The Decision That Was Never Made: Remedying a Ratified Decision in Agentic Administration
METHODS · WORKING PAPER — Governance. The third paper of the Agentic Government programme. Analytical, grounded in dated law as of July 2026; builds on Who Acted, Under Whose Authority? and Assisting or Deciding? This is analysis of governance approaches, not legal advice: how a ratified decision should be treated as void, voidable, or remediable varies by jurisdiction and forum.
When a decision about a citizen is found to have been ratified rather than genuinely decided — signed by an official who could not reconstruct, comprehend, or realistically depart from what a chain of agents produced — the signature does not cure the defect. It conceals it. So the question this paper answers is what should actually follow. In short: the affected person is owed a genuine re-decision, not another signature; where the failure was structural rather than a single lapse, the chain’s standing to make that class of decision should be suspended until it can support genuine deciding; and the flawed decision does not stand merely because a human name is on it. A ratified decision is, in substance, a decision the law did not authorise — and a remedy has to treat it as one.
This is the third of three linked problems. The first paper asked whether a delegated decision can be traced back to an accountable human. The second asked, where a human is present, whether they genuinely decided or only ratified. This paper asks what is owed when the answer is ratified — and finds that the remedy loops back to admissibility: the mechanism for withdrawing a system’s standing is the same one that grants it.
Why does a ratified decision need a remedy at all?
Because in substance it is a solely automated decision wearing a human signature — and the law does not treat those as interchangeable with human ones. The protection against being subject to a decision “based solely on automated processing” turns on whether human involvement was meaningful; a signature that ratifies an untraceable determination is precisely the token involvement the protection exists to exclude. In public administration the defect is compounded: the duty to give reasons cannot be satisfied by a human who cannot state the basis, and the right to an effective remedy is empty if the decision the citizen is challenging was never really made by the person credited with it. The signature does not add a human decision to the process. It removes the visibility of the fact that no human decision occurred.
So the remedy is not a courtesy. It corrects a decision that failed a legal precondition — that a competent human actually decide — while appearing to have satisfied it. Left uncorrected, the failure is worse than an openly automated decision, because it is an automated decision that has been laundered through a human to look accountable.
Is the decision void, or voidable?
This is where honest analysis has to hold options open, because the answer depends on how a given legal regime treats the human-decision requirement — and jurisdictions differ.
If the requirement of a genuine human decision is treated as a procedural safeguard, a ratified decision is voidable: defective, not automatically null, and curable by a proper re-decision — but it cannot simply stand once the defect is identified. If the requirement is treated as a condition of the decision-maker’s competence or authority — the decision had to be made by a human official, and in substance it was made by agents — then the decision is closer to void: made without the authority the law required, and of no effect from the outset. The Institute’s analytical lean is that for decisions the law reserves to a human decision-maker, a ratified decision is at minimum voidable and must not be allowed to stand, and is void where the human decision was a genuine legal precondition rather than a formality. But which applies in a given case is a question for the relevant law and forum, not one this paper can settle by assertion. What it can say is the narrower structural point: a ratified decision is never simply valid. The signature does not save it.
The individual remedy: a genuine re-decision, not another signature
The person affected is owed a re-decision by a human who genuinely decides — meaning one who satisfies the conditions the second paper set out: they can reconstruct and comprehend the basis, have the time to consider it, can reach a different outcome, and engage rather than default. A re-decision by another ratifying human is no remedy; it repeats the failure with a fresh signature.
This has a hard consequence that follows directly from the first paper’s rule. A genuine re-decision requires a reconstructable basis — a delegation record. Where that record exists, the case can be re-decided on it. Where it does not, the decision cannot be cured by review at all, because there is nothing to review; the agent-produced basis is unusable, and the case must be decided afresh or set aside. No record, no standing applies to remedies too: a decision no one can reconstruct cannot be repaired by a human looking back at it, only re-made. This is not a technicality. It means an agency that failed to keep the delegation record has not merely made a traceability error — it has forfeited the ability to cure its own ratified decisions by any means short of deciding them again from the beginning.
The systemic remedy: when re-deciding one case fixes nothing
A ratification can be an isolated lapse — a comprehensible case on a quiet day where the official happened not to engage — or it can be structural: a chain, or an operating condition, under which no official could genuinely decide. The second paper’s signals distinguish them: an override rate that never diverges, a decision-to-time ratio past the point where comprehension is possible, concurrence that is never recorded as engaged. Where the finding is structural, re-deciding the litigated case is necessary but changes nothing systemic — the chain will keep producing ratifications, and the population of past decisions is suspect, not just the one that surfaced.
