Chapter 10
The Exception
Two Histories
Aa
On the day the world ends / A bee circles a clover, / A fisherman mends a glittering net... / And those who expected lightning and thunder / Are disappointed.
The Mills Before Midnight
A nationwide steel strike was scheduled to begin at one minute after midnight on April 9, 1952. The United States was fighting in Korea, steel moved directly from mills into weapons and defense production, and President Harry Truman believed even a short interruption would place soldiers and the wider economy at risk. On the evening before the strike, he signed Executive Order 10340, directing the Secretary of Commerce to take possession of the plants and facilities of seventy-four steel companies and keep them operating. 1
The order was extraordinary in its reach and ordinary in its physical effects. The furnaces remained where they were. The companies' managers were asked to continue running them, now under federal direction, while the Secretary of Commerce acquired authority over operation and employment conditions. Nothing in the mills announced that constitutional authority had changed hands. Steel continued to emerge from the same machinery, under the same roofs, while the legal order around the machinery became suddenly uncertain.
Truman did not conceal his reason. In a national address that night, he described a grave danger and insisted that the mills could not be allowed to close. Congress had provided other tools for labor emergencies and defense production, but none supplied the seizure power he claimed to need on the timetable before him. He therefore acted under what the government described as the President's inherent constitutional authority, informed Congress the next day, and invited legislation after the fact. The emergency came first. The legal basis would have to catch up. 2
The steel companies went to court. Less than eight weeks later, the Supreme Court held the seizure unauthorized. The decision did not rest on the proposition that steel was unimportant, that war production faced no danger, or that the President had imagined the crisis. The emergency was real enough to divide the Court sharply. The constitutional question was whether the emergency itself supplied a power that Congress had not granted and had, through other legislation, allocated differently. Six Justices concluded that it did not. 3
Justice Robert Jackson's concurrence supplied the account that outlived the immediate dispute. Presidential power, he argued, changes with its relation to Congress. It reaches its maximum when the branches act together, enters a zone of uncertainty when Congress is silent, and falls to its lowest ebb when the President acts against Congress's expressed or implied will. Truman's seizure belonged to the third category because Congress had legislated in the field without authorizing the remedy he chose. Crisis intensified the stakes without dissolving the allocation of authority.
Youngstown is often remembered as a limit on executive power, which it is, but the event matters here for a more exact reason. The constitutional order did not answer the emergency by ensuring that no one decided. Truman decided to seize, the companies decided to resist, courts decided whether the claimed authority existed, Congress retained the capacity to authorize another course, and the mills continued operating while those decisions collided. The achievement lay in preventing the first decision from becoming the last simply because it came first.
The previous chapter ended by distinguishing the ordinary hard case from the extraordinary claim that ordinary rules can no longer govern. In the penumbra, the institution still accepts the rule and disputes its application. In the exception, someone asserts that the rule, the ordinary process, or the existing allocation of authority cannot meet the danger in time. Interpretation remains inside the order. The exception asks permission to move beyond it, or at least ahead of it.
That distinction is easy to announce and difficult to preserve. A disputed interpretation can be inflated into an emergency whenever the ordinary process is inconvenient, while a genuine emergency can be forced through procedures that guarantee the procedure will finish after the interest it was meant to protect has disappeared. The label resolves nothing. It transfers attention to the person or institution authorized to apply it.
The constitutional question is therefore not whether someone will decide. Someone will. It is whether the actor who declares the emergency also controls the response, the evidence, the duration, the review, and the moment at which normal authority returns. An emergency becomes sovereign when one hand can complete that entire sequence.
Youngstown supplies the first answer. The emergency actor need not own the emergency.
