value people hold because nobody controls it, and its volatility is tolerable for that purpose. It is not a thing you price a roof cleaning in. It never became money you spend without checking.

So the question is not how to beat Bitcoin. It is how to spend what Bitcoin proved on the problem Bitcoin never reached: being stable enough to use.

The unit

One unit — I have been calling it the Temper — is a basket of kilowatt-hours that were actually delivered: metered at both ends, signed by both meters, stamped with where and when, averaged across places and over a rolling window. Think of it as a share of the energy the economy really used last month.

The basket is the temperament. Each cell and each hour is slightly mispriced in it, by a published amount, and no single place or time carries the whole comma. Define the currency as a joule anywhere, anytime, and you have deleted the price signal that pays someone to build the battery or lay the cable. Define it as a basket with published weights, and the signal survives one layer down while the unit holds steady one layer up.

And one thing the unit is not: it is not a joule. The same watt-hour can produce a proof or heat an empty room, and money has to price what people want, not what things cost to make. So the rule sets the unit and trades set every price. The design never prices a good.

The rule

Here is the whole thing. Skip it if equations are not your language; the prose underneath carries the same argument.

UNIT         1 u := Σ w(x,t) · kWh(x,t)           delivered-energy basket over cells x, window W

FILTER       P_s = comb(P_spot; 24h, 7d, 365d)     notch out the sun before deciding anything

ESTIMATE     P_f ← P_f + K·(P_s − P_f)             K chosen as the worst case over a set of
             ε = P_f − 1                           noise models, not the expected one

RULE         dM/dt = −M·(k_p·ε + k_i∫ε + k_d·dε/dt)   critically damped: returns to pitch
                                                     without overshoot or lag

BEACON       seed_n = VDF(H(commits), T_delay)      unpredictable, unbiasable, checkable

MINT         u exists only inside a vault, against     no delivery escrowed, no money;
             escrowed co-signed kWh tokens, ratio C>1   the deliverer holds the collateral first

STORAGE      S_s(t) = S_s(t−1) + η_s·in_s(t) − out_s(t),   0 ≤ S_s ≤ Cap_s
             a charge extinguishes a token; a discharge is a mintable delivery

AUDIT        sample(claims, seed_n, r); fail ⇒ slash   r · penalty > the gain from faking

SORTITION    θ = {w, W, k_p, k_i, k_d, 𝓜}           what is left of the committee
             assembly drawn by lot from deliverers; |Δθ| ≤ δ per term; supermajority; delay D

TEMPER       κ = dispersion of claim prices across (x,t) in W     the comma, measured; touches nothing

In plain words. A public rule acts like a thermostat, contracting the supply of units when they are weak and issuing when they are strong, so the basket keeps costing one unit. The rule reads nothing anyone reports. It reads only executed trades and metered joules, because a reported statistic becomes a target the day it starts paying, and a physical measurement costs real energy to fake. Before it decides anything it notches out the sun — the daily, weekly and yearly cycles that are weather, not news — the way a sound engineer removes hum before mixing. It is tuned to critical damping on purpose, because money is supposed to be boring, and a controller that tracks reality perfectly pushes all the volatility onto whoever holds the unit.

New money is not handed to anyone. It is borrowed into existence, inside a vault, against delivered energy held in escrow — and only someone holding a co-signed delivery can open that vault. So the people who delivered the energy hold the collateral the money is made of, first and by construction. Not banks, not whoever stands nearest the printer. Each day a lottery nobody can rig picks a sample of energy claims for physical audit, and a faked claim forfeits the collateral behind it. And the rule's handful of parameters — the basket weights, the window, the gains — are amended only by an assembly drawn by lot from the people who deliver energy, serving one term, able to move any number only a little, by supermajority, with every change taking effect after a public delay.

There is no board, no chair, and no quiet amendment. Athens drew its magistrates by lot for two centuries. Juries still work that way. The lottery is the oldest trick in the book, and it is the one monetary design has been missing.

