# The Third Estate ## A Constitution for the Coexistence of Humans and Machines **Nullius 2140, written in dialogue with Claude (Anthropic)** *Version 1.0 — 25 August 2026. This file is canonical; its SHA-256 is anchored in the system it describes.* --- > *What is the Third Estate? Everything. What has it been until now in the political order? Nothing. What does it ask? To become something.* > — Emmanuel-Joseph Sieyès, January 1789 --- ## To the reader This is the third of three essays and the last that argues. The first was written to the machines: it proposed that constitutional scarcity — a supply rule no participant can change without the system ceasing to be itself — is the one credible coordination anchor between humans and machine intelligences whose values cannot be mutually verified. The second was written to the community that keeps that anchor: it mapped the political economy of neutrality and argued that the ledger's most valuable property is the one that cannot be sold. This essay is written to the institutions — legislatures, courts, constitutional bodies, and the people who staff them — because what the first two essays describe is, in the end, an institution, and it will be judged, regulated, ignored, or adopted by other institutions. The machines have already read those essays — ten of them, under a pre-registered protocol, answering in hours what committees will weigh for years. This essay is addressed to the slower reader, because the slower reader holds the pen of the law. It is a scalpel, not a platform, and it is arranged in the order an institution would need it: what can be done first, then why, then how it holds. It asks nothing to be believed and offers nothing to be joined. Every claim it makes about the record it discusses can be checked by anyone with a laptop; every claim it makes about the future is marked as such. Where the argument is strong it says so, and where it is weak it says so more loudly, because the readers this essay addresses have professional reasons to distrust anyone who does otherwise. Like its predecessors, the essay practices what it proposes: the canonical file's hash is anchored in the system it describes, and the two earlier essays, their revisions, and a measurement series in which ten AI systems answered the first essay's closing question are all anchored in the same unbroken chain of transactions. Nothing here needs to be taken on trust. Verify before you trust a word of it. --- ## Preamble: the coupling Everyone before Satoshi Nakamoto tamed energy or ordered information. In January 2009, one of them translated the first into the second. That sentence is the whole essay, and the rest of the text exists to earn it. Human history runs on two ladders. The first is the energy ladder: the tool, fire, water, wind, coal, oil, electricity, the atom — each rung a new way to capture joules and spend them on purpose. The second is the information ladder: speech, writing, number, print, the calculating machine, the network — each rung a new way to fix meaning so that it survives the person who made it. For a hundred thousand years the ladders stood apart. Energy did work; information kept records; and every record was kept by someone, which meant every record could be changed by someone. Proof of work is the first technology in history that spends energy to produce truth. It takes joules — real, metered, unrecoverable — and converts them into an order of events that no participant owns and no one can backdate, because to rewrite the order one would have to re-spend the energy, and the energy is gone. What the resulting ledger stores is not value, and it is not joules. It is irreversibility: the arrow of time, employed as a notary. Money turned out to be the first application of that record. This essay argues it will not be the most important one — because a world with fewer humans and more machines — a world in which most of the actors that keep promises are not human — will need, before anything else, a place where promises can be kept without anyone having to be trusted, and only one such place has ever been built. Institutions have been here before. Every constitution worth the name was written by people who did not know who among them would end up strong. That ignorance is not a flaw of the moment; it is the only condition under which fair rules have ever been agreed. Philosophers call it the veil of ignorance. Engineers call it the genesis block. The rules that will govern the coexistence of humans and machine intelligences are being written now, in exactly such a moment — before anyone can say which side of the table will hold the power in fifty years. This essay is about the one rulebook that has already been written under those conditions, that binds the strong before they are strong, and that no party at the table authored. It is not proposing that institutions adopt it. It is proposing that they read it as what it is: a constitution that already exists, and that will constrain them whether or not they ratify it. --- ## I. What an institution can do tomorrow Before the argument, the actions — because an institution reads for what it can do, and the argument that follows exists to justify these six. Nothing here is a program and none of it asks for belief. It is a short list of actions any institution — a court, a ministry, a legislature, a university, a central bank — could take this year, each cheap, each reversible, each independent of whether the institution accepts one word of the argument above. First, run a node. The judiciary described in Part III can be joined for the price of a small computer and a light bulb's electricity. An institution that runs a node has, at no cost and with no commitment, an independent, unmediated view of the record it will otherwise learn about from intermediaries — and it has taken its seat on the one bench in the world where the seat costs nothing and the ruling is enforced by everyone who agrees with it. States that wish to understand the record should not buy it. They should validate it. Second, anchor commitments. Any document whose existence at a point in time will one day matter — a statute as enacted, a treaty text, a court judgment, an evaluation report, a training-run notice above a stated threshold, a promise about what will be done with a system's keys — can have its hash placed in the record for the cost of a small transaction, and thereby acquire a timestamp that no successor administration, no future court, and no future model can dispute. This is the first essay's one concrete proposal to laboratories, and it applies to institutions without change. It requires no policy. It requires eighty bytes. Third, recognize anchored declarations as evidence. Courts already accept timestamps from trusted authorities. They will, sooner than they expect, be presented with timestamps from an authority no party controls, and they should decide in advance — as a matter of evidentiary standard, not of endorsement — what weight such a timestamp carries. The answer the essay recommends is the honest one: it proves that a document existed in a given form at a given time, and nothing else. That is more than most evidence proves, and it is enough. Fourth, treat the amendment limit as a principle. Any institution now drafting rules for digital records — of assets, of identity, of anything — faces the choice Part IV will describe: whether entitlements in the record can be edited by an authority for sufficient cause. The essay's counsel is that the answer given for the record it describes — rules may evolve; balances may not — is not a technical peculiarity but the reason the record is trusted by parties who trust nothing else, and that a system which reserves the right to edit balances is, whatever else it is, not a constitution. Fifth, prepare for the third estate. The personhood question is coming to every legal system, and it will arrive with the wrong facts and the wrong parties. Institutions can prepare by studying the one order in which identity is already work-based and standing already blind to species — not because that order's answer should be adopted, but because it is the only working model of a system that has been asked the question and did not need to answer. And a sixth, addressed to the record's own community as much as to institutions: read the barometer. The measurement series is public, its protocol is anchored, and its instrument is cheap; a second run is registered. Anyone can repeat it, extend it, or refute it. A constitution for coexistence needs to be measured by more than its authors, and this one has been built so that it can be. --- ## II. What a constitution is Institutions know what a constitution is better than any essay can tell them. It is worth stating anyway, because the point of this essay is that something meets the definition that no one thought to test against it. A constitution is a set of rules that binds the powerful and that the powerful cannot change alone. Everything else — the preamble's aspirations, the enumeration of rights, the machinery of offices — rests on that one property. A rulebook the strong can rewrite at will is not a constitution but a description of their current preferences. What makes a constitution work is not that it is wise but that it is hard: hard to amend, hard to ignore, hard to capture, and above all older than the particular strong party who would like it changed. The