Legal Engineering
Defined
A framework & working toolkitSearch the guide
Explore the chapters

Terms, sources, and limits

This guide combines a proposed professional framework, explanations of established methods, and a description of a local toolkit implementation. These are different kinds of claims. A synthetic example illustrates a design; it does not establish a result in legal practice.

A short glossary

TermMeaning in this guide
Process architectureA view of the organization's connected areas of work
As-Is / To-BeA model of current practice / a proposed future process
BPMNBusiness Process Model and Notation, a formal language for process models
ManifestThe structured, versioned definition from which the toolkit generates execution and visual artifacts
CapabilityA bounded kind of work a workflow needs to perform
Domain capabilityA Layer 1 method for one kind of legal work, such as research, cite-checking, drafting, review or extraction
Capability resolutionDelivery's record of how each stated need is met: a reviewed binding reused, an installed skill wrapped, a new method built, or the need left unresolved
Design sign-offThe process owner's confirmation that the proposed process is right, given before delivery resolves capabilities
Release sign-offThe process owner's approval of the exact compiled workflow version that may run on matters
SkillReusable instructions, schemas, references, and any maintained helpers for performing work
BindingA declared connection between a required capability and an implementation or available execution mechanism
HostThe active AI environment supplying reasoning and connected tools
CheckpointPortable records needed to continue a particular workflow run
IncrementA usable addition to a product that meets its Definition of Done
C4Software architecture views covering context, containers, components, and code
CI / CDContinuous integration / continuous delivery or deployment, depending on the stated meaning

Public reference points

The guide's definitions of legal engineering and its three-layer stack are proposals by Kevan Wee. They are not presented as definitions adopted by these organizations.

Implementation boundary

The current toolkit is local and session-driven. It uses a Python runner, JSON contracts, and SQLite state. The active host supplies the model and tools. It does not include an unattended hosted worker, authenticated skill endpoints, enterprise identity enforcement, or full BPMN-engine compatibility.

Packages target Codex, Claude Code, ChatGPT, and Claude chat. Package structure and checkpoint tests are distinct from live deployment in those products. A host without Python execution can support discussion and standalone methods but cannot perform this deterministic local run.

The recorded local checks include four-target package assembly, integrity verification, synthetic workflow routing, and independent BPMN import for the example models. Live skill selection, connector permissions, commercial source access, and substantive legal accuracy were not established by those checks.

Reading and reuse

This website is publicly readable. The source repositories remain private while the framework develops. Original framework and guide material is © 2026 Kevan Wee, all rights reserved. Public access does not itself grant a reuse license. Separately licensed software components and third-party materials retain their own terms.