A constitution may be amendable and practically impossible to amend. A cooperative may hold democratic votes and depend on a lender it cannot challenge. A public may have the right to organize and lack the time, money, information, or institutional leverage to make its demands consequential.
Each arrangement satisfies a formal test of revisability. None gives the people subject to it effective corrective power. The difference is not a detail of implementation. It is the difference between a right and a remedy.
This essay states what has to be true for correction to be effective, how a claim travels when the cause sits above the deciding body, and what an institution’s ordinary operation does to the capacity for correction over time. It extends the account of case and structural corrigibility and the dependence and capture arguments already in the corpus.
The corrective power chain
Contestability is often treated as a binary: a challenge route exists, or it does not. That test is too weak. Correction runs through five conditions. A break may severely weaken correction, though another mechanism can sometimes compensate.
| Condition | The question | A failure looks like |
|---|---|---|
| Detection | Can people establish what the arrangement is doing? | A black-box model, an unauditable record, or a cost that never appears in any account |
| Standing | Can affected people challenge it without permission from those responsible? | A grievance route that runs through the person being grieved |
| Agenda control | Can they require the underlying problem to be considered? | A process that can correct the case but cannot ask whether the rule is wrong |
| Decision power | Can their intervention alter the rule, authority, or resource allocation responsible? | An advisory committee, a consultation, or a review with no binding effect |
| Enforcement | Can the resulting decision be implemented despite resistance? | A remedy that requires the losing party to execute it, with no consequence for refusal |
The chain is a test, not a score. An institution can be strong at detection and weak at enforcement, or grant standing and deny agenda control. The point is to name the link that fails before calling the arrangement corrigible.
Law VI already separates technical, operational, and institutional reversibility. The chain adds the political conditions that sit between them. A halt control that works is not a correction if the people who need it cannot reach it, if the challenge cannot name the rule, or if the decision can be ignored.
The operational form is MEC-26 Effective corrective power.
Structural reachability
Many failures are experienced at one level and caused at another. A worker contests a scheduling decision. Management explains that staffing limits prevent change. The staffing limits follow from a budget. The budget follows from a financing arrangement. The financing arrangement reflects an investment decision outside the enterprise.
The worker’s complaint may be entirely justified, and no decision-maker inside the organization may hold enough authority to address its cause. A framework that stops at the employing institution turns structural constraint into a personal failure to persuade the right manager.
One remedy is jurisdictional escalation: the claim travels to the body that sets the constraint. That works where a single higher forum exists. It fits political economies much less well. A hospital’s operating constraints may originate at once in reimbursement policy, debt financing, labor markets, procurement contracts, software infrastructure, and professional accreditation. There is no single forum above them.
The stronger requirement is structural reachability: affected people can identify and reach the decision-making arrangements responsible for producing the constraint, including when responsibility is distributed across organizations. The claim does not automatically win. It gains standing at the point where the constraint can actually be changed.
The pattern carries a constraint map: the binding constraint, the decision point that sets it, the standing at each node, the route between them, and the point at which the route stops. Escalation is a special case, where the map has one higher node. MEC-26 Effective corrective power carries the chain and the map.
The rule has a reciprocal limit. Higher-level institutions should not automatically appropriate decisions that people can govern effectively at a lower level. Reachability is for constraints that cannot be reached locally, not a warrant for centralizing everything.
Error is not disagreement
The principle that people an arrangement gets wrong should be able to revise it is persuasive when there is an identifiable mistake. Many political conflicts are not mistakes. A housing policy can benefit tenants and impose real costs on small property owners. An environmental restriction can protect future residents and eliminate present livelihoods. Both sides can have genuine standing, and their claims cannot always be reconciled.
If every burdened group could reverse the decision that burdened it, collective decisions would be impossible. Corrigibility cannot mean a universal veto.
The distinction is between an error and a settled disagreement. An error is a failure against the institution’s own rules, evidence, or stated purpose; it calls for a remedy. A disagreement is a conflict between legitimate claims; it calls for a legitimate procedure, a reasoned decision, and a settlement that does not destroy the losing side’s capacity to challenge the next decision.
