Summary

  • The OECD is consulting on a shared framework for machine-executable representations of enacted law. It says the authoritative legal text remains binding and that interpretation, discretion and individual decisions remain with the institutions responsible for them.
  • A reusable rule set could reduce repeated conversion work, but a link to a statute alone cannot tell a person which source version, interpretation, exception, approval and deployed software version shaped a service result.

The answer appears before the record

A public-service checker returns a negative result. The screen says that an applicant does not qualify, or that a filing needs another document. The response may be fast, consistent and generated by software. Yet the people who need to understand it still face a basic question: what rule did the system apply?

The service may point to a statute. That is a start, not a complete answer. The statute may have changed after the software rule was written. A transitional provision may apply only to applications filed before a date. A reviewer may have interpreted an exception. A vendor may have deployed a newer rule bundle to one service while another still runs an earlier release. If the record ends at a link to the law, the person cannot tell which of those facts produced the result.

This gap is at the centre of the OECD’s 2026 consultation on a shared reference framework for Law as Code. Its definition is deliberately bounded: authoritative law already in force is transformed into machine-executable representations, linked to its source and made available as shared public infrastructure. The OECD says the legal text remains binding. The code does not change law or interpretation, reassign institutional responsibilities, or decide individual cases. Discretion and evaluation should remain visible rather than being silently reduced to a fixed condition.

That distinction avoids one mistake, but it creates a more operational one. If software is not the source of legal authority, the institutions using it still need to account for the path between source and execution. A citizen’s remedy may depend on the enacted text, while the immediate service outcome depends on the representation and release the agency actually ran.

One rule, several versions of the truth

The OECD argues that public bodies often translate legal logic separately for each digital system. It says that repetition can create duplication, inconsistent implementation, limited transparency and interoperability, and dependence on individual systems or providers. A shared reference framework could make reuse easier. That is a prospective rationale, not a reported cost study or a claim that a common code base has already delivered savings.

Reuse moves work as well as reducing it. Someone must decide how a provision is represented, review the interpretation, publish a version, tell consuming services that it changed and explain what happens to decisions already made. If each department builds its own logic, each bears the cost of that work. If many departments rely on one shared representation, the maintenance function becomes more central and a stale or mistaken update can travel farther.

The distinction between identifying law and executing it is useful here. The European Legislation Identifier gives legal documents stable, machine-readable identifiers, structured metadata and an exchange format. It helps systems find and synchronize legal materials. It does not encode an authoritative interpretation of those materials into executable rules. Identity and execution are separate jobs; a common identifier is necessary for tracing a source, but it does not prove that a software rule reflects the source correctly.

Australia’s GovCMS announced in August that its Rules as Code capability was available on its platform. The first production example it named is an eligibility checker for whether an agency can onboard to GovCMS; the rule set is open source. That is a useful, limited illustration of a rule tool operating in production. It is not evidence that an enacted statute is being authoritatively executed for members of the public across government. GovCMS itself says its tools do not give legal advice, make binding determinations or replace human judgement where discretion is required.

The encoded rules should be read alongside the authoritative source.

Canada’s Digital Service described a mining-application prototype and raised a more ordinary but critical problem: regulations can change while a tool retains an old version. That is not proof of a live failure. It shows why a rule representation needs an owner and an update path, even when the underlying logic is straightforward.

What a defensible lineage would contain

The minimum useful record is more than a citation. For each rule, it should identify the jurisdiction and authoritative instrument, the exact provision and text version, its effective dates, the services and circumstances in scope, and the exceptions or human decisions the representation does not settle. It should also identify the executable release, the reviewer and authority that approved publication, and the version a specific service used at a specific time.

Those links answer different questions. A legal citation says where a rule came from. A version and effective period establish which law was in force. Scope and exceptions show whether the rule covered the person’s case. An interpretation record explains the choices needed to turn legal language into conditions. A release record shows who allowed the machine-readable representation to be used. A deployment record connects that approval to the service that delivered an outcome.

No single record can settle every dispute. Courts may remain the forum for contested legal meaning; agencies may retain discretion; individual services may have their own review processes. But the record should make those boundaries findable. It should distinguish an automated calculation from a human assessment, identify when the system cannot decide, and give people a route to request reasons or correction. Otherwise, the authority remains formally clear while practical control shifts to whichever code path is hardest to inspect.

Testing also has two levels. A technical test can show that the same input produces the same output. A legal test asks whether the encoded condition tracks the authoritative text, including exceptions, amendment dates and discretion. Both matter, and neither should be presented as a substitute for the other. When the policy is ambiguous, encoding can expose the ambiguity; it cannot resolve it without an authorised interpretation.

A framework will distribute responsibility

The OECD consultation does not prescribe one platform or institutional model. It says successful implementation needs cooperation across legal, technical, administrative, political, academic, economic and civil-society roles, with a constitutionally sound allocation of responsibility. The open question is how a common reference framework can make those responsibilities visible without pretending every legal system assigns them in the same way.

The consultation’s first phase, which closes on 16 October 2026, maps existing initiatives. A later phase is planned to gather principles and requirements, followed by collaborative development of a draft framework in 2027. The final publication is expected in late 2027 or early 2028. As of 14 October, those later steps are plans, not findings. The current deadline is a chance to map what exists; it is not approval of a standard.

For agencies, the practical test is whether a person can reconstruct the path from a legal source to the result on a screen. For procurement teams, it is whether a reusable rule set has a named maintainer, release authority, exit option and budget for updates. For oversight bodies, it is whether they can compare the approved version with the deployed one. For the public, it is whether a rule can be challenged without needing to reverse-engineer a service.

Law as Code can make the translation from legal language to digital services more reusable and inspectable. That benefit depends on treating the executable representation as governed infrastructure with a lifecycle, while keeping the law and the institutions empowered to interpret it in view. A statute link establishes ancestry. Accountability begins when the record also shows what was transformed, who approved it, where it ran and how it can be corrected.

Sources