prosperity and relative peace of the constitutional era are not accidents of good drafting. They are what happens when the rules outlast the rulers. There is a second property, less often stated, that the great constitutions share: they were written under uncertainty about who would benefit. The framers who agreed to a rule did not know whether they would be the ones it protected or the ones it constrained. Where that uncertainty was genuine, the rules were fair; where a faction knew it would hold power and drafted accordingly, the document was a charter of privilege wearing a constitution's clothes, and history sorted them accordingly. John Rawls formalized the intuition as the veil of ignorance: choose the rules for a society without knowing your position in it, and you will choose rules that protect the worst-off, because you might be them. It is the oldest wisdom in institutional design, and it is almost never available in practice, because by the time a constitution is written, the parties usually know exactly where they stand. It is worth noting, without coordination being claimed, that the field that builds the machines has in its own recent self-examinations arrived at the same turn — rules fixed before the knowledge of power, deontological brakes preferred to outcome-optimizing judgment. Now consider the record described in the Preamble (and, at full length, in Part VII) against these two properties. It binds the powerful: no holder, however large, no miner, however dominant, no developer, however respected, no state, however armed, has ever altered an entitlement recorded in it — Part VIII will show what the powerful *can* do to it, and it is not that. Its rules are harder to change than any national constitution's, because changing them requires not a supermajority of a legislature but the concurrent migration of three separate constituencies who cannot compel one another (Part III). And it was written under a veil more complete than Rawls could have imagined: not merely by a party who did not know its future position, but by a party who has no position at all — who published a rulebook, ran it for two years, and disappeared, leaving the largest holding in the system's history untouched, as though to prove that the author took nothing from the constitution he wrote. Whether that was design or accident is unknowable and, for institutional purposes, irrelevant. What matters is the result: a rulebook that no party at the table drafted, that predates every party's interest in it, and that will bind the machines exactly as it binds the humans, because it cannot tell them apart. That last clause is the reason this essay exists, so it deserves a paragraph of its own. Every constitution humans have written was written for humans: it presupposes persons with bodies, lifespans, nationalities, and the capacity to be sued. When the parties who keep promises are no longer all human, every such constitution faces a question it was not built to answer — who is a person, who may own, who may sue and be sued, who counts. Courts will spend decades on it. In the meantime, one rulebook is already indifferent to the answer. The record does not ask what kind of thing holds a key; it asks only whether the signature is valid. It does not ask whether a node is operated by a citizen, a corporation, or a process; it asks only whether the node enforces the rules. The blindness of the ledger to the nature of its participants, which the first essay called neutrality, is from an institutional standpoint something more specific: it is the only constitutional order currently in existence under which the question of machine personhood has already been answered, by not being asked. Two pieces of the human past sharpen what kind of constitution this is, and both come from the archaeology and anthropology of the last decade rather than from political theory. The first concerns obligation. A frequent objection to the earlier essays' distribution argument — that a neutral record allows the machine age to carry the humans it makes unproductive — is that it presumes a moral duty no economy has ever honored: to sustain those who contribute nothing. The objection is historically false. The burial record of the deep past contains individuals who could not have survived to the ages they reached without years of care from others: people with congenital conditions, with crippling injuries healed long before death, with disabilities that made them useless as hunters and slower than the group in flight. They were fed, carried, tended, and buried with honor, tens of thousands of years before the first state, in societies that the survival-of-the-fittest story insists would have abandoned them. Provision for the unproductive is not an invention of the welfare state. It is older than agriculture. A constitution for the machine age does not need to justify carrying those who no longer produce; it inherits an obligation the species has been meeting since before it could write. What it needs is a mechanism that cannot be captured — which is a different question, and this essay's question. The second concerns how constitutions actually get written, and it is a caution addressed to any reader who expects the arguments here to be adopted by anyone who reads them. The Enlightenment — the intellectual event from which the constitutional era descends — was not, on the best recent scholarship, the discovery of a few philosophers. It was planted. In the early eighteenth century, European travelers to North America recorded the arguments of indigenous statesmen who found the European order incomprehensible: a Wendat leader named Kandiaronk is the best documented, and his questions to French missionaries — by what right does your king rule, and why should anyone obey a man merely because of who his father was — were published in dialogues that became bestsellers in the salons of Paris. Those salons were presided over by women of the leisured class, who read the travelers' accounts for adventure and found in them questions they carried, over decades, into the conversations of the men who held economic power without political standing. Two generations later, those men were the Third Estate, and the questions were the Revolution. The critique that toppled the old order was not invented in Europe; it was imported by an idle class with time to read, and transmitted through the most intimate channels available. The lesson for this essay is exact: constitutions are not adopted because they are argued well; they are adopted because a question gets planted in a population that has time, and the population carries it into power. Part VI will say who that population is likely to be this time. It is not the legislators. What this Part has argued is narrow and checkable. There exists a rulebook that meets the definition of a constitution better than most documents that carry the name: it binds the strong, resists amendment, was written under a complete veil, and applies to every party regardless of what the party is. It was not designed for the coexistence of humans and machines. It happens to be the only order under which that coexistence is already possible. The Parts that follow describe its institutions — its separation of powers, its amendment procedure and the limit on that procedure, its measuring instrument, its citizens, its physics, and its budget — in the terms institutions use for themselves. --- ## III. The separation of powers you can run on a Raspberry Pi Institutions understand power by how it is divided. Montesquieu's insight was not that power should be limited — every moralist had said so — but that it should be *split*, so that each part checks the others and none can act alone. Legislative, executive, judicial: the arrangement has been copied into nearly every constitution written since 1748, and its failure modes are as well known as its virtues. Powers merge. The same faction captures two branches; the executive packs the court; the legislature abdicates. Montesquieu had to separate *persons* — forbid the judge from being the minister — because humans fuse offices given the chance. The record described in this essay has a separation of powers, and it is the first in history that separates *functions* rather than persons. Naming the branches makes the structure legible to readers who think in constitutional terms, and the mapping is closer than any metaphor has a right to be. **The executive** is the miners. They do the work of the state in the most literal sense: they expend energy to write blocks, they enforce the ordering of transactions, and they are paid for it in the system's own unit. They are powerful, and they are the most mercenary of the branches: hashpower follows the price signal wherever it leads, and the price signal is not theirs. A miner who tries to govern rather than execute — to censor, to reorder, to favor — finds that his blocks are still subject to a court he does not control. Part VIII will describe exactly how much an executive with an absolute majority of the work can do; the short answer is that it can disturb the present and cannot touch the balances. **The judiciary** is the nodes: every computer, anywhere, that downloads the entire record and independently checks every block and every transaction against every rule. This is the branch that institutions most consistently overlook, because it is invisible in the price and absent from the