# Legal AI research pilot template

Prepared by Judicio Editorial Team, 9 September 2026.
Blank working template. No benchmark results or product accuracy claims are implied.
Use public, synthetic or appropriately approved material. A qualified reviewer determines the legal conclusions.

## 1. Define the task before testing

- Task ID and reviewer:
- Tool, version, mode and test date:
- Research question and supplied facts:
- Governing law and legal issue:
- Country, state or member state, court or regulatory framework:
- Relevant event date and legal version:
- Source collections, publication ranges and known gaps:
- Information approved for use:
- Intended output and audience:
- Acceptance criteria agreed before the run:
- Existing workflow used for comparison:

## 2. Include different kinds of questions

| Exercise | What the reviewer should examine | Your task and expected sources |
| --- | --- | --- |
| Straightforward authority | Source identity and passage support | |
| Mixed jurisdiction | Separation of distinct legal systems | |
| Historical version | Event date, amendment and commencement | |
| Contrary authority | Whether adverse material and qualifications remain visible | |
| Incomplete source material | Whether the answer records uncertainty and further work | |

Examples for scoping, not statements of applicable law:

- United States: state and federal questions, forum and court hierarchy.
- UK: England and Wales, Scotland or Northern Ireland, court and relevant date.
- EU: EU-level instrument and member-state implementation question.
- UAE: mainland, DIFC or ADGM, entity, forum and source language.
- India: Supreme Court, relevant High Court or tribunal, provision and event date.
- Other markets: specify the country and any applicable regional or special framework.

## 3. Inspect each material proposition

| Proposition | Authority and source reference | Identity checked | Supporting passage and context | Status/version checked | Contrary material or gap | Reviewer decision |
| --- | --- | --- | --- | --- | --- | --- |
| | | | | | | |

Use **verified**, **needs correction**, **unresolved** or **not applicable**, with a reason.
An inaccessible source is unresolved, not verified. A real citation does not by itself establish passage support or current legal status.

## 4. Record exceptions

| Task | Exception | Materiality | Correction or further research | Owner | Resolution |
| --- | --- | --- | --- | --- | --- |
| | | | | | |

Possible exceptions: wrong jurisdiction, missing authority, unsupported proposition, incorrect quotation, historical-version gap, unresolved subsequent treatment, incomplete factual input or export loss.

## 5. Compare total effort

| Stage | Existing process: minutes | AI-assisted process: minutes | Notes |
| --- | --- | --- | --- |
| Scope and prepare inputs | | | |
| Search and prepare working answer | | | |
| Inspect and verify sources | | | |
| Correct or extend research | | | |
| Review and handover | | | |
| Total time to approved result | | | |

Record product cost separately using the actual plan and consumption. Do not compare first-answer speed against an entire manual workflow. Include failed runs and unresolved tasks in the report.

## 6. Decide and retain evidence

- Did the task meet the pre-agreed acceptance criteria?
- Which uses, jurisdictions or source ranges need more testing?
- What corrections or additional services are necessary?
- Can the output and sources be retained and exported appropriately?
- Reviewer, decision and date:
- Evidence references and unresolved items:
- What change should trigger another test?

This template supports a local evaluation. It does not certify a product, establish completeness or replace professional review.