PROJECT SUMMARY

I want to tell you about a problem that has been following me for years.

An institution can have everything it is supposed to have: human reviewers, appeal procedures, forms, protocols, case numbers, even a telephone number at the bottom of a page saying that somebody will answer.

And still be unable to correct a decision before its consequences become difficult or impossible to reverse.

The existence of a procedure does not prove that a real capacity for correction exists.

Consider a hypothetical but entirely plausible case.

A person receives a notice saying that their housing benefit will be suspended. They believe the decision is wrong and have documents that support their objection. They call the number on the notice and are told to submit a form.

The form enters a system. It receives a reference number. The reference number enters a queue. Eventually, the case reaches someone who can correct a data point but cannot review the criterion that produced the decision. That criterion was established somewhere else, by someone else. The reviewer may understand the problem and still have no authority to stop what is happening.

Months later, a person with sufficient authority reviews the case. The institution acknowledges the error and issues a correction.

But the housing has already been lost. The employment opportunity has passed. The medical care did not arrive when it was needed.

A favorable answer does not necessarily restore what was lost.

This is not necessarily a story about cruelty or negligence. It is about a gap: the gap between detecting a discrepancy and changing the decision that the discrepancy calls into question.

That gap is what I want to study.

The project will have one central practical component: an open assessment matrix. It will reconstruct the path between a signal and a correction. What appeared. How it was read. What remained of it after passing through forms, categories, and levels of decision-making. How far it travelled. Who could actually act. And how much time remained before correction stopped being useful.

The matrix will not certify systems or produce a universal score. It will try to make visible something that often remains hidden: whether an institution retains an effective capacity to correct its decisions, or whether that capacity exists only in its formal procedures.

WHAT ARE THIS PROJECT’S GOALS, AND HOW WILL YOU ACHIEVE THEM?

I want to turn two ideas into a method that other people can use, discuss, and test.

The first is corrective reachability.

It is not enough for an objection to exist. It is not enough for it to be recorded or formally admitted. To produce a correction, an objection must preserve what it was questioning as it travels through an organization.

It must also reach a person—or a combination of people, rules, and procedures—with a real capacity to change the decision.

Someone may be called a reviewer and still have no power to change anything important. They may correct a data point without being able to review the criterion. They may recommend a change without being able to suspend its consequences. They may recognize the error and still have no path for carrying it to the point where institutional action is determined.

The second idea is timely corrigibility.

A decision is not correctable merely because it can be reviewed in theory. What also matters is how long that review takes and how much time remains before the consequences become practically irreversible.

An objection may be true. It may be well-founded. It may eventually be accepted. But if it arrives after it can no longer change what it needed to change, its corrective capacity has been reduced or has disappeared.

Time is not external to correction. It is part of it.

During the six months of the project, I will:

• Define the indicators and conditions for applying the framework.

• Develop an assessment matrix and a guide for using it.

• Reconstruct between three and five documented cases.

• Identify failures of detection, translation, authority, implementation, and timing.

• Submit the framework to independent criticism.

• Publish a preliminary protocol that does not claim to certify systems.

• Prepare the conditions for a possible later phase with a suitable institution.

I do not want the project to appear at the end as though it had moved forward without friction. The process will be public.

During the first month, I will publish the initial definitions, the research protocol, and the criteria for selecting cases. In the second month, the first version of the matrix will be available.

Months three and four will be devoted to reconstructing and comparing the cases. During the fifth month, I will incorporate the external reviews and produce a second version of the matrix. In the sixth month, I will publish the research paper, the public guide, the preliminary protocol, and a proposal for a later phase.

I am not interested in preserving the framework merely because it can be described clearly. It will make sense only if it reveals a difference that other approaches leave less visible.

I am also not interested in adding new names to problems that have already been adequately described. If these ideas do not change the diagnosis of a case, the comparison between cases, or the intervention that should be recommended relative to approaches such as human oversight, contestability, explainability, or AI risk management, they will have to be reduced, integrated into existing concepts, or abandoned.

WHERE THIS PROJECT COMES FROM

I want to explain where this project comes from.

I did not arrive at this question through reading alone. I arrived at it through a kind of attention that is difficult to describe in a funding proposal: the attention that notices when something has been described too cleanly.

A form says “additional documentation required,” but the phrase may hide the fact that nobody with sufficient authority will read the documentation. A protocol says that an appeal is possible, but the appeal may have nowhere effective to go.

