The Research Roadmap for Digital Twins of Legislation: 2024 to 2030

Laws are written, reviewed, and promulgated by lawyers. Afterwards, public and private organizations have to adapt their software systems, and they have to do so anew with every amendment, with considerable effort, long delays, and divergent results. The digital twin of legislation starts at exactly this point: administrative law is published as a synchronized, machine-interpretable, deterministic, and certified artifact that yields the same legal conclusion as applying the provisions and explains it by reference to the norms. This article describes the roadmap of the research program I lead on this topic at Johannes Kepler University Linz: the starting point in 2024, the milestones up to 2027, the results that are already available, and the target picture for 2030.

Starting point: the conceptual model in the book

In the book Der selbstfahrende Staat (Springer Gabler 2024, with Patrick Pils and Philipp Seper-Ambros), we described digital twins of legislation as a building block of a state that automates procedures without relinquishing responsibility. The first experiments were sobering and instructive at the same time: XML-based representations of the IT collective agreement missed the automation and cost targets because automated reasoning requires a more comprehensive model. A prototype for part of Austrian income tax law was built as a basic digital twin, and the Upper Austrian landscape levy (Landschaftsabgabe) on the surface extraction of mineral raw materials was implemented as a digital twin in Prolog.

From this preliminary work emerged the research hypothesis of the program, namely that "representing administrative law as a synchronised, machine-interpretable, deterministic and certified artefact can preserve rule-of-law guarantees while reducing the administrative burden of implementation."

Why language models alone are not enough

The motivation is threefold. First, laws are published exclusively as natural language, and every authority, every vendor, and every business interprets and codes the same provisions all over again. Second, this decentralized translation is redundant and inconsistent: it duplicates effort, produces divergent results, and is expensive to maintain with every amendment. Third, language models alone are not a solution. Blair-Stanek and Van Durme showed at ICAIL 2025 that large language models give unstable and frequently wrong answers to legal questions and cannot fulfill rule-of-law obligations on their own.

A demonstration from the program makes the difference tangible. Without a connection to a twin, a language model guesses a business's tourism levy, for instance with an invented rate of 0.45 percent and a result of EUR 2,250. With a connection via the Model Context Protocol, the twin computes the correct amount of EUR 2,000, and the model is left with only the explanation.

Method and conditions

The program follows the Design Science Research methodology of Peffers et al. (2007): the artifact is designed, built, and evaluated. The requirements come from a Straussian Grounded Theory study with nine senior experts, analyzed through open, axial, and selective coding until theoretical saturation. This yielded four feasibility conditions and four design principles:

• Selective automation. Only procedures with deterministic logic and little discretion are automated; personal judgment remains with the decision-makers.
• Semantic and ontological clarity. A uniform, machine-readable legal vocabulary ensures clear terminology, unambiguous references, and interoperability.
• Computable legal structures. Laws are transferred into modular, executable structures that are maintained in parallel with the text.
• Institutional and procedural reforms. The legislative process becomes iterative and multidisciplinary; every change to the text is reflected in the digital representation.

Three propositions summarize the conditions for automated decisions that comply with the rule of law: semantic alignment is a necessary condition. Computable structures improve feasibility only if governance capacity supports synchronized updating. And automation is viable only for deterministic, parameterizable sub-decisions.

The architecture: four layers, one platform

A Digital Twin of Administrative Law (DTAL) is a continuously synchronized, machine-interpretable representation of administrative law that captures the normative content of statutes, regulations, delegated acts, and administrative guidelines at both the descriptive level (concepts, definitions, relations) and the decision level (eligibility checks, parameterized computations, outcomes). Given complete and correct case data, it yields the same legally relevant conclusion that would follow from applying the provisions and explains it by reference to the governing norms.

Technically, each twin consists of four layers, inspired by the structure of smart contracts: text (the unaltered statutory text), ontology (a knowledge graph in OWL/RDF with stakeholders, legal objects, and relationships), configuration (axioms and interface definitions, such as income thresholds and tax rates), and logic (calculation rules and decision rules including exceptions, in Python). A Digital Execution System hosts any number of twins and exposes them via REST to e-government applications and enterprise software, and via the Model Context Protocol to agentic AI. The platform is deterministic, fully traceable, synchronized with amendments, and designed as open source.

What this looks like in practice is shown by the platform lawdigitaltwin.com: there, available twins can be searched, executed in a form, and integrated via REST or MCP. The calculator for the Upper Austrian tourism levy determines the contribution, including a breakdown, from municipality, business activity, and revenue.

Three generations of the pipeline

Within the program, the translation of statutory text into a twin passes through three generations:

1. Manual translation. Experts read the law and code the rules by hand. Accuracy is high, the effort enormous: a civil law corpus required around 900 person-days, and the work is repeated in every organization.
2. The language model proposes, humans review. This is the current state. A language model drafts the artifacts in six staged steps; experts validate and approve each of them. The pipeline exists as a research prototype and has not yet been released to third parties.
3. Multi-agent pipeline. The target picture: an orchestrator with a planner, shared memory, and tool routing coordinates specialized agents that segment laws, model the ontology, externalize rates and thresholds into the configuration, synthesize logic and tests, verify against reference data, and establish traceability via ELI and Akoma Ntoso anchors. Legal and domain experts approve, correct, or reject at approval gates. The published bundle of text, ontology, configuration, logic, and tests is stored under version control and exposed via API and MCP.

The state of the third generation, the agentic pipeline with its benchmark on the municipal tax (Kommunalsteuer), is described in the article Digital Twins of Legislation: Where the Research Stands Today.

