position paper / Legal AI and Computational Law
A swarm instead of a platform: a new architecture for national legal systems
From one monolith to replaceable workspaces and an explicit artifact standard
An architectural thesis: legal infrastructure can use many controlled agent workspaces when they exchange verifiable work packages through a shared contract.
In brief
- Interoperability should apply to artifacts and evidence, not one interface.
- A local workspace must respect institutional roles, access, and accountability.
- A pilot should first test one workflow and whether its package can be reconstructed.
Starting point
A conventional platform tries to place many roles, procedures, and institutions inside one interface. In legal work, that often means averaged workflows, long change cycles, and dependence on a central supplier. The alternative is not an uncontrolled collection of bots. It is a federation of bounded workspaces that follow a shared exchange contract.
This edition develops an argument first published in Ukrainian. It is an architectural position and a pilot design, not a description of an operating national infrastructure.
The unit of exchange is a work package
A workspace may use a different model, interface, and tool set if its output has the same verifiable shape. A package should contain source identifiers, the material version, transformations, unresolved conflicts, test results, and the accountable person’s decision. Answer text alone cannot carry professional work safely between institutions.
| Layer | Control question | Artifact |
|---|---|---|
| Source | Where did the material come from? | source register |
| Analysis | What did the process change? | transformation log |
| Evaluation | How was the result checked? | test report |
| Decision | Who accepted responsibility? | approval record |
Local workspace boundaries
Each organisation retains its roles, access rules, and procedures. An agent operates inside an assigned contract rather than acting as an abstract “digital lawyer.” Local control makes it possible to stop a task, replace a tool, and audit the result without waiting for the whole ecosystem to be rebuilt.
A minimal pilot
The first experiment should cover one workflow, two workspaces, and a small explicit scenario set. A second professional receives the package and attempts to reconstruct the path from source to conclusion. The pilot measures completeness, citation fidelity, reconstruction time, missing elements, and the points where a person must stop automation.
Failure modes
A swarm without a contract becomes disorder. A shared format without version control creates only apparent interoperability. A central catalogue that does not allow local rejection recreates monolithic dependency. The most serious failure is a package that looks complete while hiding a missing source or an outdated state of the material.
Decisions still required
Implementation requires decisions on the legal basis for exchange, data classes, accountability, procurement, a minimum security standard, and withdrawal of a defective version. Architecture can make these questions explicit. It cannot answer them on behalf of legislators, institutions, or legal professionals.
LIM
Limitations and scope
This is an architectural position and pilot proposal, not an operating national system or a procurement recommendation. Implementation requires legal, security, procurement, and accountability review.
SRC
Sources and external edition
Original or earlier Medium material
AI tools assisted with structure and editing. The final text underwent human editorial review of facts, sources, conclusions, and attribution.
LOG
Change history
- First expanded edition in the owned archive.
- Review of structure, limitations, and evidence links.