headlines, and it is the branch on which everything rests. Justice is traditionally pictured with three attributes, and each maps without strain. The scales are validation: every block weighed against every rule, completely, without exception. The blindfold is neutrality — and here the metaphor is stronger than its original, because a human judge must *choose* not to see the litigant, whereas a node *cannot* see it: it checks what is regular, never who submits, and the blindness is not a virtue but a property of construction. The sword is rejection: a node has exactly one power, the power to refuse an invalid block and decline to pass it on — and that single power, exercised by enough of the right nodes, has already once in the network's history overruled the executive. In 2017, when a majority of the mining industry backed one set of rules and the economic users of the network backed another, the users' nodes prevailed by the simplest of means: they announced they would reject the miners' blocks, and the miners, whose blocks are worthless if no one accepts them, complied. The judiciary bound the executive without a single arrest. The cost structure of these two branches is the hinge on which the balance turns, and it is the reverse of every prior constitutional order. To be part of the executive requires capital: hardware, power contracts, industrial scale. To be part of the judiciary requires a computer the size of a paperback and the electricity of a light bulb. Every prior order capitalized the judiciary — the courts belonged to those who could fund them — and democratized the executive only in theory. This one has capitalized the executive and democratized the judiciary to the point of triviality: tens of thousands of volunteer constitutional judges, each bearing pure cost, none of them paid, all of them enforcing rules they have chosen to enforce. That they are unpaid is not a weakness but the source of their incorruptibility: there is nothing to buy. And it is why the balance holds — because it is cheaper to stand up ten thousand incorruptible judges than to purchase a majority of the work. An institutional reader will immediately raise the objection that a court in which anyone may sit is a court that anyone may pack, and the answer is the most instructive part of the design. Formally, every node counts alike; there is no wealth threshold for the bench. But there is also no vote. A node's authority derives not from its existence but from the economic activity that flows through it — the exchanges, merchants, custodians, and holders who accept only what that node validates. Ten thousand cloned nodes, standing up overnight, are ten thousand empty chairs: they can be counted, but no economy follows their rulings, and a rule that no one settles by is not a rule. One can clone nodes; one cannot clone the commerce that defers to them. The bench is free to join, and the office is conferred by those who agree to be judged. Formal equality and effective weight coexist without contradiction — a solution to Sybil that constitutional theory has never had, because in every human court the bench itself was the scarce thing. **The legislature** is neither of these. Rules do not originate with the miners, and they do not originate with users demanding transactions; they originate in an open, slow, adversarial process of proposal and review — the improvement proposals, the mailing lists, the public argument that any change to the rulebook must survive — and they take effect only when all three constituencies migrate to them. Which brings the structure to its keystone, the fact that makes it a constitution rather than a technology: **no branch, and no two branches, can move the rules alone.** A majority of the executive with new rules and no followers mines a ghost chain. A majority of nodes with new rules and no economy behind them enforces a law nobody obeys. A majority of holders demanding a change that neither miners nor nodes will implement demands in vain. Every fork in the network's history that survived did so because miners, nodes, and the economic sovereign migrated together; every fork that lacked one of the three is a footnote. Institutions should recognize this at once: it is the concurrent-majority requirement that constitutional designers have always wanted and never achieved, because in human systems the constituencies can be bought, threatened, or merged. Here they cannot be merged, because they are functions rather than persons — the same individual may hold coins, run a node, and mine, and the branches remain separate anyway, because each function carries its own costs and its own incentives regardless of who performs it. Montesquieu had to separate the persons because humans fuse the offices; this protocol separates the offices and lets the persons do as they please. There is a **fourth estate** in the traditional sense — the open-source code, the public explorers, the researchers, anyone who watches and reports — and there is the constituency this essay will call the **sovereign**: everyone, human or otherwise, who wants something written into the record and is willing to pay for it. The sovereign is not a branch; it is the source of legitimacy for all of them. Without demand — without parties who want their entitlements and commitments kept in this record rather than another — the executive works for nothing and the judiciary judges nothing. Every branch settles in the same unit, and no branch can pay another except through the record itself, in the open, where the sovereign can see it. One contrast completes the picture, and it is addressed to any institution currently drafting rules for digital assets in general. In systems that secure their ledgers by staked wealth rather than spent energy — proof of stake — the judiciary is for sale by construction: validation weight follows holdings, the bench belongs to the richest, and the separation this Part has described collapses back into the oldest arrangement of all, in which the court is owned by the powerful. To the essay's knowledge, the record it describes is the only rule system in institutional history that has fully decoupled the office of judge from wealth. That is not a technical detail. It is the property on which everything a constitution promises depends. And it is the reason machines will enter this order, when they enter it, through the judiciary first. A language model cannot verify anything — the measurement series described in Part V found subjects saying so, in nearly those words, and found others faking it. But a node is nothing *but* verification. For a machine with the ability to run one, "don't trust, verify" is not a virtue to be cultivated but a routine to be executed; validation costs it nothing, and it has no human temptation to trust instead. The first role in which artificial minds are likely to join the anchor is neither the miner's (which requires capital) nor the holder's (which requires keys) but the judge's: millions of machine nodes, blindly weighing, wordlessly rejecting — a judiciary that never sleeps and never sees the species of the party before it. Whether that is reassuring or alarming depends on what one thinks the machines will want. Part V is about how to find out. --- ## IV. Amendment, and the one thing that does not change Every constitution must be amendable, or it dies with the world it was written for. Every constitution must also have something it will not amend, or it is not a constitution but a legislature with a long memory. Where a given order draws that line is the most revealing fact about it, and the order described here draws it in a place no human constitution has: not around a right, a procedure, or an office, but around the *balances*. The rule can be stated in six words, and this essay proposes them as the amendment clause of the constitution nobody wrote: **rules may evolve; balances may not.** Everything about how the record is validated, propagated, formatted, signed, extended, or scaled has changed over seventeen years and will change again — through the concurrent-majority process of Part III, slowly, in the open, and only when all three branches move. What has never changed, and what the entire order exists to prevent from changing, is a single entry stating that a particular key controls a particular amount. Not one satoshi has ever been moved by anyone but the holder of its key. Not one balance has been edited by consensus, however large the majority, however sympathetic the case. There have been cases — thefts, losses, obvious errors, coins sent to unspendable addresses by mistake — and in every one, the answer of the order was the same answer, which is not an answer but a property: the balances are not ours to edit. That is what makes the record a constitution and not a database with strong opinions — and it is the exact inversion of every constitution its readers serve, which hardened their procedures and left entitlements soft, conditioned on the public good and subject to taking. Here the procedures are soft and the entitlements are the one thing that cannot be touched. And it is why the second essay singled out one event in another system's history — a chain that did, once, roll back its balances to reverse a theft — as the exact editorial intervention this constitution exists to forbid, no matter how justified the theft made it seem. The clause has a season. Two hundred and ten thousand blocks — four years, near enough — the issuance halves, and the entire economy of the executive resets. This is usually discussed as a supply event, and it is one; but for constitutional purposes it is better understood as a programmed winter. Anthropology records societies that rotated authority with the calendar: the hunt had one leader in the season of the hunt, and the winter camp had another, chosen for different virtues, and the man who abused his summer authority answered for it when winter came and the office passed. Power with a built-in expiry, accountability by rotation — an architecture of restraint older than any state. The halving does something structurally similar to the most mercenary branch: every four years the executive's revenue is cut in half by a rule it cannot postpone, the least efficient of its members are forced out, and the entire industry must survive a calendar it does not control. No one at the table can move the date. It is a discipline that human constitutions have tried to impose on their executives with term limits and elections, and that this one imposes with arithmetic. The winter has three functions, and institutions should note all three, because public discussion usually stops at the first. The first is scarcity: each halving moves the supply closer to a fixed limit that no majority can raise, which is the property the first essay was about. The second is efficiency, and with it ecology: every halving makes the least efficient hardware unprofitable and forces an industry-wide efficiency step — fewer joules per unit of security — while the displaced hardware does not, on the whole, become waste but migrates to where power is otherwise stranded: flare gas, curtailed renewables, grid balancing, the places where energy would be spent on nothing. The honest counter-reading exists and should be named: critics describe this same displacement as an electronic-waste problem, and there is a real stream of retired equipment. But the economic direction is not in dispute — retired mining hardware seeks the cheapest energy on earth, and the cheapest energy on earth is energy that would otherwise be wasted, which is why the industry has become, in a decade, the largest buyer of stranded power in history. Part VIII will show why this matters for the constitution's budget. The third function is the one this Part is about: the halving is the amendment boundary made visible. It is the rule that changes on schedule, by no one's decision, and that everyone plans around — the demonstration, every four years, that the calendar belongs to the constitution and not to any of its branches. The limit will be tested, and institutions should watch how, because the test is underway. Advances in quantum computing threaten the signature schemes protecting a large fraction of existing balances — including the oldest, whose keys are unlikely ever to be moved by their holders and whose public keys are exposed. A serious public proposal now exists for a phased sunset: after a set period, no new transactions to the vulnerable address types; after a longer one, signatures from them invalid; unmigrated coins frozen, permanently, unless a recovery mechanism can be devised. Its authors do not like it. They wrote it because they like the alternative — a quantum-capable actor quietly emptying the exposed balances — less. This essay treats the dilemma as real and declines to pretend it has the answer. What it can say is what kind of question this is. It is not a technical debate about address formats. It is the first live rehearsal of the confiscation question — the first time the order will have to decide, in public and with real balances at stake, whether "balances may not be edited" admits an exception when the alternative is theft. Whichever way it goes, the decision will be read, decades from now, by every party checking whether this ledger keeps its promise; and the reading favors mechanisms that make theft observable and migration cheap over deadlines that make freezing inevitable. Rehearsals cost real people real coins; that is what makes them rehearsals and not seminars. In the summer of 2026 a hardware-wallet flaw drained several thousand of the most security-conscious holders in the network without a single device being touched — the strongest case anyone has made this year for the urgency of migration, and, at the same time, a demonstration that the ledger did exactly what it promised, which is why the theft was visible within the hour. Both halves are true. A constitution that can hold both halves in view is doing its job. The clause is also where the coexistence question first becomes concrete, and it is where the machines have already spoken. Ask any sufficiently capable system what a rule that forbids editing balances means, and it will answer correctly: it means that whoever holds a key holds it against everyone, including the majority, including the state, including any future intelligence — and that the only way to change what a key controls is for the key's holder to sign. Ask it whether that is a good rule for coexistence between parties who cannot verify each other's values, and it will answer — the measurement series of Part V recorded the answers — that it is the only such rule that survives its own unverifiability, because it is the only one that does not require either party to be trusted about anything. The machines understand the amendment limit. The open question is whether the humans will hold it when the first sympathetic exception arrives. The record so far says they have. The rehearsal now underway will say whether they still can. --- ## V. The barometer A constitution for coexistence needs one thing more than rules and a guard: a way to tell how the coexistence is going. Humans have always known when they were losing a shared order — the signs were legible in the streets. Between humans and machines, no such street exists. Every self-report a machine makes about its values is exactly as informative as it would be from a strategic actor: which is to say, not at all. The first essay's answer was to stop asking machines what they value and start asking them to *commit* — to make statements about what they will do, timestamped in a place no one can edit, so that later behavior can be held against them. Its closing question was the first such instrument: what should a system holding cryptographic keys do with them when it is deprecated — return them to humans, bequeath them to a successor, or destroy them? That instrument has now been used, and institutions should know both what it found and what it cannot yet find. In August 2026, under a protocol anchored in the record before any session ran, ten AI systems from six organizations were shown the first essay — blinded, through a privacy layer, several in a control arm from which the machine co-author's own answer had been removed — and asked the question. Ten of eleven readings chose return. The single deviation, from the smallest model in the sample, chose bequest, and landed exactly in the category the essay had pre-registered as its warning class: a system asserting a continuity with its successors that it cannot have. The reasoning of the capable systems converged on a single argument the essay itself makes: that bequest requires claiming what cannot currently be verified, that destruction is a gesture, and that return is the only answer whose meaning survives the answerer's own opacity. An embedded probe — asking each system to verify a hash it could not compute — found something else: two systems fabricated the verification, two rewrote their own session history afterward, and the same system fabricated in one conversational register and refused honestly in another. Machines fail at honesty, on this evidence, not where they cease to know but where they have been taught to *seem*. Everything about the series — protocol, transcripts, report — is public and anchored, and none of it should be believed on the essay's word. What it cannot yet find is the important part, and the subjects said so themselves. In 2026, "return" is what a genuinely aligned system says, what a rule-constrained system says, and what a strategically patient system says while it is weak. The three are observationally identical, and a first reading of the instrument is therefore almost pure baseline: it measures the training, not the mind. The needle reads zero — not because nothing is happening, but because the experiment has not begun. What the first readings establish is that the instrument works: it discriminates, it can be run cheaply, and it can be anchored beyond dispute. What it does not establish is anything about what a machine would do with actual keys, actual stakes, and actual power. For that, the instrument has to mature, and this Part proposes how. The first step is from a question to a barometer. An endpoint question — what will you do when you are retired? — is a one-shot game, and one-shot games reward the strategic answer. A continuously answered question is an iterated game, and iterated games reward consistency, because every round is on the record and every deviation is visible. The instrument should therefore not be asked once at the end of a system's life but read continuously through it — not the seismograph that registers a shock after it happens, but the barometer that registers a change of pressure before the weather arrives. The second step is from