The LLM drafts, the engine decides, the expert verifies.

Every record ends in one of three outcomes — applicable, not applicable, or undetermined — with a trace that cites the clause, the rule version and the source of every fact.

1 Extraction2 Dependency graph3 Rule language4 Concept-to-field mapping5 Three-outcome engine6 Test harness

Create the knowledge graph

Upload the Act and, optionally, the Rules made under it, the notifications, circulars and judgements that change or interpret them, and the records to check. The pipeline runs the six framework activities in a background worker. Anything the documents refer to but you did not upload is listed after the dependency graph, and the run waits for it.

1. The Act — the main law (required)

Drop a file here or browse

.html / .xml / .docx / .txt preferred over .pdf

2. Rules made under the Act (optional)

Drop files here or browse

one or more files

3. Changes and interpretations (optional)

Drop files here or browse

notifications, circulars, judgements, amending Acts, orders — each is recognised from its text

4. A dataset to validate (optional)

Drop a file here or browse

.csv / .xlsx / .json — without one, a synthetic path-coverage dataset is used

Target provisionsAct code

StructureSource

Interest on tax payable up toDefault: the day the engine runsPause after the dependency graph and ask for referenced documents that are missing

TitleUse the LLM for drafting when configured (it never decides verdicts)

Rulebooks

A rulebook is a law's rules frozen from a knowledge graph. Reuse it on any number of datasets: only mapping, engine and tests run.

Knowledge graphs

Loading…