The Decision Outside the Rule
Carl Schmitt compressed the opposing view into one sentence: "Sovereign is he who decides on the exception." 4 The force of the sentence lies in the word decides. A legal norm presupposes a sufficiently normal world in which the norm can be applied. When the presupposition fails, Schmitt argued, the norm cannot restore the conditions of its own authority. Someone must determine that the ordinary situation no longer exists and act before law can resume. The person capable of doing so is the sovereign, whatever the constitution says about where sovereignty formally resides.
Schmitt's later service to the Nazi regime belongs to the meaning of the theory rather than to an incidental biography. He used the exception to attack liberal restraint and justify concentrated power. The diagnosis can nevertheless be taken seriously without accepting the prescription, because every order does encounter situations in which delay, procedure, and divided responsibility appear to endanger the thing the order exists to preserve.
The Protocol Republic cannot answer by saying that code will remove the problem. Mathematics can reject an invalid signature, enforce a time lock, and prevent an unauthorized state transition. It cannot decide whether the signature rule should yield because the signing key was compromised, whether the time lock should be shortened because an exploit is draining assets, or whether a state transition that satisfied every encoded condition has destroyed the relation those conditions were meant to protect. The exception begins where the mechanism's own success becomes part of the danger.
Nor can the answer be that no one acts. Inaction can decide as conclusively as intervention. Refusing to pause a protocol permits the next block. Refusing to alter a defective rule lets the defect continue. A constitution that treats omission as politically innocent merely hides its preference for the existing trajectory.
The stronger answer is to divide the decision Schmitt gathered into one person. Declaration, stabilization, alteration, review, renewal, and return need not belong to the same actor. One institution can identify the emergency, another can take a narrowly preservative step, another can test the evidence and authority, and the extraordinary measure can expire unless a different body renews it. None of these divisions makes judgment disappear. They prevent judgment at one point from becoming control of the entire constitutional sequence.
This is the tradition Clinton Rossiter tried to recover in his study of constitutional emergency. Liberal orders had repeatedly created extraordinary powers that were limited by purpose, territory, duration, authorization, or review, with mixed and sometimes disastrous results. His central warning was not that emergency power could be eliminated. It was that the power said to preserve the republic must be prevented from becoming the instrument through which the republic is permanently remade. 5
Youngstown is a better constitutional model than the fantasy of a rule-complete order. Truman could act, but his action had an author and an instrument. The companies could challenge it. A court outside the executive chain could say that the asserted authority did not exist. Congress could still legislate. The Court's judgment did not rewind the weeks of government operation or prove that every alternative was preferable. It denied the executive the power to convert urgency into its own source of law.
Computational emergency requires the same separation in a different material. A pause key, guardian council, emergency multisignature, client release, oracle override, or irregular state change can be indispensable. It can also become the hidden constitution of a system that advertises rule-bound governance while reserving a quiet route around every rule that matters. The fact that an emergency path is written in code does not make it less political. It identifies the location at which politics can re-enter most quickly.
The relevant design aim is not emergency without decision, and not even emergency without power. It is emergency without monopoly over the full cycle of power. The sovereign function is not abolished. It is broken into pieces that do not fit comfortably in one hand.
Block 1,920,000
The DAO was not quite the managerless machine later memory made of it. In the spring of 2016, it sold approximately 1.15 billion DAO tokens for about twelve million ether, then valued at roughly $150 million. The organization was described as a for-profit virtual entity whose token holders would vote on projects and share in their returns, with contractual terms formalized and enforced through software. Yet human curators selected by Slock.it controlled which proposals could be placed before token holders and performed what promotional materials called failsafe functions. The Securities and Exchange Commission would later describe their power as considerable. 6
This resistant detail matters. The DAO did not begin as pure code and later discover governance. Governance was already there, distributed among source code, token votes, curators, interfaces, developers, and the expectations created by people promoting the project. The code carried much of the institution, but never the whole of it.
