Legal corpus management relies on a paradoxical blend of rigidity and flexibility. Like software codebases, laws are incrementally maintained by distributed actors—parliaments, courts, legislators—each adding or modifying provisions over decades. This creates a palimpsest where a 1995 statute coexists with 2021 amendments, all anchored by stable fine-grained addresses. These addresses, such as 'Section 12(2) of Act X,' act as permanent pointers, ensuring contracts, court rulings, and regulations can reference specific legal text without disruption. The system’s resilience hinges on this addressing scheme, which outlasts amendments by using hierarchical paths rather than mutable references.

The legal structure mirrors a tree in format but functions as a graph in operation. Statutes are printed as linear hierarchies (parts, chapters, sections), but their meaning relies on cross-references that span branches. A tax law might cite an EU regulation, while a constitutional provision overrides a distant section. This graph-like behavior—enabled by references, overrides, and dependencies—requires tools that reconcile the printed text (tree) with its semantic meaning (graph). Early efforts like Akoma Ntoso and LegalRuleML separated structure from analysis, but modern challenges demand reproducible text-layer computation before semantic tools can operate effectively. Law’s text must be a reliable, replayable substrate for interpretation.

Amendments are operations, not edits. Instead of replacing text, they specify a target address, an action (replace, repeal), and a payload. This precise vocabulary—‘Section 12, subsection 2 is amended to read…’—is standardized across jurisdictions, from Finland to the EU. However, legal meaning changes transcend text operations. Interpretive overlays, delegated legislation, and conditional applicability create semantic layers that text compilers ignore. For example, a repealed provision might still govern old contracts, or a constitutional clause might require layering interpretations without altering the base text. A legal virtual machine (VM) focuses on compiling the text state, leaving normative interpretations to downstream tools. This separation ensures reproducibility but requires semantic tools to handle meaning shifts outside the text layer.