Two people argue about healthcare. One says public, one says private. They bring evidence, examples, two coherent theories of who can be trusted with the money. Neither notices they have skipped a question. Every design they might propose — and the same is true of centralized or decentralized AI, worker-owned firms or investor-owned ones, universal or targeted welfare — is an arrangement in which some institution will hold power over people who depend on it. That arrangement has to answer for itself before either theory of ownership gets a turn.
The question
What must be true of an institution's relationship to the people who depend on it before its exercise of power over them is legitimate?
It is not a question about who should rule. It comes before that, the way a building code comes before an argument about architecture: whichever design wins, the code constrains what it was allowed to build. This archive has spent two hundred essays answering the question in pieces. Put the answers together and they form a grammar — nine constraints that any arrangement of power must satisfy, whatever it is called.
What failure costs
The first three constrain how an institution may account for itself.
Its success is not yet evidence. An institution cannot be credited with working until someone has asked whose compensatory labor, exhaustion, and risk produced the appearance of working. The system runs smoothly because somebody absorbs its failures quietly; the smoothness is the failure, relocated. An account that omits them is not an account of the system but of the system plus whoever is subsidizing it.1
Its blame is not yet justice. Where the institution assigns responsibility, responsibility tracks authority: a person may be answerable for an outcome only to the degree they had the power to alter it. Assigning blame to the participant with the least power disposes of accountability rather than exercising it.2
Its exceptions are not yet noise. The recurring workaround, the appeal, the case the rule mishandles, the human who quietly repairs what the rule broke — these are the visible part of whatever the institution's model of the world systematically fails to represent, and an institution that treats them as noise has chosen blindness.3
What consent counts
The next three constrain how an institution may claim its subjects agreed.
Participation is not consent while refusal is ruin. The fact that people keep using the system is weak evidence of authorization when saying no costs them their income, care, or standing. Where exit is catastrophic, usage measures captivity, and captivity is not a vote.4
A right is not yet a right. Formal entitlements count for what exercising them actually costs. If using the procedure requires time, money, knowledge, and endurance that ordinary people cannot supply, the institution has built a right for people who do not exist and a procedure for everyone else.5
Dependence is not a title. The more people must rely on an institution, the more it owes them — not the freer its hand. Reliance generates obligation. Read as licence, it converts a fact about the subjects' need into a right over the subjects, and the arithmetic runs backwards.6
What correction requires
The last three constrain what an institution must be able to survive: being wrong.
Power must be able to lose. No arrangement eliminates error, so legitimacy cannot depend on getting things right. It depends on what happens after being wrong: whether decisions, rules, and finally the arrangement itself can be revised by the people its errors fall on, or whether the institution may absorb every valid finding and continue unchanged.7
Usefulness does not convert into jurisdiction. However well an institution performs, performance in one domain is not authority over another. Capability, expertise, necessity, dependence, prediction, and previous authorization are all facts about the system. None authorizes. Authority has to be granted across that gap — someone performs the promotion, on the record, and is answerable for it.8
Strength is the point. The destination is strong institutions: strong enough to be depended upon, constrained enough not to convert dependence into domination, and corrigible enough that the people affected by their mistakes do not have to destroy themselves to get them corrected.9
One constraint, promoted
Set the nine against the familiar arguments and a pattern appears that the arguments themselves cannot see. Each ideology's account of legitimacy is one of the nine constraints, promoted to stand for all of them.