On June 17, an attacker exploited the DAO's recursive split function. The contract transferred ether before updating the internal balance, allowing the function to be called again before the first call finished. Approximately one-third of the DAO's assets moved into a child DAO. The code had not been bypassed. It had been induced to repeat an action its designers had not intended to be repeatable in that way. A built-in waiting period meant the ether could not immediately be withdrawn, leaving the community a short and unusually visible interval in which to decide what, if anything, Ethereum should do. 7
The first answer was a soft fork that would prevent the affected contracts from moving the ether. That response looked modest. It would freeze rather than reassign, preserving time while the argument continued. Then researchers found that the proposed intervention created its own denial-of-service vulnerability, allowing attackers to impose costly computation on miners without paying the ordinary gas cost. The emergency mechanism had become another attack surface. The soft fork was abandoned. 8
The episode contains a lesson that disappears when the story is told only as immutability against mercy. Emergency tools are mechanisms too. They are written under pressure, tested against incomplete knowledge, and deployed into the same adversarial environment that produced the crisis. A safeguard rushed into place may widen the vulnerability it was meant to contain. The exception does not suspend the need for verification. It increases it at the moment verification is hardest to perform.
The eventual proposal was more invasive and more exact. At mainnet block 1,920,000, clients supporting the DAO fork would execute an irregular state change, transferring ether from a specified list of DAO and child-DAO accounts into a withdrawal contract through which token holders could recover funds. The fork did not alter the Ethereum Virtual Machine, transaction format, or ordinary protocol rules. It inserted one exceptional change into the state those rules governed. 9
The Ethereum Foundation said that no single entity could make the decision and used a community signaling process to determine the default behavior in its Geth client. Developers implemented both paths. Miners chose which client and configuration to run. Application operators, node operators, exchanges, wallets, and users made their own consequential choices around the fork. On July 20, the irregular state change executed, with the Ethereum Foundation reporting that roughly eighty-five percent of miners were initially mining the forked chain. 10
The non-fork chain did not disappear. It continued as Ethereum Classic. Holders of ether before the split possessed assets on both chains, while replay risk and diverging infrastructure made the practical meaning of that duplication less simple than the ledger suggested. The Ethereum Foundation later committed its attention and resources to the forked chain based on several measures of community and ecosystem adoption, while acknowledging the continuing non-fork chain. 11
The event is often narrated as a vote between two principles. On one side stood code as law, according to which the attacker had done what the contract permitted and state history should remain untouched. On the other stood social consensus, according to which the transaction defeated the evident purpose of the arrangement and could legitimately be undone. The opposition is real, but it is incomplete. The DAO had curators, promotional representations, security assurances, token-voting rules, a user interface, and a wider legal environment. The question was never only what the bytecode allowed. It was which of these overlapping sources of authority Ethereum would allow to govern its own state.
The hard fork answered that question for one branch. It did not answer it for everyone.
Who Decided Ethereum?
No single actor commanded the DAO fork. That fact is constitutionally important and easily romanticized.
Core developers could write and release the client changes but could not make miners run them. Miners could secure a chain but could not determine which applications, exchanges, wallets, and users would treat it as the continuation of Ethereum. Exchanges could assign the familiar ticker and concentrate liquidity without writing the consensus rules. The Ethereum Foundation could recommend, fund, communicate, and set defaults in the clients it maintained, but it could not extinguish compatible software running elsewhere. Users could choose, though their choices arrived through interfaces, service providers, and defaults created by actors with far greater technical and economic power.
The sovereign function had become a relay. Proposal, implementation, activation, validation, naming, market support, and continued use were separate acts performed by different institutions. No participant controlled the whole sequence. Several controlled indispensable parts of it.
That is not sovereignty absent. It is sovereignty decomposed and unevenly distributed.
Defaults illustrate the difference. The Foundation's process did not compel a user to support the fork, but the result helped determine how widely used software would behave unless someone acted to change it. A default is not a command. It is also not neutral, especially in a system where most participants do not compile clients, inspect release diffs, or maintain competing infrastructure. The path requiring no intervention usually acquires an advantage before deliberation begins.
