LawGraph

POC/thought experiment for a JSON-based intermediate representation for the logical structure of legislation. It captures statutory text, the rules embedded in that text, and known points of ambiguity in a single machine-readable document. The format is jurisdiction-agnostic, tool-agnostic, and designed so that every derived annotation can be traced back to the source language.

Quick Example

A .lawgraph.json file for 17 U.S.C. § 107 (fair use) shows all three layers:

{
  "id": "17-usc-107",
  "citation": "17 U.S.C. § 107",
  "title": "Limitations on exclusive rights: Fair use",
  "jurisdiction": "us-federal",
  "text": "Notwithstanding the provisions of sections 106 and 106A, the fair use of a copyrighted work ...",
  "sections": [
    { "id": "17-usc-107-1", "label": "(1)", "text": "the purpose and character of the use ..." },
    { "id": "17-usc-107-2", "label": "(2)", "text": "the nature of the copyrighted work;" }
  ],
  "effective_date": "1992-10-24",

  "provisions": [
    {
      "id": "17-usc-107-p1",
      "type": "defense",
      "text": "the fair use of a copyrighted work ... is not an infringement of copyright.",
      "conditions": [
        { "id": "17-usc-107-p1-c1", "text": "the use qualifies as 'fair use' under the four-factor balancing test", "type": "factual" }
      ]
    }
  ],

  "ambiguities": [
    {
      "id": "17-usc-107-amb-1",
      "type": "undefined_standard",
      "location": "17-usc-107-p1",
      "text": "for purposes such as criticism, comment, news reporting, teaching ...",
      "description": "The word 'such as' makes the listed purposes illustrative, not exhaustive.",
      "source": "human"
    }
  ]
}

Layer 0 (top) is the statutory text and structure. Layer 1 (provisions) decomposes the text into individual rules with conditions and logic. Layer 2 (ambiguities) captures interpretive questions and competing readings.

The Three Layers

• Layer 0 — Structure. The statutory text and its formal organization: section IDs, headings, sub-sections, defined terms, cross-references, effective dates. Everything here can be extracted automatically from official sources like U.S. Code XML.
• Layer 1 — Logic. Individual provisions (obligations, prohibitions, permissions, defenses, exceptions) with their conditional logic. Semi-automatable with NLP; typically needs human review.
• Layer 2 — Analysis. Ambiguities in the text, competing interpretations, and how courts have addressed them. Primarily human-authored. Every annotation carries provenance and, for machine-generated entries, a confidence score.

Each layer builds on the previous one, and a valid LawGraph document can contain any subset of layers.

Try It

Browse parsed statutes in the web viewer: dvelton.github.io/lawgraph

Example Statutes

Statute	Citation	Layers
Fair Use	17 U.S.C. § 107	L0 + L1 + L2
Section 230	47 U.S.C. § 230	L0 + L1 + L2
DMCA Safe Harbors	17 U.S.C. § 512	L0 + L1 + L2
CFAA	18 U.S.C. § 1030	L0 + L1 + L2
Copyright Definitions	17 U.S.C. § 101	L0 + L1
Exclusive Rights	17 U.S.C. § 106	L0 + L1
DMCA Anti-Circumvention	17 U.S.C. § 1201	L0 + L1
Sherman Act § 1	15 U.S.C. § 1	L0 + L1
Sherman Act § 2	15 U.S.C. § 2	L0 + L1
FTC Act § 5	15 U.S.C. § 45	L0 + L1
Patent Infringement	35 U.S.C. § 271	L0 + L1
Stored Communications Act	18 U.S.C. § 2701	L0 + L1

Using the Parser

The reference parser converts U.S. Code XML into LawGraph JSON.

cd parser && pip install -e .
lawgraph parse "17 USC 107" --output fair-use.lawgraph.json

Options:

lawgraph parse CITATION [--output FILE] [--layer 0|1|2]

--layer controls parsing depth: 0 for structure only, 1 (default) for structure + logic, 2 for structure + logic + analysis (not yet implemented).

The parser also includes lawgraph fetch (download U.S. Code XML by title) and lawgraph validate (check a .lawgraph.json file against the schema).

Schema

The full schema is at spec/schema.json (JSON Schema draft 2020-12). The specification document at spec/SPEC.md covers the field reference, design principles, and a complete annotated example.

Building on LawGraph

LawGraph is designed to be extended. A few directions where contributions would be useful:

• Parsers for other jurisdictions. The format is jurisdiction-agnostic; the reference parser targets U.S. federal statutes, but the same schema works for state codes, EU directives, or other statutory systems.
• Renderers and analyzers. The web viewer is one way to consume LawGraph data. Others might include diff tools for statutory amendments, cross-reference graphs, or compliance checkers.
• Layer 2 annotations. The example statutes have Layer 0 and Layer 1 coverage. Adding ambiguity analysis and judicial treatment data to more statutes would make the corpus more useful for legal research.

See spec/layers.md for detailed implementation guidance on each layer.

License

MIT