Engagement format
A Legal AI research pipeline with versioned sources and citations
Legal AI workflow design with corpus boundaries, source currency, citation QA, evaluation, and qualified human review—not legal advice or automated conclusions.
Describe the taskProblem framing
What must be resolved
A large legal corpus loses value when a team cannot show source currency, excerpt provenance, retrieval completeness, and the path from source to conclusion.
Fit
Suitable for
For LegalTech teams, universities, and organisations analysing legislation, decisions, or case material.
Deliverables
What remains after the engagement
- Corpus plan
- Processing pipeline
- Citation QA
- Reproducibility package
Process
How the work proceeds
- 01Legal question
- 02Unit of analysis
- 03Corpus
- 04Test
- 05Report
Scope
From input to a decision
When it is needed
This format addresses legal or case material where textual similarity must not be mistaken for a legal conclusion. Jurisdiction, temporal scope, unit of analysis, and the person able to stop a result all matter.
Input material
Inputs include the research question, jurisdiction and time, corpus currency, unit of analysis, and a qualified reviewer. Without those boundaries, a system may sound confident without providing a basis for professional work.
What we do
The work designs a corpus plan, versioning, extraction and retrieval pipeline, citation QA, counterexamples, and a professional review path. Document, fact, rule, interpretation, and decision remain separate in both the model and the result.
Artifact and outcome
The team receives a corpus plan, pipeline, citation-QA protocol, and reproduction package. The artifact states what was found, which material version was used, what it does not cover, and who reviews it before professional use.
Acceptance criterion
Acceptance means every material statement in a representative package can be traced back to an admissible source, version, passage, check, and reviewer decision.
Boundary of responsibility
The workflow supports qualified research and review. It does not provide legal advice, claim regulatory compliance, or turn semantic similarity into a legal qualification.
Decision record
The work record should keep a research question separate from a professional conclusion: it identifies jurisdiction and time, corpus scope, source version, citation checking method, and the person conducting review. If any part is unknown, the result must not be presented as legal advice.
Boundaries
Risks checked early
- Outdated sources
- Mixing fact and interpretation
- No citation audit
Evidence path
Existing material that demonstrates the scope of work
FAQ
Practical questions
Does the system provide legal advice?
No. It supports research and professional work; the qualified user remains accountable.
Can similar text be treated as a legal conclusion?
No. The pipeline separates material retrieval, its version, citation, interpretation, and professional decision. Textual similarity can be a cue for review, but it does not replace qualified review or legal advice.
Where does responsibility sit in this workflow?
The workflow owner names the jurisdiction, time period, unit of analysis, and person able to stop or approve use of a result. The package should let that person move from a statement back to material and citation control.
Scope and cost
Evidence before an estimate
I do not publish a fictional ‘from’. Scope and estimate follow a review of the workflow, material access, risk, expected artifact, and acceptance criterion.
Start by describing the workflow
The first response will identify whether this format fits and what is needed for an honest scope.