The same is true of names. Which chain kept the ETH ticker, which one exchanges treated as the principal market, and which one developers expected applications to support shaped the value of each constitutional answer. These decisions did not alter the minority chain's code. They altered the world in which that code had to live.
Fork rights therefore accomplished something narrower and more defensible than rule by pure selection. They prevented the majority branch from making its state change universally final. Dissenters could continue the prior history, and the existence of Ethereum Classic demonstrated that the Foundation, miners on the forked chain, and the exchanges supporting it could not erase the alternative altogether.
They could nevertheless make the alternative more expensive. A minority may keep the ledger and lose the economy that made the ledger valuable. It may retain keys while losing application support, liquidity, developer attention, familiar names, and the network in which those keys had practical force. Continuation is not equality.
Nor does market success settle legitimacy. The chain attracting more capital and use has won adoption, not a philosophical argument. Network effects, defaults, institutional support, coordination capacity, and risk tolerance all shape the result. Selection reveals what participants did under those conditions. It does not transform the majority's choice into the correct constitutional answer.
The DAO fork thus answers Schmitt only in part. No actor possessed an uncontested power to end the disagreement. The disagreement was allowed to become two systems. Yet each system still required decisions about what state it would recognize, which software it would run, and which actors could alter that software again. Forking pluralized the exception. It did not remove decision from it.
The proper question is no longer whether a hidden sovereign exists somewhere in the protocol. It is whether any actor or coalition can control every stage by which an emergency becomes a new normal, and whether dissenters possess a practical route that keeps disagreement from becoming disappearance.
The Emergency Charter
A constitution cannot specify every emergency in advance without redefining the emergency as an ordinary case. It can specify the powers through which an emergency is handled.
That distinction is the basis of an emergency charter. The charter does not predict the exploit, market failure, compromised key, corrupted oracle, or cascading agent error that will arrive. It allocates authority before the arrival so that the first institution able to act does not acquire the rest by momentum.
The initial power is declaration. Someone must state what has happened, which interest faces imminent harm, what evidence supports the claim, and why ordinary procedures cannot respond in time. The declaration may later prove mistaken. Its constitutional function is to commit the declarant to a contemporaneous account before the result of the intervention makes another explanation convenient.
The next power is stabilization. Where possible, emergency action should preserve options rather than choose the final allocation. A pause, quarantine, rate limit, temporary withdrawal cap, or narrowly scoped revocation may stop compounding harm while leaving the underlying dispute open. The abandoned DAO soft fork was an attempt of this kind. Its vulnerability also shows that a stabilizing measure must be independently tested and capable of being withdrawn quickly when it creates a second danger.
Alteration is a different power. Reassigning balances, rolling back state, invalidating a completed transaction, or changing the rule for future conduct resolves interests rather than merely preserving them. It should therefore require a higher burden, a broader constituency, and an account of why less irreversible measures would not suffice. The DAO hard fork crossed this line. It did not merely stop the child DAO from moving ether. It changed the state from which ownership claims would proceed.
Review must sit outside the chain that benefits from the intervention. The actor proposing a pause may possess the best technical knowledge and the strongest institutional interest in defending what it proposed. Neither fact makes it the proper final judge. An emergency council reviewed only by its own members is ordinary discretion wearing a faster clock.
Duration is its own constitutional object. The emergency measure should expire automatically unless renewed through a separate act supported by current evidence. Renewal cannot be treated as the quiet continuation of the first decision, because the burden changes as time passes. What was necessary for six hours may be indefensible after six days, and an emergency authority that cannot identify the procedure and date of its own disappearance is not temporary authority.
Finally comes normalization. The charter must specify who restores ordinary rules, what happens to actions taken under the emergency, whether affected parties receive restitution, and which parts of the record survive. An emergency that stops causing visible interventions while leaving its keys, committees, or exceptions permanently available has not ended. It has become latent government.