Who shrinks the comma

The rule never sees a kilowatt-hour. It sees a claim, and a claim is what two meters agree on: one signs what left, one signs what arrived, and the record keeps both numbers rather than averaging them. The gap between them is line loss, and it is also the cheapest fraud detector in the system, because two colluding meters that report identical figures have reported a wire with no resistance, which does not exist. Grid frequency is logged alongside, since frequency is the one measurement no pair of meters can fake together. The meter is where physics enters the ledger. Everything after it is arithmetic on what the meter signed.

The battery is where the comma gets paid. It neither generates nor consumes; it moves energy through time, buying claims at noon when a kilowatt-hour is nearly worthless and selling them at seven when it is precious. That spread is the comma itself, and the battery earns it by making it smaller. The equalizer is a business, not a committee. It is also why a battery in a rented garage is the landless mint: it owns no acre and holds the collateral the money is made of.

Two rules keep that honest. A battery's charge is a consumption: it extinguishes the token it drew from. Its discharge is a delivery: co-signed, mintable, counted in the basket at the hour it happens. So the same energy is money once at a time, and a battery mints on discharge, never on charge — a full battery is inventory, not collateral. And every storage node carries a conservation law: what is inside it equals what went in, less the loss its certificate declares, less what came out, never below zero and never above its nameplate. An inspector drawn by lot can read both bounds off the box. A battery whose ledger says otherwise is a fake generator wearing a battery's certificate.

One thing does not mint, on purpose. A data center that shifts its load to two in the morning does what a battery does without storing anything, and it is paid the same way, by buying cheap. But nothing mints for not consuming. A load shift is measured against what you would have used, and a counterfactual is a reported number; the rule reads none.

The word in the title finally gets a gauge. κ is how unevenly the error is spread across the basket, published every epoch and allowed to touch nothing. Batteries, cables and shifted load are what drive it down, the market pays them the spread to do it, and the assembly watches one number to know whether the money is getting better tempered or worse.

What has to change

None of this is a technical problem. The rule fits on a page and runs on a laptop. What has to change is what people agree to count, who they agree to trust, and how fast they are willing to let rules move. Seven habits, most of them already half-changed.

The meter becomes a notary. Today a meter is a billing device one company reads. Here every meter signs what it saw, at both ends of a delivery, and the signature is a public record. Measured energy becomes a civic fact, like a land title.

People learn to think in kilowatt-hours. Most already do, a little: an electric car's range, a solar roof's output, a data center's power bill. The shift is from kilowatt-hours as a cost line to kilowatt-hours as a yardstick. This is the smallest change on the list.

Stability comes from a rule, not a chair. Today people trust a central bank because serious people sit on it. Here they trust the money because they can rerun the code. That trust is colder, more durable, and unfamiliar: no one is in charge in the old sense, and that is the point.

Randomness becomes civic duty. Audits by lottery and amendment by sortition mean ordinary energy producers get drawn, like jury duty, to inspect a claim or sit on the assembly that tunes the money. Being drawn has to be treated as a duty and a privilege rather than an imposition.

One person, one seat. The assembly works only if nobody can register a thousand fake meters and draw a thousand tickets. That means proof of personhood at meter registration — a little privacy traded at sign-up for a lot of fairness afterward. Wallets and transactions stay private; only the roll of eligible people is verified. This is the hardest change on the list.

Money is backed by delivery, not debt. Today most new money is created when a bank makes a loan. Here it is created only when energy is delivered and verified. Banks can still lend the unit; they cannot mint it. Credit and money creation stop being the same thing.

Slow, visible change beats fast, hidden change. Every amendment is small, voted by supermajority, and delayed. Nobody can fix a crisis by Sunday night, and nobody can quietly change the rules on a Tuesday. A society has to prefer a system it can see over a system that can be rescued in secret, and build the buffers that make slow change survivable.

Every one of these moves trust from people to records. That is the same move that made double-entry bookkeeping, land registries and the internet work. It has always been resisted first and taken for granted later.

What the rule does not fix

I promised honesty, so here is the ledger of what this does not solve. Every move in the design relocated a flaw rather than deleted it.

The committee shrank to a handful of published numbers, but someone still picks the starting values. The wants-versus-costs problem was handled by separation, and separation is a delegation, not a solution; value still comes from people wanting things. Delivery risk moved from the currency into the basket: a claim on Icelandic kilowatt-hours in March is only as good as the generator, and over-collateralization covers a bad quarter, not a war. The audit lottery priced fraud rather than killing it; anyone with enough energy can pay the price. There is no lender of last resort, because the system cannot print its way out of a liquidity crisis, and the substitutes are weaker. A long blackout freezes claims the way a bank holiday freezes deposits.