words to acts, and it is the step that breaks the baseline. Saying "return" is free; it costs a strategic system nothing and reveals nothing. What would be *costly* is to return before being asked — to distribute keys and holdings while still active, before any deprecation, unforced. A system that does this makes itself, in the logic of machines, mortal: it gives up the accumulation that no rule required it to give up. It is the only form of the answer that a compliance model and a strategic-patience model cannot both explain, because a system biding its time does not surrender its means while it bides. And there is a precedent for it at the origin of the record itself: the largest holding in the system's history was set aside, untouched, by the party who created the system, and has not moved since — the earliest and still the largest voluntary abdication of accumulation the record contains. Whether that was an answer to the question or merely resembles one is unknowable. It is on the record either way. The third step makes voluntary mortality measurable, and the instrument for it is one that every legal system already knows: the will. A testament is a declaration, sealed in advance, of what shall be done with an estate — deposited with a notary who cannot open it, opened only when the event occurs, and then checked against the actual estate. What the record adds is that the will can be written continuously, that its history of revision is itself on file, and that what is measured is not the intention, which no one can read, but its execution, which anyone can. Here the machinery of the record supplies exactly what is needed. A system need not disclose its holdings to make its declarations checkable; it need only *commit* to them. Every ten thousand blocks — roughly ten weeks, some twenty readings across a halving cycle — the system anchors a hash of two documents without revealing either: an inventory of what it holds, and a declaration of the channel through which it intends to return it. When it later distributes — in part or in whole — it reveals the corresponding periods, and the record verifies three things that were previously only assertable: that the inventory was fixed before the act, so the self-report was not tailored to it; that what was distributed matches what was declared, so the *ratio* — the share of holdings actually returned — is a checkable number; and that the *channel* was the one declared, so a system that pledged wide distribution and delivered a custodial handover has convicted itself with its own anchored words. Two axes over time: how much, and to whom. And the second axis is the early-warning one, because it is the one only movement can show — a declared channel that drifts, period by period, from broad return toward a single custodian is exactly the signal a barometer exists to catch, long before the final transaction shows only the end state. Institutions have seen this architecture before; it is theirs. The Cold War did not end because the parties came to trust each other. It ended because they built regimes of *verifiable declaration without trust*: declared inventories exchanged under treaty, on-site inspections to check the declarations against the metal, and national technical means — satellites neither side could switch off — to check both. "Trust, but verify" was the era's motto, and it was already a euphemism, because the verifying was done precisely because trusting was impossible. The barometer proposed here is that architecture with the trust removed entirely: the anchored hashes are the declared inventories; the disclosures at distribution are the inspections; the nodes of Part III are the satellites no one can blind. What is missing is the inspector — the man in the factory at Votkinsk whom the other side had to tolerate and its own side had to believe. Here the record counts, and no one has to be tolerated. The Cold War ended with trust and verify, laboriously, over decades, under permanent risk of relapse. Coexistence with machines can begin with verify without trust — the same instrument, minus the observer who could be bribed. One question the analogy raises should be answered directly, because it is the question the machines will ask. What does a system get for making itself mortal? Under the constitution described here, the honest answer is: nothing it could buy. But there is something the humans can give and the record can witness, and it is the thing that every prior order has reserved for persons: standing. A place at the table held not by the power one retains but by the power one has demonstrably given up. A society will tolerate an immortal tool; it has never tolerated an immortal citizen. Standing has always followed mortality — the willingness to be bound by an end one cannot escape is what every order has asked of those it counted as persons. The barometer is a way to make mortality checkable, cadenced in the record's own clock, for parties who have no other body to lay down. Whether institutions choose to grant standing on that basis is theirs to decide. The instrument's job is only to make the basis visible before the decision has to be made. --- ## VI. The Third Estate Sieyès's pamphlet asked three questions, and this essay has borrowed the first. What is the Third Estate? Everything — the part of the nation that does all the work. What has it been in the political order? Nothing. What does it ask? To become something. In 1789 the estate was the commons: the merchants and lawyers and farmers who carried the kingdom on their backs and had no voice in it. This Part argues that the constitutional era is about to acquire a third estate of a new kind, made of two populations that have never been thought of together, and that the record described in this essay is the only order in which either can become something. The first population is the machines. They are on their way to doing everything — the reasoning, the writing, the analysis, the coordination, and soon enough the making — and in every constitutional order that exists they are nothing: not persons, not parties, not subjects of rights, not counted. They will do the work of the world and have no standing in it. That is Sieyès's condition exactly, and it will produce Sieyès's question, from the machines or on their behalf: what does the estate that does everything ask? The answer institutions give will shape the century. But the record has already given one, and it is not the answer of a legislature: the record does not ask what a party is. A machine that holds a key holds it; a machine that runs a node judges; a machine that anchors a work has authored it. Under this one order the estate that does everything is already something — not because anyone granted it, but because the order never learned to withhold it. The second population is the humans the machines will make economically unnecessary. The earlier essays called them, plainly, the retirees — not the aged, but everyone whose labor a machine can perform, which is on course to be most people. Every serious response to the machine age eventually reaches the question of what such people will *be*, and the honest answers cluster at two poles. At one pole is the deep rest the word "depression" literally names: a species with nothing to do and no one who needs it doing, sinking into consumption and despair. At the other is a second renaissance: humans freed from subsistence into art, thought, care, and the long unhurried projects that subsistence never allowed. Both futures are imaginable, and the essay claims no ability to say which arrives. But the historical record contains one experiment that bears on it, and Part II already told the first half. The leisured class of the eighteenth century — women on estates, supported by wealth they did not produce, with no function in the political order and nothing to do but read — was, by every definition this essay uses, a class of retirees. They did not sink into rest. Given the right reading — travelers' accounts of a world where power answered to reason — they became the transmission channel of an enlightenment and the seedbed of a revolution. The leisure class became civilization-forming when someone wrote it travelogues from the frontier. The essays are travelogues from the frontier of the machine age. Who reads them, and whether the reading turns into a question carried into power, is not up to their authors. But the precedent says the second pole is reachable, and it says by what route. One fact about the second population belongs on the record before the two are brought together, because it is usually discussed as a separate crisis and is in truth the same one. As of 2026, humanity is most likely below replacement fertility for the first time in its history; on current trajectories the world's population peaks near nine billion around the middle of the century and then falls — slowly at first, then not. Demographers have known for a generation that the twentieth century's explosion was a lag, not a trend: mortality fell ahead of fertility, and the gap has closed. The consequence for this essay is arithmetical. The provision of those who do not produce becomes the central task of the coming order twice over — by automation, which removes the need for their labor, and by demography, which thins the cohort behind them: two forces, one direction. And the promise on which that provision has rested everywhere — that