These powers should not belong to one actor. The body that declares should not automatically execute. The body that executes should not control review. The body that reviews should not be able to extend duration without a new public act. The institution that normalizes should preserve the evidence needed to evaluate what occurred rather than erase the exceptional record in the name of returning to normal.
Every intervention should therefore produce an emergency receipt. The receipt identifies the triggering condition, the authority invoked, the exact systems and state affected, the measure taken, its intended duration, the alternatives considered, the evidence relied upon, the actors participating, the review route, and the procedure for return. A receipt does not make the intervention lawful or wise. It creates the object through which lawfulness and wisdom can be disputed while the institutional memory is still intact.
The design principle is one of least constitutional violence. Pause before rollback where pause preserves the interest. Quarantine before confiscation. Use the smallest scope capable of stopping the harm. Preserve evidence. Make irreversible action carry a heavier burden than reversible action. These presumptions may yield when delay would itself make recovery impossible, but yielding should be visible as a constitutional choice rather than smuggled inside the language of technical necessity.
Youngstown and the DAO sit on opposite sides of this architecture. Truman acted through a named order and faced external review able to terminate the asserted power. Ethereum's response was publicly specified and inspectable in code, but authority over proposal, client defaults, validation, naming, and market support was informal and uneven. Each system possessed something the other lacked. The constitutional problem is to join the inspectability of the protocol to the separation and review of public law without importing the opacity and delay of either.
The Last Resort
Forking belongs at the end of the emergency sequence, not at its beginning. A protocol that resolves ordinary disagreement by splitting has failed to maintain a common order, just as a state that treats every political loss as cause for revolution has ceased to govern.
Normal rules should decide the ordinary case. Interpretation should govern the penumbra. Amendment should change rules that the community still accepts as common. A bounded emergency process should contain immediate danger and preserve a route back. Forking becomes legitimate when the disagreement concerns the constitutional identity of the system itself and no shared procedure can produce a resolution that both sides remain willing to inhabit.
Locke's right of revolution supplies the analogy, but the analogy has limits. Revolution in a territorial state exposes bodies, homes, families, and institutions to violence that cannot be duplicated onto a second chain. A protocol fork can copy code and state at a chosen point, allowing rival continuations to begin from a common history. That difference is morally significant. It does not make the fork cheap. 12
Albert Hirschman's three terms are useful here because the third is often dropped. Voice is the attempt to repair the institution from within. Exit is departure. Loyalty is the attachment that makes voice worth attempting and exit costly enough to reserve for serious failure. A community with no exit can convert loyalty into captivity. A community in which every inconvenience produces exit never acquires the continuity needed for common judgment. 13
Fork rights strengthen voice when they remain credible in the background. A governing coalition that knows dissenters can continue elsewhere has reason to treat minority objections as more than ritual. Yet the threat works only if departure does not require abandoning every asset, identity, relationship, and accumulated claim that made membership valuable. Otherwise the supposed right disciplines no one.
A fork is therefore closer to constitutional secession than to an ordinary vote. It does not establish which faction is correct. It refuses to let either answer become exclusive merely because one coalition controls the current institution. Two orders emerge, each accountable for the state it chooses to recognize.
Selection follows, but selection should not be confused with justification. Participants move, builders choose, markets form, and one branch may become dominant. Those acts reveal commitment and determine practical survival. They can also reflect wealth, inertia, defaults, and control of infrastructure. A popular branch may be unjust. A principled branch may fail. Constitutional legitimacy cannot be reduced to market capitalization without turning every successful power into a rightful one.
The legitimacy of a fork instead depends on the path to it. Were less destructive mechanisms genuinely available and used? Was the disagreement constitutional rather than opportunistic? Was the divergence point disclosed? Could the minority continue without sabotage or deception? Did the institutions controlling clients, names, bridges, and markets disclose their positions? Could affected participants understand what each branch would do to their existing claims before they were forced to choose?