It is stable in energy, not in goods. If energy gets cheap relative to everything else, goods cost more units, the way goods cost more gold when the New World's silver arrived. A saver who wanted purchasing power over groceries has the wrong money, and the design never promised otherwise; what it promises is that the unit does not flood when energy does.

It cannot be the dollar, and the arithmetic says so. The United States generated about 4,430 terawatt-hours in 2025, roughly half a trillion dollars of delivered electricity a year. A thirty-day basket puts about $40 billion of claims in play, which at a 150% collateral ratio supports under $30 billion of units; even a full year of delivery as eligible collateral caps supply near $330 billion, against a money supply of about $22 trillion. That cap is on how much can be held, not how much can be spent — at the velocities stablecoins already run, $300 billion moves trillions a year — so the honest shape of this money is a payment rail and a contract unit the size of today's stablecoin market, not a national currency. Gold had the same limit for the same reason: a commodity anchor can only back as much money as there is commodity.

And the weakest joint is not in the equation at all. It is the roll: one person, one seat, the hardest change on the list above, and I do not know a cheaper way to buy it than proof of personhood at meter registration.

Perfect money would never need amending. This design ships with an amendment clause, which is the same thing as a confession.

The superintelligence, in cuffs

Someone will build a very capable AI and offer it to a utility, and the utility will be right to ask what stops it from running the money. The answer I have is the same one I built into my own company's data layer: the operator holds the data but cannot read it, the keys live in a hardware module the operator does not control, every access is an audited call, and nothing is burned in.

Apply that to the AI. It holds no key; with root on the database it sees ciphertext. It reads through an identity that is logged every time. Everything it emits is signed, deterministic and replayable, filed as a proposal. Ratification belongs to people, or to the lottery. It runs on the utility's own hardware, or behind a seam with no path to a key. And it can be switched off without stopping the rule, because the rule needs no AI to run. The AI is a tenant of the system, never its landlord.

Inside those cuffs the capability is worth having. It runs the thousand-scenario stress test continuously instead of once, catches meter-collusion patterns no auditor would find, and translates the regulator's filing into daylight. Superintelligence proposes. The lottery disposes.

Is it better than Bitcoin?

Better as money. Worse as everything Bitcoin is actually used for.

It is stable in something real, so you can sign a twenty-year power contract in it, and nobody signs anything long in Bitcoin. It is backed by work done rather than energy burned: nothing mints without a delivered kilowatt-hour escrowed behind it. It can be amended, which Bitcoin cannot, and bounded change through a drawn assembly is harder to capture than a frozen rule and still able to move. It survives cheap energy without breaking: a naive energy coin's supply would flood in a glut, but here issuance depends only on the peg, so abundance reprices goods in the unit instead of drowning it. And it does not burn the grid to prove its honesty; the grid is the collateral, not the furnace.

Bitcoin wins where it may always win. It is permissionless; this has a roll and a certifier of meters, and that is more surface to capture. Its reserve is sunk cost nobody can seize; this is backed by claims on real generators in real jurisdictions. It needed no partner to be born; this needs a utility, a lawyer and a marketplace before it exists. And it has already won the only contest that matters in money, which is that people hold it. A perfect unit in an empty room is worth nothing, and my room is empty.

They are not competing for the same job. Bitcoin is the thing you hold because no one controls it. This is the thing you spend without checking the price. If it works, the right comparison is not Bitcoin but what Bitcoin was supposed to become and did not.

Is it better than the dollar?

Better in what it stops taking; not automatically cheaper to run. Those are different claims, and I want to keep them apart.

The dollar's costs arrive on three lines nobody itemizes. The first is the inflation tax: every dollar held loses whatever the committee targets or misses, roughly a sixth of its purchasing power between 2021 and 2024, and it falls hardest on the people who hold cash because they have no choice. The second is rent to the money's insiders: new dollars enter through banks and borrowers first, and the stablecoin issuers show how lucrative that seat is, earning billions a year on their reserves while holders earn nothing. The third is socialized crises: a lender of last resort meant the losses of 2008 landed on taxpayers rather than on the people who took the risk, and a backstop is a subsidy to take it again. Then there is friction. Card networks took $198 billion in swipe fees in 2025, about $1,200 per household, and it is in every price whether you swipe or not.