That last condition protects the people who lose a decision. The minority does not need to win. It needs to retain the resources, organization, information, and standing to survive the loss and contest what comes next. A settlement that impoverishes, disperses, or disorganizes the losers has not resolved a disagreement; it has removed one side from future politics.
This is why MEC-27 Protected disagreement is a pattern about what the losing side keeps, not about who wins. It also marks the limit of this framework. Corrigibility is a condition of legitimate collective decision-making. It is not a theory of justice, and it does not decide how scarce resources should be distributed.
Non-self-extinguishing authority
A constitutional order can be amended into an order that abolishes independent elections, restricts opposition, and makes further amendment practically impossible. The original decision may be procedurally authorized. Its result destroys the conditions of future correction.
If everything is revisable, can people legitimately revise away the capacity for revision? The framework has reason to say no. That answer does not freeze existing constitutions, property relations, or governing institutions. It distinguishes changing an arrangement from extinguishing the continuing capacity of the people subject to it to participate in changing it.
Law XII states the narrower version: no system may erase the conditions of its own contestability. The wider version is a commitment, not a derivation. It says that no exercise of constituent or delegated authority may permanently eliminate the effective means through which those subject to it can contest and revise its future exercise.
The commitment is normative, and it should be stated as one. It is also incomplete without a theory of which changes count as extinguishing capacity. A change that narrows a particular route may leave other routes intact; a change that looks procedural may remove the material conditions that made the route real. The test is not whether the old form survives. It is whether the people subject to the new form retain effective means of future correction.
The distinction also protects the opposite case. An institution may legitimately cease to exist. What may not disappear with it is the capacity of the people it governed to participate in what comes next. The commitment is to the capacity, not to the institution’s survival.
Independent corrective capacity
An institution can become harder to challenge without formally restricting contestation. It can develop skills useful mainly inside its own arrangement, make livelihoods depend on its continued operation, define which grievances are recognizable, reward adaptation to its categories of success, and make alternatives difficult to imagine, organize, or finance.
The result can be fewer complaints, higher satisfaction, better measured performance, and less capacity for transformative change. A conventional corrigibility assessment can mistake that for success. The absence of opposition is not evidence of legitimacy.
The test is therefore not only whether people successfully challenge existing arrangements. It is whether they retain the developed capabilities, independent resources, collective organizations, and practical alternatives needed to challenge them. This is the dynamic version of the consumption of adaptive capacity: an institution’s ordinary operation can consume the conditions of its own future correction.
MEC-28 Reproductive conditions treats those conditions as an object of governance. The measure is difficult, and an independent evaluator can impose a contested judgment about what people ought to want. The pattern therefore stays with observable conditions of agency — skills, resources, organizations, alternatives, and the time to use them — rather than claiming access to anyone’s authentic preferences.
Two standards follow, and they should not be collapsed. Institutional non-depletion: an institution should not systematically undermine the independent resources and capacities through which people can challenge it. Systemic reproduction: the wider institutional order should sustain the capabilities, resources, and countervailing organizations needed for effective collective self-government. A workplace cannot secure housing, healthcare, legal aid, independent media, and alternative employment on its own. Requiring it to do so would produce the comprehensive dependence the framework criticizes. The institution is responsible for not depleting; the wider order is responsible for reproducing.
The capacity inventory is not a measurement until it compares plausible alternatives. The research design is longitudinal: track the capabilities, their distribution, the successful challenges that changed governing authority, and the counterfactual — what would have been available under a different arrangement? A single indicator cannot answer the question. Union membership can fall while income, mobility, and access to legal assistance rise; participation can rise while the only channels are controlled by the incumbent. The question is whether people gain or lose the capacity to challenge, and whether the institution makes the next challenge easier or harder.