A fork that bypasses these questions is not redeemed by open-source code. It may be a seizure accomplished through software distribution. A fork that follows the exhaustion of common procedures can preserve freedom where agreement has become impossible.
The right is to continue, not to win.
What Survives the Split
Calling the elements that survive a fork invariants promises too much. A fork can change balances, contract state, governance rules, or the practical meaning of an asset. It can cause one key to control different things on different branches. What survives is not an untouched world but a set of capacities from which alternative worlds can be built.
The first is cryptographic agency. The holder retains the key and can sign for the corresponding address on each compatible branch. The fork cannot extract the secret merely by changing state. Yet the economic and institutional force of that signature depends on what each branch recognizes. The key persists. Its jurisdiction changes.
The second is a common, auditable past up to the divergence point. Each branch can preserve the blocks, transactions, and receipts that preceded the split, making the act of divergence visible rather than allowing one side to rewrite the existence of the other. This shared history matters because a constitutional secession should declare where and how the accounts part. A branch that obscures the divergence or fabricates continuity deprives participants of the information needed to choose.
The third is practical continuity. Here architecture becomes difficult. Code and chain state may be copied, but liquidity, stable assets, oracles, bridges, domain names, user interfaces, developer attention, legal recognition, and social trust do not duplicate automatically. Each supporting institution may select a branch, support both, or withdraw. A fork can preserve a person's private key while rendering the surrounding world in which the key mattered unusable.
The DAO split made this visible immediately. Pre-fork holders possessed claims on both chains. That apparent symmetry arrived with replay risk, duplicated assets, competing names, and institutions deciding which branch their systems would recognize. The ledger created two continuations before the ecosystem had created two coherent worlds.
A serious fork right therefore requires infrastructure. Transactions need replay protection so that an act intended for one branch is not silently repeated on another. Names and identifiers must distinguish the branches. Participants need access to archival state and client software. Credentials and receipts should disclose the branch and rule set under which they were issued. Bridges and asset issuers must state what they will recognize. Migration tools should not be controlled exclusively by the coalition favoring one side.
Minority continuation also requires nonretaliation. The dominant branch should not corrupt the minority's software, conceal information needed to run it, or use common infrastructure to make continuation technically impossible. It need not subsidize the rival order forever, but it should not turn superior coordination into a right to erase dissent.
Even with these protections, exit remains unequal. The well-capitalized participant can run infrastructure, hedge assets across branches, employ counsel, and wait for markets to settle. A smallholder may depend on a hosted wallet whose default was selected elsewhere. A developer may possess the source code and lack the community needed to make the application useful. A person whose identity or livelihood is bound to one ecosystem may have no meaningful ability to inhabit the other.
This inequality cannot be solved by declaring every participant sovereign. It can be reduced through public clients, portable credentials, transparent defaults, common data formats, independent verification services, and migration paths designed before crisis. The aim is not to equalize every consequence of a split, which no architecture can do. It is to prevent technical dependence from making one faction's constitutional choice unavoidable for everyone else.
A fork preserves plural continuation. It does not preserve equal power, equal value, or equal prospects. Any constitution that offers fork rights must say all four things.
The Emergency That Stays
Giorgio Agamben's warning concerns the emergency that stops behaving like an interruption and becomes an ordinary technique of government. The exception no longer suspends the normal order for a bounded purpose. It supplies the normal order's recurring method. 14
Protocols can reach the same condition without soldiers in the street. An emergency multisignature remains active years after launch. A guardian pauses transfers whenever volatility rises. An upgrade council repeatedly invokes security to bypass the slower governance path. A temporary restriction is renewed automatically, while the original evidence grows stale and the extraordinary route becomes easier to use because everyone has grown accustomed to its existence.