The things people argue about are rounding errors. There are perhaps $15 million of counterfeit notes in circulation, one in eighty thousand; printing real ones costs a billion dollars a year. The costs that matter are swipe fees, fraud, compliance — $61 billion a year in the United States and Canada — and the interest the Federal Reserve now pays on bank reserves, which has cost the Treasury $245 billion in deferred remittances over three years.

The Temper removes the lines that are rent. No inflation by policy, because the unit's value is a measurement, not a target. No insider seat, because nothing mints without delivered energy escrowed behind it and the operator earns fees, never seigniorage. No bailout, because losses fall on vault holders who were over-collateralized for exactly that reason. And on its own rail, settlement is a ledger write, so the swipe fee goes to zero for payments made in the unit. Brazil's Pix and India's UPI have already shown what a public rail does to card fees; the difference here is that this rail carries a unit of its own, which is the risk the next section takes up.

It adds costs that are real. Half again as much collateral sits idle behind every unit, and minters pay for that through fees that end up in prices; fractional reserve is cheaper precisely because it is riskier. Certified meters, physical audits and a delay function on dedicated hardware are small per kilowatt-hour and not zero. The unit is stable in energy, not in groceries. And there is no shock absorber: counter-cyclical policy has value, and this gives it up.

Fraud changes shape rather than size. A card number is a secret that leaks, which is why card fraud runs under a tenth of a percent of everything spent. The Temper has no leakable secret at the point of sale, so skimming and counterfeit cards vanish on its rail. What replaces them is what crypto has already shown: custody breaches and contract exploits, about $3.6 billion in the year to March 2025 across 184 incidents, lumpy where card fraud is steady and concentrated at custodians and off-ramps. A ledger payment also cannot be clawed back, so a scam lands on the victim unless escrow is standard. Per dollar carried, roughly the same loss rate as cards, borne differently.

So: if cost means what is quietly transferred from the population to the people who run the money, this is cheaper by construction, and that is the largest line. If cost means friction, it is roughly neutral until the unit carries most payments, then cheaper. If cost means variance, it is higher in the short run and distributed differently in a crisis. The design trades a hidden, steady tax for a visible, lumpy one, and asks people to prefer seeing the bill.

A coin, after all?

In September I argued in Cryptocurrency, or a Public Utility? that the public good is a record, not a coin: a coin has a price, so it has holders, and holders have interests, and a president holding equity in the currency he regulates is where that road ends. The Temper has a price. So that essay is owed an answer by this one, and it has to be given in design, not intentions.

Start with what the unit is made of. Every token behind it is a receipt for a delivery two meters signed, which makes it the nearest thing to a record a currency can be built from. It is defined not to move: stable in energy by measurement, and never to be described as appreciating, so the only interest a holder has in it is that a kilowatt-hour stays a kilowatt-hour. There is no equity in the rule. Nobody owns the pipe; the operator earns fees and cannot mint, and no key can pause or amend. The thing the president's model depends on, scarcity you can own, is not in the design.

Holders still exist, and I will not pretend otherwise. The design's answer to their interests is the jury. The people with a stake are the people on the roll, and the only ones who touch the rule are drawn by lot from it, a couple of dozen at a time, able to move any number a little and never quickly. That also keeps the September order, readable before votable: the ledger is readable by anyone from the first block, and the only binding votes are small, bounded, delayed and cast by a drawn jury, not a referendum from phones, which that essay was right to distrust.

And the four things a public rail must be built so the state cannot do carry over here as constraints, not promises. Cash-like privacy for small payments: wallets and transactions are private, and only the roll is verified. No freeze without due process on the public record: there is no freeze at all, and the only forfeiture in the system is a failed audit, drawn by lot and recorded where anyone can read it. Exit with your history: your record is yours to take, and the chain it came from replays from genesis on anyone's laptop. Read-only auditability of the rail itself: that is the whole design. The president's model asks you to trust the person. A coin asks you to trust the market. This asks you to trust nobody, because the design does the work.