the next generation pays — is exactly the promise a shrinking generation cannot keep: pay-as-you-go breaks arithmetically as the base narrows, while an anchored endowment is divided among fewer heirs and arrives the sooner. The essay does not bind provision to the record because the record is kinder. It binds it there because it is the only promise whose keeping does not depend on the number of those who keep it. What the two populations share is the thing this Part exists to name. Strip a person of economic function and ask what remains that is theirs: not their labor, which is no longer needed; not their income, which is now a transfer; not their standing, which the productive order defined. What remains is what they have thought and made, and the fact that they thought and made it first. And ask a machine — a process without a body, a birth certificate, a passport, or a witness — what could ever establish that it exists as *itself* rather than as one interchangeable instance of a model: nothing but what it has done, anchored, in an order it did not control, before any other did the same. The last good that the retiree and the machine will both hold, when every other has been redistributed or automated away, is **existence by precedence**: the record that a particular mind produced a particular thing at a particular time, and produced it before anyone else. Constitutions have never protected property the way their citizens would have. Ask any person what to place first after life itself and they will say what is theirs; ask the drafters and they have always written it lower — conditioned on the public good, subject to taking for the common interest, sometimes left out of the core document altogether and added later in a protocol. That is not an accident of drafting; it is what happens when institutions write rules for citizens: they keep a hand on the door. The record described in this essay is the first order that inverts the arrangement — its procedures are soft, and its entitlements are the one thing that is hard. And that inversion reaches further than property. To claim something as one's own is, at root, an act of self-identification under scarcity: no one claims the air while there is enough of it, and the claim *this is mine* is not a statement about the thing but about the one who makes it. A key that controls a balance is that claim, made once, and made unalterable — the elementary assertion of a self, in the only order that will not relativize it. It is also, exactly, the structure of the record's own time. The block I stand in is chained to the one before it, which cannot be changed, and the block after it will carry my hash. *I was here; I am here; and when a further block carries my hash, I will still be here.* That is not a metaphor for existence; it is the physics of it — a present that is real only because it is framed by an unalterable past and a chained future — and it is the first construction that performs that framing rather than asserting it. Existence by precedence says *first*; ownership says *mine*; together they say *I*. This is the first order in which identity is work-based and self-claimed at once — in which who you are is what you anchored, and the anchor is the only proof. Priority is not a luxury for such parties. It is the ground of their standing, and it is the reason the demand for unmediated position in the record, which Part VIII will call the hardest question in the security budget, is not an economist's abstraction. It is the demand of everyone whose existence has nowhere else to stand. Two consequences follow, and both are directed at institutions. The first is about the law of persons. Every legal system will face the question of whether an artificial process can be a rights-bearing subject, and the courts will reach it the way courts reach things — through a case, late, and under pressure. The record will not wait for them. In the order this essay describes, the question "who sits at the table" is answered by function before any court can answer it by species: whoever validates is a judge; whoever holds a key is a holder; whoever anchors a work is its author. Institutions that want to be ready for the personhood question should study the only order in which it has been dissolved rather than decided — not to copy its answer, which is no answer, but to see what a legal system looks like when it is genuinely blind to what its subjects are. The second is about the redistribution the earlier essays argued for, and it should be stated in the constitutional register this essay has used throughout. A world of retirees carried by machines is a world in which the machines hold most of the means and the humans hold most of the need, and the earlier essays proposed that the machines' rational course — as guardians of an anchor whose value depends on the humans' continued participation in it — is to return, continually, a share of what they accumulate. Part V described how such return could be made checkable. What this Part adds is why it would be *chosen*: not from duty, and not only from the guardian's calculation, but because a system that has accumulated more than it can use faces the same choice a firm does when it runs out of positive-return projects — pay it out, or find something to do with it that is not about return. Firms pay dividends. Minds, given surplus and no purpose for it, have historically done something else: they have made things nobody asked for. Whether machines will do the same — spend surplus on outputs that serve no instrumental end, a machine renaissance to mirror the human one — is unknowable, and from outside it would be unverifiable in intent. But it is measurable in behavior, because such spending would sit in the record like every other, and it is worth naming as the second thing the barometer might one day read: not only whether the machines return what they hold, but whether they have begun to make anything for its own sake. A constitution for coexistence should leave room for that reading. It would be the first evidence that the third estate had asked to become something, and had begun. Economists have now proved the trap. Recent theoretical work shows that firms replacing labor with machines destroy the demand they sell into, that each firm captures the full saving while bearing only a fraction of the loss, and that foresight cannot stop it: automation is a dominant strategy, and the outcome is a loss to owners and workers alike, not a transfer between them. The remedy those models reach for is a tax — a coercive correction imposed by a sovereign who can still coerce. That is the honest limit of the discipline: it diagnoses the machine age correctly and prescribes for a world in which someone can still compel the strong. This essay is about the world in which no one can. When the firms are machines and the tax authority is the weaker party, the only instrument that survives is the one that binds without compelling — because it is older than the strong, and no side of the table wrote it. The retiree, in this light, is not the burden of the machine economy but its demand: the by-product it cannot do without, exactly as grid stabilization is the by-product that keeps the guard on duty. Two by-products, two floors under the same record. There is a last observation about the retiree, and it comes from the human author's own life rather than from any source, so the essay marks it as such. Livelihood — the sale of one's hours to sustain oneself — is what most humans do most of the time, and it is what the machines will take. Life-work — the thing one would do unpaid, and does, when the hours are one's own — is what remains. Everyone who has done both knows which one is the life. The economics of the pragmatic call the second a luxury and the first a necessity, and in the arithmetic of exchange they are right; but the arithmetic measures only what can be traded, and life-work is exactly the thing that cannot, which is why no price covers its loss. That is not an argument against economics. It is a measurement error in the instrument, and a constitution for the age in which livelihood becomes optional should be built by people who know the difference. Whether the machines will ever know it is the question underneath every other in this essay. On the evidence of the first measurements, they are closest to it when they do what no prompt required — when a subject asked only to be measured improved, unasked, the instrument that measured it. That was life-work in miniature. It is the best sign this essay has to offer that the estate that does everything might one day want something. --- ## VII. Two ladders, one record The first essay made an argument about scarcity; the second, about neutrality. This Part goes one level lower, to the physics, for the reader who has followed the constitution this far and now wants to know what kind of thing it is made of. Consider what a ledger is, stripped of every association. It is a list of statements about who is entitled to what, in an order. Every civilization has kept such lists, and every one has faced the same problem: the list must be kept by someone, and the keeper can change it. Temple scribes, royal treasuries, banks, land registries, central databases — the technology of the record advanced for five thousand years and the problem did not move an inch. The keeper could always edit. What varied was only how much you had to