These powers may be necessary. A system handling consequential assets would be reckless to pretend that no exploit, compromised key, or implementation failure will ever require rapid intervention. The constitutional danger lies in treating necessity at one moment as proof of authority for every later one.
Exit is not a sufficient discipline. A governance council that invokes emergency powers too often may lose participants, but network effects, tax consequences, locked assets, identity, and dependence can make departure slower than the abuse. By the time economic selection delivers its judgment, the exceptional institution may have consolidated the very advantages that make exit harder.
The emergency record must therefore be cumulative and public. How often was extraordinary authority invoked? By whom? For how long? Which ordinary process was bypassed? What evidence supported the declaration? Which interventions were later found unnecessary or overbroad? Were the same actors declaring, executing, and reviewing? Did the emergency path disappear when the charter said it would?
The answers should be queryable across incidents, because domination can hide in a pattern no single intervention reveals. Each pause may look defensible. A system paused every month under standards only insiders can apply is governed by the pause key, not by the rulebook displayed to users.
Repeated emergency use should also raise rather than lower the burden. Renewal requires new evidence. A second extension should not inherit legitimacy from the first. Longer duration should trigger broader review and stronger rights for those bearing the cost. The officials or delegates who designed the failing ordinary mechanism should not control the postmortem by default.
Normalization deserves the same care as declaration. Returning to rule means more than toggling the system back on. Exceptional permissions must expire, emergency signers lose their temporary authority, altered parameters are either ratified through ordinary governance or restored, affected parties receive an account and any available remedy, and the record remains available for later institutional learning.
The emergency ends only when the power created by the emergency disappears.
The Return to Rule
Schmitt's challenge cannot be answered by pretending that no one decides. Youngstown contains decisions all the way down. The DAO fork depended on decisions by developers, miners, exchanges, applications, and users. The constitutional question is whether one decision can claim the emergency as its own and convert the ability to move first into authority over everything that follows.
A Protocol Republic should answer no.
It does so by separating the exceptional sequence. The trigger is declared in public terms. Stabilization is narrower than final alteration where the facts permit. Review lies outside the institution that benefits from the act. Time limits operate automatically. Renewal requires another decision. Irreversible changes carry a higher burden. Every step produces a receipt. Normalization removes the extraordinary power rather than leaving it dormant behind the ordinary constitution.
None of this eliminates the originating judgment. Someone must decide that an event crosses the emergency threshold, and no prior rule can make that judgment entirely mechanical without moving the penumbra into the trigger. Infinite regress is avoided by accepting the political moment while making its author, evidence, scope, and consequences contestable. The threshold is not outside constitutional scrutiny merely because it cannot be generated by a formula.
Fork rights remain the final safeguard when a common return has become impossible. They deny one coalition the capacity to make its resolution universal, provided the minority possesses the tools and standing needed to continue. Their value lies in preventing final monopoly. Their cost lies in the destruction of a shared world.
The DAO episode does not prove that Ethereum escaped sovereignty. It shows a network in which the sovereign function was divided among actors whose authority depended upon one another and whose decisions could not eliminate an alternative chain. It also shows how defaults, infrastructure, capital, and coordination can make that plurality profoundly unequal. Both findings belong in the constitutional account.
Youngstown supplies the corresponding lesson from public law. The President acted because he believed delay intolerable. The Court did not need to deny the emergency in order to deny the claimed power. The order could preserve itself only if the institution remained able to stop the person who said preservation required it.
An emergency constitution should therefore be judged less by the speed with which it concentrates power than by the reliability with which it gives power back. The exception begins when an institution claims that ordinary rules cannot wait. It ends when the extraordinary authority has expired, the reasons remain inspectable, remedies remain possible, and the ordinary distribution of power can function again.
A republic survives crisis not by avoiding decisions but by refusing to let the decision become the regime.
The next chapter turns to the order to which emergency power must return: many centers, overlapping jurisdictions, and authority that remains bounded in ordinary time.
The Fractal Polis.