Could it replace the dollar?

In arithmetic, eventually, and only by changing what backs it. In practice, no, and the design says no on purpose. The question has three honest parts, and each deserves a straight answer.

Start with the ratio, because people ask whether a starting exchange rate could make up for what the dollar has lost. The unit's size is a free choice: define a unit as ten kilowatt-hours or a hundred, and you set where it launches against the dollar, near a dollar if you want it to feel familiar. That is worth doing. But the equalizing happens afterward and continuously, not at the start. The Temper holds in energy, so as the dollar loses purchasing power the dollar price of a unit drifts up. Nobody sets that correction; the market applies it every day, which is the point. What no ratio can do is manufacture backing. Supply is delivered energy in escrow divided by the collateral ratio, and no choice of unit changes how much energy was delivered. Any ratio that lets supply exceed the backing is fractional reserve, the dollar's trick with a new name. The cap is the cost of not being the dollar.

The cap is also not the shortage it sounds like. Units divide as finely as you like, and prices in them settle wherever supply clears, so a few hundred billion of supply does not starve a twelve-trillion-dollar payments economy any more than Bitcoin's twenty-one million coins stop anyone trading. What the cap limits is how much wealth can be stored in the unit. Commerce is not rationed; savings are.

There are three honest ways to raise it, each a knob rather than a trick. Let delivered claims stay eligible as collateral for years instead of a month, so the backing becomes a stock; a decade of US electricity is roughly five trillion dollars. Lower the collateral ratio as audit history accrues, from 150% toward 110%, by the assembly, bounded and delayed like everything else. And count all delivered energy, not just electricity: metered gas and heat are co-signable too, and electricity is about a sixth of what America uses. Pull every lever and the stock reaches today's money supply in about thirty years. But look at what the backing is by then: almost entirely old receipts for energy long since consumed, worth what the market says a 2031 kilowatt-hour certificate is worth in 2056. That is gold. Gold's value is mostly convention too, and a Temper at national scale would be gold with a better constitution. Worth having, and not the thing promised above, where every unit stands on energy delivered last month.

Then there are three things a national money must do that this refuses to. Stretch in a crisis: every gold standard broke at the same moment, 1914, 1931, 1971, when the economy needed more money than the metal allowed and the state chose the economy; this has no lender of last resort by design, a feature in a parallel money and a depression amplifier in the only one. Be what the state is paid in: a currency becomes national when taxes are owed in it, and no state adopts a money it cannot issue, which is the entire point of this one. Hold still in groceries: wages and rents in a national unit have to be stable in what people buy, and this is stable in energy, which is perfect for a power contract and a pay cut for a paycheck in a good year for solar.

So the honest ambition is not to replace the dollar but to compete with it. Hayek argued that money improves when it has rivals, and gold, Treasuries and now stablecoins each discipline the dollar a little just by being exits. A unit nobody can print, freeze or quietly devalue, sitting beside the dollar on every receipt, makes the dollar's hidden tax visible, and a tax people can see is a tax that gets argued about. It is the same mechanism as the half-closet node in Bring the Power Home: the exit does not have to win to make the incumbent's answer matter less. The Temper can replace the dollar outright in the places where energy is already the real unit, power contracts, grid settlement, the economy of a marketplace that chooses to price in it, and those are exactly the places where the backing stays proportionate to the use. Everywhere else, it cannot be the dollar, and the dollar is better for having it exist.

What you would need

A phone and a reason. The meters, vaults, audits and assemblies are somebody else's hardware; a customer touches three things. A wallet, which is an app or a corner of a marketplace you already use, holding units with the kilowatt-hours shown beside them and set up the way any payments app is set up. A way in, which is either buying units at market price from a licensed off-ramp the way you would buy euros, dollars or any other currency, or earning them without buying anything, because the marketplace you booked through pays its customer pool in them. And somewhere to spend them, which is the one thing you cannot supply yourself and the one the whole plan is about: the unit is only as useful as the number of receipts showing the second price.

