A model evaluation establishes properties of a model. It cannot establish the alignment of the system that model is deployed into, because that system includes the people who notice what the model got wrong, interpret what it left ambiguous, and absorb what it broke. Where those corrections are routine rather than exceptional, the people making them are not users of the safety architecture. They are part of it.
People who are part of the safety architecture are owed a route into it. That claim is stronger than a general appeal to participation. The companion essay, dependence runs both ways, takes the reverse case: where people cannot leave an institution’s system, the institution’s obligations rise, not its license to direct them.
Safety work without authority
An institution deploys an automated system into a population and then relies on that population to catch what the system misses. A caseworker who quietly re-enters a rejected form. A nurse who double-checks a dose the scheduler set. A customer who calls three times until someone reverses the charge. Each of these people is performing a safety function. The institution’s error rate, as reported, is the rate after they have done their work.
Usually none of them holds any authority. They cannot halt the system, cannot change its parameters, cannot make it reconsider a class of decisions, and often cannot make it acknowledge that the correction happened. The institution has delegated a safety function without delegating the power to exercise it. It asks for judgment and refuses the standing that judgment implies.
The asymmetry is not a matter of degree. An institution can depend on a population’s labor and pay for it; that is an ordinary economic relation. An institution that depends on a population’s corrective capacity and gives it no route into the system’s governing state has put the population in a position where its competence is required and its voice is not. The political theory term for a relation of that shape is domination: one party’s welfare depends on the other’s forbearance, and the dependent party has no recognized means of contesting how the forbearance is exercised.
Corrective labor erases its own evidence
A familiar argument holds that everyone affected by a system should have a say in it. But affectedness is everywhere, and a system that had to consult everyone it touched would never act. Institutions rightly reject that argument.
Corrective labor is a narrower ground, and it survives that rejection. Ordinary productive labor makes a system valuable. Extracting it without a say is a distributive wrong, and distributive wrongs can, in principle, be paid off. Corrective labor does something further. Absorbing a failure competently removes the evidence that it occurred. The metric improves because someone caught it. The institution therefore loses not only the labor but its own capacity to know how the system is performing.
This is a fail-silent state, a failure that no indicator registers, and the people absorbing it are what keeps it silent. Nothing in the institution’s instruments registers the failure, because the failure was handled before the instrument looked. Dependence on correction is self-concealing in a way that dependence on production is not. A factory that runs on unpaid labor at least knows it has a workforce. An institution that runs on unacknowledged correction may sincerely believe its system is healthy, and its dashboards will agree.
People whose corrections the institution depends on hold information the institution cannot otherwise get, and withholding standing from them is not only unjust but epistemically self-defeating. A clean dashboard cannot be read as evidence of health until absorption has been measured separately.
Theory, drawn
A failure fixed by hand leaves no record
Competently absorbing a failure erases the evidence that it occurred. Clean dashboards cannot be read as health until absorption is measured separately.
What standing has to mean
What the institution owes them has to be more than consultation. Being asked for feedback does not change the relation; the institution can still ignore what it hears, and the corrective labor continues regardless. What changes the relation is standing: a recognized route by which the person bearing the error can put that error into the system’s governing state, where it has to be answered and can produce a change.
Standing in this sense is procedural. It does not mean a veto over the system. It means that a challenge, once raised by someone with standing, must be received by an identified responder, answered by a stated deadline against a stated standard, and capable of producing one of a defined set of outcomes: the decision stands, is narrowed, is suspended, is reversed, or the authority that produced it is reconsidered. That is what turns an observation into a control. Without it, the observation is a complaint.
This also says what standing should not track. Standing that attaches to current indispensability rewards staying indispensable, which is the opposite of what a safety architecture wants. Standing should attach to dependence that has been documented, and it should survive the dependence being engineered away. Otherwise the doctrine preserves the fragility it exists to expose.
Standing proportional to reliance
Authority follows demonstrated dependence, so it is proportional rather than unlimited. The institution is not obliged to give everyone who touches the system a voice in its design. It is obliged to give the people whose corrections it relies on a route into its state, with force proportional to how far it relies on them.
That proportionality needs an instrument, and what it measures is not whether the system has users who sometimes complain but how far the system’s reported performance depends on a human subsidy that is not being counted. Where that share is large, the population is doing the institution’s safety work, and the case for standing is strong. Where it is small, the case is weaker and the institution can say so with evidence rather than assertion.
Challenges in volume are a budget line, and institutions fear drowning in them. The essay on challenge density answers with measured response capacity and stratified halt force, not narrower standing.
What the laws take from this
Law IV: every consequential delegation creates a correction obligation. An institution that delegates authority to act has also delegated the capacity to be wrong, and someone will bear it.
Law VII: error-bearing parties require standing proportional to exposure, in the procedural sense described above.
Law VIII: observability without state transition is theater. The point of standing is that what the error-bearing party sees can change what the system does.
The constraints do not depend on the word domination. A system whose failures are absorbed silently is a system whose failure rate its operators do not know, and a nurse who catches a wrong dose has no way to tell them.