The cost of correction
Automation is usually adopted to raise throughput while holding verification labor flat. If every increment in consequential agency required a linear increment in human review, the economics of the deployment would fail, and an institution under competitive pressure would hollow out the safeguards first. That is the throughput-verification asymmetry, and it is a real design constraint rather than a failure of will.
The Ethotechnical invariant compares the rate at which agency accumulates with the rate at which the capacity to answer it does. It does not require a linear ombudsman army. It requires capacity proportional to consequential agency, and four levers can restore the proportion:
- Scope. Narrow what the system is allowed to decide. A tool that ranks cases for review delegates less than a tool that closes them.
- Autonomy. Lower the level of authority. A recommendation that a person must act on is a different delegation from an automatic denial.
- Volume. Reduce the number of consequential decisions, even if the system keeps processing the low-stakes ones.
- Verification. Automate inspection, challenge intake, and reconstruction, so the capacity grows without growing the labor linearly.
An institution that pulls none of these levers and instead turns its safeguards into a new layer of paperwork has not met the invariant. It has renamed the gap. The economic question is not whether the institution can afford to verify everything. It is which decisions are worth delegating at the capacity the institution can actually maintain.
The corpus already describes the accounting failure this creates: the antihuman metric shows a dashboard improving while the people behind it run down, and what outcomes hide shows the production path disappearing from the terminal measure. MEC-28 Reproductive conditions records the capacity; this section adds the constraint that the capacity has to be bought by changing the delegation, not by pretending the review is free. The queue arithmetic is in Capacity planning for consequential decisions, and MEC-30 Active epistemic scaffolding carries the interface patterns that raise fidelity without adding passive friction.
The correction window
Correction is usually imagined as a feedback loop: an arrangement acts, an error is discovered, and the arrangement changes. Institutional consequences do not necessarily remain reversible while correction is organized. An ecosystem can collapse. A community can be displaced. A capability can disappear. A technical standard can create dependencies that make replacement prohibitively expensive.
An arrangement can be fully amendable in principle and effectively irreversible in practice. The relevant question is temporal: can affected people recognize the harm, organize a challenge, and secure a response before the consequences become irreversible or the cost of correction becomes prohibitive?
The answer is not a universal veto against irreversible choices. It is a distinction among the gravity of the potential loss, the availability of alternatives, and the distribution of the risk. Where the loss is severe and the alternative is available, the burden of justification shifts toward precaution, staged implementation, reversible experimentation, and protection for people who cannot yet participate, including future generations.
Technical infrastructure is constitutional
Software architectures, protocols, identity systems, data models, and platforms determine who can do what, which actions are observable, and what kinds of change remain feasible. A political community can democratically control an organization and depend on proprietary systems that make meaningful revision impractical.
A municipality may hold the legal authority to redesign a benefits system and lack the ability to inspect, modify, migrate, or replace the software through which benefits are administered. Formally, the rule is revisable. Operationally, its implementation is not.
Where technical infrastructure materially governs people’s lives, correction requires more than a legal route. It requires the ability to inspect the system, challenge its outputs, modify or replace it, and leave it without unacceptable loss. Inspectability, portability, and interoperability reduce dependence. Forkability can enable an alternative, but it is not equivalent to the power to govern: a watershed, a municipal grid, or a social insurance system cannot be forked without undermining the collective good, and privileged groups can exit more easily than disadvantaged ones. The priority is credible collective control over essential infrastructure, with exit and alternatives where they do not intensify inequality.
Technical capacity is not democratic authority, and democratic authority does not supply technical capacity. The pattern has to hold both.
When the rule cannot be reconstructed
The framework assumes that a decision can be traced to a rule that can be inspected and amended. That assumption is safe for a deterministic expert system or a supervised classifier. It is not safe for a multi-agent chain with stochastic sampling, dynamic tool use, and long-context reasoning. The path to the decision is a trajectory, not a policy line. A post-hoc explanation may be a rationalization rather than a reconstruction.