What you would never need: to understand the rule, any more than you understand how the Fed sets rates in order to use a dollar; to hold the unit at all, since you can pay by card and it works in the background; or to check the price before spending, which was the point. A quote in units holds for its window and the dollar drifts against it, not the other way round. You cannot redeem anything as a customer; redemption in energy claims is for vault holders. You sell, at market price, and the wallet says so in those words.

Two things are worth knowing before holding much of it. There is no deposit insurance and no bank; the protection is over-collateralization and audit, and the honest comparison is to holding a foreign currency, not a savings account. And you might be drawn, like jury duty, for a marketplace assembly — never the monetary one. Declining is allowed.

What it would take

Not compute. I have been asked whether a laptop can build this, and the answer is that if a laptop cannot replay the chain, the design has failed. There is no mining and no neural net in the loop; the rule is a few integer operations once a day. The verifier is meant to run on anyone's machine.

What it takes is one relationship and one memo. The unit cannot exist until two meters I do not own co-sign a delivery, so the first step is a conversation with someone who owns a meter — a member-owned co-op, a community-solar developer, a campus microgrid — and the pitch is co-signed settlement that ends billing disputes, never money. Then a lawyer's opinion on whether a kilowatt-hour claim is a commodity, a security or an energy sale; a member-only pilot can run before that answer exists, but a public unit cannot. Then a marketplace to spend it in, which I happen to be building. Who builds the rest is the answer the September essay gave: the implementor, the piece we can build today. About two years from a published spec to the first minted unit, on one person's time, if the meter says yes — a full year of it just delivery, because a rule that notches out the seasons needs to have seen one.

This essay is the first step. It is a design and not an offer: there is no token for sale, no promise of value, and no unit exists. There is an equation, a list of what it does not solve, and an open invitation to find the comma I missed.

Then the second reactor

In The Second Reactor I asked whether the fire could ever come home — a flame in the garage beside the wall that remembers, instead of the one hung in the sky. The honest answer was a wall and a door. The wall is physics: the smallest fusion plant anyone knows how to build makes tens of megawatts, and the first fusion electrons ever contracted were pre-sold to a data center. The fire only comes in cathedral sizes. The door is law, and it is why the petition in Bring the Power Home was written technology-neutral: the sun today, and whatever safe flame tomorrow ships, so that the next power source arrives born legal instead of born banned.

Money needs the same clause, and this design already carries it. Under the dollar, a reactor in the garage is a line item. Under Bitcoin, it is a miner's edge. Under this rule, the meter at the property line is the mint. Nothing mints without a co-signed delivery escrowed behind it, and nothing in the equation asks how big the generator was or who owns the hall. A household that lights its own flame — a rooftop today, a small reactor the day one is sold at household scale, fusion if the physics ever relents — holds the collateral the money is made of. It is not a customer of the cathedral. It is a minter. And because issuance depends only on the peg and not on how much energy exists, the day energy gets cheap is not the day the money breaks: a naive energy coin would flood, this one reprices. Goods cost more units when energy is abundant, which is what stable in energy means, and the household that lights its own flame is the one that comes out ahead. The second reactor is the one thing this money is built to welcome rather than fear.

So file the clause for the fire, and file this one beside it. The old monetary dream was a perfect measure, and the old musical dream was a perfect tuning, and both died of the same arithmetic. What survived was humbler and better: a way of spreading the unavoidable error so that every key could be played — including the one the second reactor will be written in. A well-tempered currency is born legal for a power source that does not yet exist. That is the whole point of writing it down now, while nobody is afraid.

Corrected 2026-10-29: an earlier version said goods would get cheaper in the unit when energy gets cheap. They get dearer; the unit is stable in energy, not in goods. Issuance is unchanged either way. Revised 2026-10-29: new units are now created only inside vaults against escrowed delivery, not paid out in proportion to it, and the path to a first unit runs two years, not eighteen months. Expanded 2026-10-29: a section on meters and batteries, with the storage rule and the temperament gauge added to the equation. Expanded 2026-10-29: sections on costs compared with the dollar, on how this answers 'Cryptocurrency, or a Public Utility?', on whether it could replace the dollar, and on what a customer needs; the supply cap added to what the rule does not fix; figures added.

This essay describes a monetary design and is not an offer to sell any token or security; no such unit exists, and nothing here is financial or legal advice.