Reconstructable Decisions

A protocol for public authority in the digital state

TLDR;

States increasingly act through systems, but digital systems do not automatically preserve the record, reasoning, and reviewability that paper-based administration once left behind. A decision is not truly accountable just because it is logged. It must be reconstructable from the rule applied, the evidence used, the method of inference, the authority responsible, and the review path available. Reconstructable decisions are the operational form of accountability in the digital state.

---

Modern states increasingly act through systems. Benefits are approved or denied through workflows. Taxes are computed through filings. Licenses are granted through portals. Inspections are triggered by risk scores. Enforcement is shaped by data, models, and digital procedures. In each of these cases, public authority is not disappearing. It is being mediated. The question is not whether the state acts. It is how it acts.

In a paper-based system, the exercise of authority usually left a trail. Files, notes, signatures, and orders created a record that could be inspected after the fact. Courts, auditors, administrators, and citizens could reconstruct how a decision had been made, even if imperfectly.

In digital systems, that property is no longer guaranteed. A system may produce an outcome without producing a usable explanation. A model may influence a decision without leaving a trace of how. Data may change between steps. Rules may be updated silently. Logs may show that something happened, but not how authority was exercised.

This is the problem. A constitutional state requires more than the production of outcomes. It requires that consequential decisions be reviewable, contestable, and accountable. That, in turn, requires something more specific: decisions must be reconstructable.

What is a reconstructable decision?

A decision is reconstructable when it can be reproduced from the elements that governed it at the moment it was made. At a minimum, that means being able to identify:

• the rule that authorized the decision

• the version of that rule that was applied

• the evidence that was used

• the source and state of that evidence at the time

• the method by which inputs were transformed into an outcome

• the authority under which the decision was made

• the time at which it occurred

• the pathway for review, correction, or reversal

A log records that a decision happened. A reconstructable record shows how public power was exercised. That distinction matters. Without reconstructability, oversight becomes guesswork. Courts cannot test legality. Auditors cannot verify consistency. Citizens cannot meaningfully challenge outcomes. Institutions cannot correct errors with confidence because they cannot reliably reproduce the decision that was taken. With reconstructability, the exercise of authority becomes inspectable in a structured way.

The decision record

To make decisions reconstructable, they must be represented in a structured form. A consequential public decision should produce a decision record containing at least the following:

• Rule reference - What rule authorized the decision?

• Rule version - Which version of that rule was applied?

• Evidence sources - Which registries, documents, or datasets were used?

• Evidence snapshot - What exact facts were used at the time of decision?

• Inference method - How were rule and evidence combined to produce the outcome? This may include rule logic, thresholds, matching procedures, scoring methods, or models.

• System or model version - If a system or model was used, which version influenced the outcome?

• Authority - Which official or system acted, under what delegation or authority?

• Timestamp - When was the decision made?

• Outcome - What was decided?

• Review pathway - How can the decision be challenged, reviewed, recomputed, suspended, or reversed?

This is not an attempt to create a new layer of paperwork. It is a way of ensuring that decisions produced through digital systems retain the properties required by administrative law and constitutional governance.

Where discretion moves

In traditional administration, discretion was visible, at least in principle, in the acts of officials.

In digital systems, discretion often moves elsewhere:

• into how rules are encoded

• into how thresholds are set

• into how data conflicts are resolved

• into how cases are ranked or prioritized

• into what counts as a valid match

• into which model outputs are treated as actionable

These are not neutral implementation choices. They shape outcomes.

A person may be excluded not because they are legally ineligible, but because the system failed to recognize them with sufficient confidence. A case may be delayed not because the law requires it, but because it was scored as lower priority. An application may be flagged as suspicious because a model treated inconsistency as risk rather than as poor data quality.

Reconstructable decisions do not eliminate discretion. They make it legible. They allow institutions to ask: which rule was applied, which evidence prevailed, which threshold mattered, which model influenced the outcome, and under whose authority the result was produced. Without that visibility, discretion becomes harder to detect and harder to govern.

A simple example

Consider a benefit eligibility system.

An applicant is denied a transfer.

To determine whether that denial was lawful, fair, and reviewable, the institution should be able to answer:

• which eligibility rule was applied

• which version of that rule

• which registries supplied household and income data

• what exact values were used

• how conflicts between those sources were resolved

• whether matching logic, thresholds, or a model influenced the result

• who was responsible for the decision

• when the decision was made

• how the applicant can challenge it

If the applicant later corrects an error in the underlying data, the system should be able to recompute the decision using the same rule and logic. Otherwise, the applicant is left with a rejection, but no meaningful way to contest it.

At that point, the problem is no longer only administrative error. It is the absence of a reconstructable exercise of public authority.

Contestability as a protocol

Reconstructability is necessary, but not sufficient. A decision must not only be inspectable. It must also be contestable. In digital governance, contestability cannot be an afterthought. It must be built into the protocol of decision-making.

At a minimum, a contestable system should ensure:

• access to the full decision record

• timeliness of review

• independence of the reviewing authority

• correction of data errors

• recomputation where facts change

• suspension, override, or reversal where warranted

This matters because many harms occur upstream. A risk score may delay a benefit. A classification may trigger additional scrutiny. A queue ranking may deprioritize a legitimate claim. A generated summary may bias the human reviewer before the case is fully read. If these intermediate steps are not contestable, formal rights of appeal may exist only on paper.

Systems as agents of authority

In many domains, systems now act as agents of public authority. If a decision would otherwise have been made by an official, then a system that produces or structures that decision must meet comparable standards of traceability, reasoning, review, and accountability. This does not mean systems must mimic human reasoning. It means they must satisfy the requirements of lawful decision-making. Where systems act, authority must remain visible.

Where this is implemented

Reconstructable decisions do not emerge automatically from good intentions. They have to be specified and implemented across multiple layers. Law and regulation can require traceable and reviewable decision systems. Scheme design can define authoritative evidence sources and decision rules. Procurement can require systems to produce reconstructable records. Technical standards can define common schemas and interfaces. Audit institutions can verify system-level properties. Shared infrastructure can embed these guarantees across applications. In practice, many of the most consequential properties of digital governance are shaped not by law alone, but by procurement, standards, and architecture. That is why reconstructability is not just a technical preference. It is an institutional requirement.

What happens if we do not do this

Without reconstructable decisions, digital governance risks becoming more opaque even as it becomes more data-rich. Outcomes may be generated without intelligible records. Errors may become harder to locate. Vendors may end up governing critical features of accountability by default. Citizens may receive decisions that can neither be understood nor meaningfully challenged. This is not a failure of technology as such. It is a failure to specify how public authority should be exercised through technology.

The protocol layer of the state

The modern state already relies on shared infrastructure. Identity, payments, registries, and data exchange have become foundational. The next step is to recognize that public decisions themselves require a shared protocol. Not to centralize authority, but to ensure that wherever authority is exercised, it remains lawful, visible, reviewable, and correctable.mReconstructable decisions are not a technical detail. They are the operational form of accountability in a digital state.