If neither the affected person nor the engineering team can isolate what would have changed the outcome, the institution cannot correct the rule. It can only adjust a prompt, retrain a checkpoint, or add another guardrail, which is the ad-hoc patching the framework exists to distinguish from structural correction.
The answer is not to demand mechanistic interpretability as a precondition. It is to require a reconstructable decision record or a lower level of autonomy. The record has to carry enough to answer a counterfactual question: what input, tool result, model version, threshold, or human action would have produced a different outcome? If the record cannot answer that, the system is not contestable at that level of delegation. The institution can narrow the decision class, lower the autonomy, or stop.
This is a test of the record, not of the model’s internal representations. A system can be opaque and still produce a record adequate to correction; a system can be simple and still fail the test because nobody recorded the boundary that mattered. Law VIII already says an observation that cannot change the governing state is theater. The reconstructability test adds the prior question: can the observation establish what would have changed the state?
MEC-29 Reconstructable delegation carries the record, the counterfactual test, and the autonomy reduction that follows a failed test.
Accountable transition
When an institution must be replaced, the people who depend on it do not disappear with it. An institution can reproduce domination and also support livelihoods, services, skills, and communities. Removing it can be justified without making every consequence of removal just.
The transition is part of the correction. It should be evaluated by how it allocates losses, preserves essential capacities, sustains alternative arrangements, and gives affected people authority over implementation. This is not an argument for preserving harmful institutions indefinitely. It is an argument against treating removal as sufficient evidence of emancipation.
The distinction the pattern has to keep is between justified accommodation and an incumbent’s attempt to preserve its position. A transition that protects people is not the same as a transition that protects the institution from change.
The rules of recursion
A recursive process needs more than the instruction to revise. It needs rules for starting, moving between levels, and treating a result as settled. Without them, recursive corrigibility fails in four ways: endless procedural contestation, entrenchment, self-abolition, or formal assurance with no practical consequence.
| Requirement | Political question |
|---|---|
| Entry | Who can initiate a challenge, and on what grounds? |
| Escalation | When must a challenge move from a decision to a rule, an institution, or a wider structure? |
| Authority | Who may resolve the challenge, and what makes that authority legitimate? |
| Closure | When may a decision be provisionally settled despite continuing disagreement? |
| Reopening | Under what circumstances can the settlement be challenged again? |
| Preservation | What prevents the process from abolishing the conditions that make further correction possible? |
Closure without reopening becomes entrenchment. Reopening without closure becomes endless contestation. Preservation without enforcement is a promise. Enforcement without preservation can destroy the capacity it claims to protect.
The rules are a design checklist, not a constitution. They say what a correction process has to answer before it can claim to be recursive.
What corrigibility does not claim
Corrigibility is a necessary condition of democratic self-government. It is not a complete account of justice or human flourishing. A correctable institution can still distribute resources unjustly, impose unacceptable burdens, or perpetuate harmful practices. Its amendability does not vindicate its present arrangements.
The stronger formulation is therefore not that every institution must be revisable. It is that an institution’s claim to legitimacy depends partly on whether those subject to its power possess effective, continuing, and enforceable means to contest and revise the decisions, rules, and governing arrangements that shape their lives. Those means must not merely exist at a given moment. Institutions must not deplete the independent capacities through which people can challenge them, and the wider institutional order must reproduce those capacities over time. The obligation is divided because no single institution can supply the conditions of its own correction.
That is the move from corrigible institutions to the reproduction of democratic self-government. It is also a research question, not a result: under what conditions does an institution’s ordinary operation increase or diminish the capacity of those it governs to correct and transform it?
The public sentence is shorter: the test is whether people can initiate and secure reconsideration of the arrangements that govern them, and whether ordinary operation leaves that capacity stronger or weaker.
The next research agenda is adversarial contestability tooling: claimant-side instruments as cheap, scalable, and inspectable as the decision systems they monitor. The pattern language can specify the record, the standing, and the enforcement route. It does not build the tool. Until that tooling exists, verification remains a slow craft against an instantaneous decision, and the asymmetry this essay describes will keep reasserting itself.