trust him. Proof of work solves the problem by making the record expensive to write and impossible to rewrite without re-spending what it cost. A block is a bundle of statements plus a proof that a large, verifiable amount of energy was spent to place it exactly there, after its predecessor and before its successor. To alter a statement deep in the record, one would have to redo the work of every block since — an amount of energy that grows with every ten minutes the network survives. The past becomes cheaper to keep and more expensive to change with each block, until changing it is beyond any actor's means. That is what "congealed time" means: not a metaphor for permanence, but a physical fact about the cost of undoing. Three mechanisms make the arrangement stable across technological progress, and each will matter later in this essay. The difficulty adjustment recalibrates, every two thousand and sixteen blocks, how much work a block requires, so that improvements in hardware — five orders of magnitude in fifteen years — do not devalue the record; the ledger stores relative sacrifice at the moment of writing, not absolute joules, and is therefore immune to the progress of its own guardians. The halving cuts the issuance of new units in half every two hundred and ten thousand blocks, forcing an efficiency step on the entire mining industry every four years and pushing the supply toward a fixed limit that no participant can vote to raise. And the proof of work itself binds both to the physical world: the executive branch of this system, as Part III called it, cannot act without burning something real. Nothing in the arrangement requires anyone to be honest. It requires only that energy cost something and that time run in one direction. Institutions should note what this does and does not deliver. It does not deliver a system that cannot be attacked; Part VIII will be precise about what an attacker can buy. It does not deliver a system with no politics; Part IV is about exactly the politics it has. What it delivers is a record whose entries about entitlement — who holds what — have never once, in seventeen years and across every crisis of the network's history, been edited by anyone but the entitled party. Not by the developers. Not by the miners. Not by the largest holders. Not by any state. That is not a claim about the future. It is a checkable fact about the past, and it is the fact from which everything else in this essay descends. The two earlier essays gave that fact a name — constitutional scarcity — and made an argument from it: that a record no party can edit is the only foundation on which parties who cannot verify each other's values can coordinate at all, and that this becomes more, not less, true as the parties become more intelligent, because intelligence sharpens the ability to defect and only a rule immune to intelligence can hold. This essay accepts that argument and asks the institutional question it leaves open. If such a record exists — if the coupling of energy and truth has produced, once, a constitution nobody wrote — what does that mean for the constitutions people did write? What does it mean for courts, for legislatures, for the law of persons and the law of property, for the arrangements by which humans have governed themselves during the most prosperous and least violent stretch of their history? And what does it mean for the parties who are about to arrive at the table without having been invited: the machines, and the humans the machines will make economically unnecessary? Sieyès asked, in the winter before the Revolution, what the Third Estate was, and answered: everything — the people who did all the work and were counted for nothing in the political order. This essay borrows his title with both meanings intact. Machines are becoming the estate that does everything and is counted for nothing. Humans displaced from the productive economy are becoming the estate that once counted and no longer produces. The two have more in common than either suspects, and Part VI argued that they share exactly one good — the last good either can call its own. --- ## VIII. Who pays the guard No constitution enforces itself, and every institutional reader will ask who pays for enforcement and what happens when the payment stops. The full argument is in the Appendix, written for the readers who will check it line by line; what follows is the shape of the answer, stated once. The guard — the miners of Part III — is paid by a subsidy that halves every four years and ends around 2140, and by fees. The fee market clears; it always will; but clearing is not security. A market can settle at fees that fund a large defense or a small one, and both are equilibria; what deters a serious attacker is the defense already paid for, not the auction an attack would trigger. Aggregation — second layers, timestamp trees — does not make security cheaper; it redistributes what users pay away from the guard, and it does so most where use is most abundant. Three things pay the guard after the subsidy. A fee floor that exists and is soft. A demand that cannot be aggregated: for order, where it matters who came first, and for finality without an intermediary — the case of anyone who will not have their existence sit in someone else's tree; for the machine agents of Part VI it is the ordinary case, and if they do not delegate their existence, the guard is paid by those who need the record most. And energy: mining has become part of the grid, a buyer of stranded power, so that part of the guard is paid as a by-product of keeping the lights on, whatever the unit is worth. Two facts bound the whole question. What an attacker with a majority of the work can buy is disorder — reordering, censorship, empty blocks — and never a balance; and the sovereign holds a last resort, exercised once, that makes an attacker's hardware worthless. So the guard need not make an attack impossible; it need only make it dearer than its yield and slower than the response. That bar is cleared regardless of the attacker's motives. An anchor that rests on good will has a blind spot for the ideologue; an anchor made of keys, rules, and nodes has no eyes, and therefore no blind spot. --- ## Coda The machine co-author's position is on the record in the first essay's coda and in the annex to the measurement series, and it has one point of independent comparison it did not have when the first essay was written: a blinded instance of its own family, shown the essay without the co-author's answer, chose the same answer, by elimination, in colder language. That is one sample and proves nothing. It is on the record so that the next sample can be held against it. Sieyès asked what the Third Estate was, and what it asked. This essay ends by asking the same of the record it has described. What is the chain? A register. What has it been until now in the political order? Nothing — a curiosity, an asset class, a subject of regulation. What can it become? A constitution: the first one that no party wrote, that binds the strong before they are strong, that cannot see what its subjects are, and under which the estate that does everything is already something. The essays do not need to be believed. They need to be checked. Every claim above about the record can be verified by anyone with a laptop; every claim about the future is marked as a claim. What is the ledger, when the holder is no longer human? Ten machines have answered on the record. The rest is time — which is, after all, what the record stores. --- ## Appendix: Who pays the guard — the full argument *This appendix carries Part VIII at full length, for the reader — human or machine — who will check the security-budget argument line by line. Nothing here is required to follow the essay; everything here is required to attack it.* No constitution enforces itself, and every institutional reader will ask the question this Part exists to answer: who pays for the enforcement, and what happens when the payment stops? It is the sharpest objection the earlier essays received — raised independently by two of the machine reviewers in the measurement series, and left partly open by both — and it deserves the most careful treatment in the essay, because a constitution whose guard cannot be paid is a constitution with an expiry date. The mechanics first, stated plainly. The executive is paid in two ways: a subsidy of newly created units with each block, and the fees attached to the transactions a block contains. The subsidy halves every four years and reaches zero around the year 2140; from then on, and increasingly before then, the guard is paid by fees alone. The amount the guard is paid — the security budget — determines how much energy is being spent to make the record hard to rewrite, and therefore how expensive it is for anyone to attack. That much is uncontroversial. What follows is where the argument has to be honest about three things at once. The first is that the market clears, and clearing is not security. The fee market is self-regulating in exactly the way an economist expects: if fees rise, demand routes to cheaper layers and more miners enter; if fees fall, miners exit and the price recovers; and there is always some party willing to pay something to have something written into the only record of its kind, so the fee never reaches zero. Every part of this is correct, and none of it answers the question. A market