I have spent years thinking about what happens when an action continues to be guided by a model that is no longer sufficient to understand its consequences. Above all, I have been trying to understand what conditions make it possible to revise that model before correction comes too late.

This is not only a question about AI. It is a question about living systems, organizations, and institutions. It concerns the difference between a procedure that exists and a capacity that functions.

The conceptual work already exists. I am seeking support to turn part of that work into an instrument and expose it to a more concrete test.

HOW WILL THIS FUNDING BE USED?

With USD 15,000, the project can be completed on a smaller scale without being left halfway. This would not be an advance toward a project that would make sense only if more funding appeared later.

The budget would be distributed as follows:

• USD 9,000 for protected research time: conceptual development, protocol design, reconstruction of three cases, analysis, and writing.

• USD 1,500 for two independent external reviewers.

• USD 1,500 for case research and documentation.

• USD 1,500 for English-language editing and preparation of publications.

• USD 1,000 for open publication and accessible design of the materials.

• USD 500 for research tools, reference management, and basic administration.

With USD 15,000, I will deliver an operational matrix, an application guide, three documented cases, two external reviews with responses, an open working paper, a public summary, and a preliminary non-certifying protocol.

The reviewers will not be paid to endorse the project. Their task will be to look for weaknesses in the framework, identify unnecessary concepts, and test whether the matrix can be used by someone other than me.

If the result is negative—if the framework reveals no relevant difference—that result will also be published.

With USD 25,000, I will expand the number and depth of the cases, the external reviews, and the public materials.

Funding of up to USD 30,000 will add an organized search for partners and a feasibility assessment for a future institutional pilot. It will not yet fund a complete institutional implementation.

The matrix, guide, and protocol will be published under a CC BY 4.0 licence. If another funder participates, no activity will be charged twice. Each contribution will be associated with different outputs or extensions.

WHO IS ON YOUR TEAM, AND WHAT IS YOUR TRACK RECORD ON SIMILAR PROJECTS?

I am Andrés Palladino, an independent researcher and writer living in Tromsø, Norway.

My manuscript Corrigibility and Continuity: A Temporal Hypothesis about Life and Mind is currently under consideration at Biology and Philosophy.

A second manuscript, When Information Cannot Correct: Corrective Reachability in Distributed Organizations, was submitted to Social Epistemology.

I do not present these submissions as accepted publications or as academic validation that does not yet exist. They demonstrate something more limited but important for this proposal: the conceptual program existed before I sought funding and was developed independently of it.

I will carry out the conceptual development, case reconstruction, writing, and general coordination.

The two independent external reviewers will have a critical role. They will be asked to identify ambiguities, redundancies, methodological problems, and limits of application.

I do not need them to confirm the theory. I need them to try to find the places where it can fail.

A later pilot will require a suitable partner, institutional access, defined ethical conditions, and a specific protocol. This project does not assume that such a partner already exists, nor does it promise an implementation for which the necessary conditions are not yet in place.

WHAT ARE THE MOST LIKELY CAUSES AND CONSEQUENCES IF THIS PROJECT FAILS?

The main risk is that the project produces a new vocabulary but no new difference.

A theory may organize a problem more clearly and still add nothing important. It may also establish distinctions that appear clear on paper but disappear when someone tries to apply them.

I do not believe those results should be hidden. The project must be able to recognize them.

If review time does not change the diagnosis of the cases, the temporal component will have to be reduced or removed.

If corrective integrity and corrective reachability cannot be distinguished with sufficient stability, they will be merged.

If the reviewers cannot use the guide to identify comparable paths, evidence gaps, or points of failure, the definitions and application rules will have to be revised.

If the method produces no practical difference relative to existing frameworks, its redundant components will be abandoned. The negative result will be documented.

There is another important limit: the case reconstructions will depend on publicly available information. If that information does not allow a case to be reconstructed responsibly, the case will be excluded. I will not fill gaps with assumptions that the sources cannot support.

The project will not certify AI systems. It will not build an AI model. It will not produce a universal corrigibility score or attempt to demonstrate causal effects through experiments.

Its purpose is earlier and more limited: to build a falsifiable framework for observing when an institution retains an effective capacity to correct its decisions and when that capacity exists only in its formal procedures.

HOW MUCH MONEY HAVE YOU RAISED IN THE LAST 12 MONTHS, AND FROM WHERE?

This project has received no funding during the last twelve months.

Total funding received: USD 0.