The market's story is the fourth constraint: consent. People chose it, and they can leave. Promote consent to the whole grammar and a market with catastrophic exit costs has answered the only question it asks and none of the others, which is why its failures arrive as the fifth and sixth — rights priced out of reach, dependence read as licence — and why they never register as failures at all. The state's story is authorization: the arrangement was voted for. Authorization is a jurisdiction claim, and the eighth constraint declines the promotion; a mandate to govern is not a warrant over every domain the governed depend on. The technocrat's story is accuracy. Accuracy is not corrigibility, so the seventh constraint asks who pays while the experts are wrong and how they get corrected, and the third points at the remainder no model can represent. The localist's story is proximity; proximity is not survivability, and close institutions dominate their members. The cooperative's story is participation, and participation without the authority to alter outcomes is what the second constraint calls disposed accountability. The socialist's story is ownership — the first ledger — and ownership, transferred, leaves the other two ledgers where they were.10
None of these stories is false. Each names a constraint that matters. The error is the promotion: an ideology is what one constraint becomes when it is asked to do the work of nine, and its characteristic failure can be read off in advance, because it is always the constraint the story has no word for. Markets have no word for captivity. States have no word for jurisdiction they were never granted. Experts have no word for being wrong at someone else's expense.
That is also why the arguments do not converge. Each side measures legitimacy in the unit its own constraint supplies — choice, votes, accuracy, closeness, participation, ownership — so each side's evidence is real and beside the point to the other. More evidence of the kind a side counts cannot settle a dispute between partial grammars. Only the full grammar can, and the full grammar belongs to no side.
None of this settles whether healthcare should be public or private. It settles a question that has to be settled first, about whichever one is built: does the arrangement externalize its operating costs onto vulnerable people, convert dependence into command, assign responsibility without authority, make refusal catastrophic, suppress the evidence of its own mistakes, and make correction practically inaccessible? Some institutional forms fail these tests repeatedly, and it shows in the mechanism rather than the slogan. You do not have to stipulate that any arrangement is intrinsically illegitimate. You can trace how it fails and watch.
The grammar
The nine are a grammar because they are prior to the ideologies, the way syntax is prior to the argument. Diagnosis first: follow the burdens, the exit costs, the quiet repairs, the exceptions. Constraint second: an institution whose power over dependents is legitimate must internalize the cost of its correction, pair every responsibility with the authority that earned it, and treat dependence as exposure rather than opportunity. Design third: staffing, redundancy, appeal, reversibility, portability, and limits on automation stop being technical details and become the substance of the politics. Implication last: whichever ownership wins the argument, the tests still apply to it — and apply identically.
The archive's subject is the people underneath the institutions — who depend on them and are governed by them. The question this essay assembles is the one all of it has been asking: how much power can an institution hold over the people who need it before they are entitled to be something more than governable by it?
Nine answers. Not one of them requires you to pick a side first.
Notes
The founding version — the person who becomes the failure, so the system can keep calling itself orderly — is Everyone is a Crumple Zone Now, with the accounting question pressed in This Did Not Have to Cost You This Much.
The principle and its sorting rule: Responsibility Does Not Track Authority.
Exceptions as the readable surface of an unrepresentable remainder: The Metric She Repairs Says She Is Not Needed and Consensus Without Conflict Is a Lie.
Consent measured by the cost of refusal: Refusability is the Future of Design, with the political form of the toothless no in Toothless Ethics: Why Principles Don’t Stop Machines.
Rights priced by the capacities they demand: You Were Free to Choose and The Undo Button You Don't Get to Press.
The five conversions this refusal stops, and the dependence arithmetic: Capability Is Not Permission; the jurisdiction half pressed against healthcare in It Isn't Greed. It's Design..
What separates transparency, participation, and appeal from actual correction: Stop the Machine, Not the Person, Dependence Without Domination, and The Rule Is Never on Trial, with the survival question for corrigible institutions in All the Machinery, None of the Correction.
The one refusal that stops every promotion from observed to authorized: Capability Is Not Permission.
The positive program — durable institutions without sovereignty over their dependents: Domination Is a Design Choice, with the care side in On Inherent Care and the structural-no-coercion side in Designing Systems Where Coercion Is Structurally Impossible.
Ownership rewrites the first ledger only, and domination survives the transfer to better ownership: The Third Distribution; the limits of collective control in Dependence Without Domination.