Here the remedy has to reach the deployment, and the mechanism is admissibility applied in reverse. Standing granted to a chain can be withdrawn on evidence that the deployment structurally prevents genuine human deciding. The systemic remedies are the four admissibility outcomes, applied to a system already in service:
- Hold — suspend the chain’s standing for this class of decision until it is shown that an official can genuinely decide within it (the delegation record exists, the decision-to-time ratio permits comprehension, override is real and protected).
- Admit-with-Limits — return the chain to service only through a narrowed scope: lower volume per official, simpler cases, or advisory-only output that cannot be ratified without independent human construction of the basis.
- Refuse — withdraw the chain from this decision class entirely, where it cannot be redesigned to support genuine deciding.
And a structural finding raises a duty the individual case does not: because the failure is a property of the system, the affected population — not only the complainant — was subject to it, which points toward a systemic review of comparable past decisions. How far that duty reaches is genuinely hard and unresolved; that it exists in some form, where the failure is structural, is difficult to avoid.
A remedy framework
The remedy is set by two things: whether the delegation record exists, and whether the failure was isolated or structural.
| Finding | Owed to the person | Owed by the system |
|---|---|---|
| Isolated ratification, record exists | Genuine re-decision on the reconstructed basis | Note the lapse; monitor for pattern |
| Isolated ratification, no delegation record | Decision cannot be cured by review — set aside and decide afresh | Restore the record before further decisions of this class |
| Structural ratification | Genuine re-decision (afresh, if no record) | Withdraw standing (Hold); redesign to support genuine deciding, or Refuse; review the affected population |
Harm caused in the interim is a separate question the re-decision does not erase: a decision later re-made does not retroactively become one that was lawfully made when it acted, and existing liability and administrative law govern what follows from that.
Two ways to game a remedy — both already foreclosed
A remedy regime invites two evasions, and the programme’s first two rules close them before this paper adds anything.
The first is to remove the human entirely, so no one can be found to have “only ratified.” But a decision with no human at all is a solely automated decision, already prohibited where it has legal or significant effect. Removing the human does not escape the remedy; it triggers a different and clearer breach.
The second is to stop keeping the delegation record, so ratification cannot be proven. But under no record, no standing, the absence of a reconstructable record is not a shield — it is itself the failure. A chain that cannot show who decided under whose authority has not earned admission to the decision, and its outputs have no standing regardless of what an audit can or cannot prove. Destroying the evidence does not defeat the remedy; it converts a curable ratification into an uncurable one.
The honest tension
A remedy regime must be proportionate, or it becomes its own failure. A single isolated ratification should not trigger the suspension of a whole administrative chain; treating every lapse as structural would paralyse administration and punish the honest keeping of records that made the lapse visible in the first place. This is why the isolated/structural distinction is load-bearing rather than decorative: it scales the remedy to the failure, and it protects the agencies that document their decisions from being worse off than those that do not. Equally, the remedy must not tempt agencies toward the two evasions above — and the safeguard against that is not goodwill but the two prior rules, which make record-keeping and a genuine human the conditions of admission, not optional virtues. The aim is precise: a citizen wronged by a decision no one really made should be made whole, and a system that cannot let anyone really decide should not keep deciding — without turning every documented imperfection into a crisis.
What this paper does not settle
Three questions are left open, and honestly so. First, the void/voidable line is jurisdictional and forum-specific; this paper argues that a ratified decision never simply stands, not that any single doctrinal label applies everywhere. Second, how far a systemic review of past decisions must reach — every comparable decision, a sample, a defined window — is a question of proportionality and evidence that rulemaking, not assertion, should resolve. Third, the standard for interim compensation, where a ratified decision caused harm before it was corrected, sits in existing liability law and is beyond this paper’s scope. The framework for what kind of remedy is owed is meant to be usable now; the precise legal contours of each remedy remain work to do — and are best decided by the relevant institutions, not by an institute.
FAQ
If a human signed the decision, why isn’t it valid?
Because a signature records approval, not deciding. Where an official ratified an agent-produced determination they could not reconstruct or depart from, the decision is, in substance, one the law did not authorise — solely automated, or made without the required human decision-maker. The signature conceals the defect rather than curing it.
What remedy does the affected person get?
A genuine re-decision by a human who actually decides — reconstructing and comprehending the basis, with a real ability to reach a different outcome. Another ratifying signature is not a remedy. And where no delegation record exists to reconstruct, the case cannot be cured by review and must be decided afresh.
What happens to the system, not just the case?
Where the failure was structural — the chain or its operating conditions made genuine deciding impossible — the chain’s standing to make that class of decision should be suspended until it can support it, narrowed, or withdrawn, and comparable past decisions may need review. Re-deciding one case does not fix a system that structurally prevents deciding.
Can an agency avoid this by removing the human, or by not keeping records?
No. A decision with no human is a solely automated decision, already restricted where it has significant effect. And under no record, no standing, the absence of a delegation record is itself a failure, not a defence — it turns a curable ratification into an uncurable one.