Milestones 2024 to 2027

The program is structured as a series of research projects with a portfolio of use cases that test one architecture against legal norms of different reach and structure.

2025: the tourism levy. The validated and published use case is the Upper Austrian tourism levy with around 29,000 affected businesses. The twin models section 43 with its percentages per contribution group and municipality class and the minimum contributions; the ontology links classes such as tourism business, municipality, and contribution group to the Legal Information System (RIS) via ELI links; and the logic computes the contribution from revenue, municipality class, and contribution group.

2026: ontologies, pipeline, municipal tax. In parallel with the evaluation, a systematic mapping study on legal ontologies from 2017 to 2025 (with Alexander Hipfl and Christoph Schütz) is under way; it analyzes 54 primary studies from ACM, IEEE, Scopus, and Springer and lays the foundation for building our own Legal Ontology. The feasibility of integrating existing legal ontologies is being assessed on this basis. The municipal tax under the KommStG 1993 was chosen as the next anchor case: a nationwide rate, collection by municipalities, practically every Austrian employer affected, highly deterministic and parameterizable, with an apportionment across several municipalities that deliberately stresses the architecture. The research partners are the Federal Chancellery, the City of Linz (Office of the Magistrate Director), Johannes Kepler University, and Binary Growth GmbH. Evaluation is carried out with expert groups and against real data from the City of Linz.

2027: industrialization. Planned are the development of an open-source multi-agent pipeline, its evaluation on the municipal tax case, guidelines and process recommendations for building digital twins of legislation, and the transformation of entire bundles of laws with the agentic pipeline. The question is no longer whether deterministic administrative law can be twinned, but how to industrialize the translation: at scale, under the rule of law, and natively consumable by autonomous AI agents.

Results already available

The burden study on the tourism levy (JURISIN 2026) compares five implementation architectures in person-days:

Central publication of a machine-executable twin, generated by a semi-automated pipeline, reduces the administrative burden of implementing legislation roughly 35-fold compared with decentralized implementation, with better correctness, traceability, and reproducibility. A purely language-model-based implementation without safeguards was not evaluated because it fails to meet the normative requirements.

For the municipal tax, the agentic pipeline is currently being evaluated with a source-traceable benchmark; the results will appear with the scientific publication. The research questions and the design of the evaluation are described in the article on the state of the research.

The road to 2030

Beyond the project horizon of 2027, the program pursues a target picture in three stages:

• 2027 to 2029: selection of suitable legal sources and agentic pipeline. For each jurisdiction, the legal sources suitable for twinning are identified and translated with the agentic pipeline. Several governmental prototypes are created every year, including internationally.
• 2028: twins as annexes to laws. The first digital twins of legislation are published outside research as annexes to laws. The twin thereby becomes part of the promulgation, no longer merely a replica of it.
• 2030: legally effective twins. The first legal texts are implemented as legally effective digital twins of legislation. The definitions that are extracted after the fact today will then be created in the legislative process itself.

The prerequisites for this are laid out in the four design principles: selective automation, semantic clarity, computable structures, and a legislative process that keeps text and twin in sync. The technical side is prepared with the pipeline and the platform; the institutional side is the subject of the guidelines to be developed in 2027.

Publications of the program

• Schnitzhofer, Pils, Seper-Ambros (2024): Der selbstfahrende Staat. Ein Denkmodell für das Zusammenleben im Staat der Zukunft, Springer Gabler.
• Schnitzhofer (2025): Towards the Design of Digital Twins for Tax Law and The Self-Driving State: Automated Decision-Making in Modern Governance, IRIS 2025, Vienna.
• Schnitzhofer (2025): Applying Digital Twin Principles on Administrative Law within the Rule of Law, ACM Symposium on Computer Science and Law 2025, Munich.
• Schnitzhofer (2025): Integrating Legal Ontologies and Digital Twin Technology, ESWC 2025 PhD Symposium, Portorož.
• Schnitzhofer, Schütz (2025): Towards a Scalable Architecture for Legal Ontologies Integrated into Digital Twins of Administrative Law, SEMANTiCS 2025 Developers Workshop.
• Schnitzhofer, Schütz (2025): Enhancing Automated Decision-Making in Administrative Law Through Digital Twins of Legislation: A Grounded Theory Approach, JURIX 2025, IOS Press.
• Schnitzhofer, Schütz (2025): Towards Translating Natural Language Normative Text into a Digital Twin of Administrative Law, NLL2FR 2025 at JURIX 2025.
• Schnitzhofer, Nikiforova, Schütz (2026): Digital Twins of Legislation for Explainable Automated Decision-Making in Administrative Law, AIOG 2026 at ICAIL 2026, Singapore.
• Schnitzhofer, Nikiforova, Schütz (2026): Reducing Administrative Burden by Automating the Translation of Administrative Law into Digital Twins of Legislation, JURISIN 2026, LNCS, Springer.
• Schnitzhofer, Pils, Seper-Ambros (2026): The Self-Driving State: AI-Driven Governance and the Transformation of Public Administration, Springer.
• In progress: Schnitzhofer, Hipfl, Schütz, Legal Ontologies between 2017 and 2025: A Systematic Mapping Study.

The contributions are listed with abstracts under Publications; the PDFs are available there after registration.

The next step

Public administrations that want to contribute a use case of their own to the portfolio will find on the Public Administration page how ReqPOOL supports projects from the selection of suitable norms through to operation. If you would like to assess which of your levies, fees, or eligibility checks are suitable as a digital twin of legislation: Book an expert meeting.