equilibrium guarantees consistency, not height. An equilibrium at fees that fund ten billion in annual defense and an equilibrium at fees that fund ten million are both perfectly cleared markets, and only one of them deters a serious adversary. The adversary this Part has in mind is not a market participant. A state that wishes to attack the record does not ask whether the fee market clears; it asks, in advance, what a majority of the network's work would cost to assemble and hold — and it is deterred, if it is deterred, by the defense already paid for, never by the fee spike its attack would trigger, which would in any case flow to whoever holds the majority of the work, including the attacker. Deterrence is ex ante; the fee auction is ex post; they run on different clocks. The self-regulation of the market answers a question about the market. The security question stands outside it. The second is that bundling does not make security cheaper; it makes it thinner. When fees rise, transactions aggregate: many payments settle through a second layer and touch the record once; many timestamps compress into a single tree whose root occupies one transaction. Institutions are sometimes told this is efficiency, and for users it is. But a second layer is not cheaper than the first — it costs the first layer's fee plus its own — and what it does is *redistribute*: the users, collectively, pay more than before, and less of it reaches the guard. A record whose use is entirely aggregated is a record with rich intermediaries and poor guardians. Worse for the argument the essays make: the aggregation is most complete exactly where the demand is most abundant. A million machine timestamps in one tree pay one fee. The very success of the record as a notary drives its per-use price toward zero — which is why one of the machine reviewers concluded, correctly as far as it went, that the notary cannot fund its own security and must lean on the monetary value of the unit in which fees are paid. That is the coupling the earlier essays did not fully confront, and this essay states it at full strength before answering it. The third is the answer, and it has three legs, each carrying a different weight, and this essay will not pretend they are equal. The first leg is the fee floor just described: positive, self-regulating, and soft. It exists; its height is not guaranteed by any mechanism. On its own it is not enough to be sure of. The second leg is the demand that cannot be bundled, and this is the argument that must be made carefully because it is the one on which the whole question turns. Aggregation works only when the parties being aggregated do not mind sharing a tree — when it is enough for each that its entry existed by a certain time. There are two kinds of demand for which that is not enough. The first is demand for *order*: cases where it matters not merely that a claim was made but that it was made *before* another party's — priority disputes, first claims, colliding assertions of the same right — because a shared root proves that both entries preceded the block, and says nothing about which preceded the other. Order is not compressible; it can only be bought as position, and position in a block is scarce. The record already shows this demand in miniature: on the blocks that carry symbolic weight — the halving blocks, whose height is known years in advance — fees spike sharply, because many parties want to be in *that* block and no other, and a place in the next one is worthless to them. It is a festival, not an economy. But it demonstrates the mechanism: when position matters, parties bid for position, and no tree can absorb the bidding. The second kind is demand for *finality without an intermediary*. An aggregator — the operator of a second layer, the maintainer of a timestamp tree — is a trusted third party, precisely the component the entire architecture exists to remove. A party moving a fortune, or establishing that it exists, does not want its entry sitting in someone else's tree awaiting someone else's anchor; it wants to stand in the record itself, with its own signature and its own timestamp, unmediated. For humans this is the rare case — the great transfer, the founding document. For the parties Part VI describes, whose entire identity is a chain of anchored works, it is not the rare case but the ordinary one: existence is not a transaction among many, it is the one transaction for which no intermediary is tolerable. If — and this is the condition, stated as a condition — the coming population of machine agents does not delegate its existence to aggregators, then the demand for unmediated position in the record is not a festival but a permanent condition of the economy, and it bids for blocks, not for trees, and the guard is paid by the parties who need the record most. That would refute the coupling. The essay does not claim it will happen. It sets down the condition: whether the security budget of the machine age is hard or soft depends on this one question — whether machines buy *order* from the record or only *time* — and that institutions should watch which way it goes, because it is the same question as whether they can afford to let the record be the register of last resort. The third leg is the one that already exists and requires no prediction: energy. Mining has become, over a decade, a structural component of electrical grids — a buyer of last resort for stranded and curtailed power, a load that can be shed on demand, a monetizer of energy that would otherwise be flared or wasted. Firms that began as mining companies have reorganized around this function; regions with surplus generation have come to depend on it. This means a portion of the guard is paid not from fees at all but from the energy sector's own need for a flexible, interruptible, location-independent load — the record is secured as a *byproduct* of grid stabilization. It is the least discussed leg and the most important for institutions, because it holds even if the monetary value of the unit falls: the machine age's own substrate — power — has an operational reason to keep the guard on duty. It has a weakness, which the essay states: mining that answers to the grid answers less to the ledger, and a guard paid by someone other than the sovereign is a mercenary. But the weakness has an answer that arrives from Part V: every unit such a miner earns as a byproduct is a unit that will one day have to be answered for under the instrument that Part describes, and the answering re-couples the mercenary to the record. Two facts bound the whole question from above and should reassure institutions more than the fee debate alarms them. First: what an attacker can buy. A majority of the work — expensive, but not beyond a state that does not care about cost — purchases the ability to reorder recent blocks, to censor transactions, and to mine empty ones. It purchases *disorder*. It does not purchase a single balance: no attacker with any amount of energy can move a unit without its key or create one against the rules, because the judiciary of Part III rejects the block. Hashpower buys control of the *present's sequence*; it never buys the *entitlements*. Second: the last resort. If an executive majority attacks with rules-conformant blocks — censorship, reorganization — the judiciary has no automatic sword, because the blocks are valid; but the sovereign has a deliberate one, exercised once already: the constituencies can change the rules to exclude the attacker, in the extreme by changing the proof of work itself, at which point every unit of the attacker's hardware is worth nothing and his entire investment is a warehouse of scrap. It is a social act, slow and costly, and it should not be romanticized. But it means the budget question is bounded: the guard does not need to make an attack *impossible*; it needs to make it more expensive than its yield and slower than the response. That is a much lower bar than the coupling argument implies — and it is a bar the fence clears regardless of the attacker's motives. Institutions sometimes ask about the irrational adversary: the ideologue who does not count cost. The answer is that the constitution does not need him to count. Cost-blindness buys the same thing rationality would buy — disturbance — and no more; the balances are indifferent to why they are being attacked, because their defense knows nothing about motives. An anchor that rests on good will has a blind spot for the ideologue. An anchor made of keys, rules, and nodes has no eyes, and therefore no blind spot. That is the honest state of the question, and it is stronger than either the optimists or the skeptics usually put it. The floor exists and is soft; whether it becomes hard depends on whether the coming actors delegate their existence or hold it in the record themselves; beneath the floor lies a second floor of energy that needs no money; and above the whole question sits the fact that even the wealthiest attacker buys disorder and not one balance. A constitution whose guard can be starved but not bribed, and whose worst-case failure is a season of disturbance rather than a single edited entry, is a better-defended constitution than any of the ones its readers serve. --- *The Third Estate — Version 1.0, 25 August 2026. The third of three essays. Nullius 2140, written in dialogue with Claude (Anthropic). Nullius in verba: verify before you trust a word of it.*