# Legal AI work-sample kit
AI Vortex · 8 September 2026 · Suggested practice framework

[AI Vortex](https://www.aivortex.io/legal/) · [Legal AI careers](https://www.aivortex.io/legal/careers/) · [Review lab](https://www.aivortex.io/legal/careers-resources/review-lab/)

Use public or synthetic material. This kit does not authorize use of client information, legal advice, a filing or a real-matter workflow. Follow employer/course rules and obtain appropriate review.

## 1. Choose the task
Write the intended user, jurisdiction if relevant, question, source set, allowed tool/environment, expected output, explicit exclusions, and who reviews it. Define what would make the result unacceptable before running the tool.

## 2. Worked miniature: synthetic chronology
This is an invented training example, not a client result.

Inputs:
- A: email dated May 6 stating that a package was delivered "yesterday".
- B: letter dated May 5 stating that the package had not arrived.
- C: undated delivery photograph with no usable timestamp.

Bad first pass: "The package was delivered on May 6, confirmed by all three documents."

| Source | What it supports | What remains unknown | Correction |
|---|---|---|---|
| A | Sender asserts delivery on the day before May 6 | Whether assertion is accurate | Preserve as attributed statement, not established fact |
| B | Sender asserts non-delivery as of the letter | Time of day and whether delivery happened later | Retain the apparent conflict |
| C | A package appears in the image | Date, location and identity | Do not use it to fix the delivery date |

Reviewed outcome: A and B contain statements that require reconciliation; C does not establish a date. Escalate before drawing a legal conclusion. No claim of time savings is made.

## 3. Reusable evidence log
Copy one record per run:
- Run ID and date:
- Task and intended user:
- Public/synthetic source identifiers and source dates:
- Tool, configuration and permitted environment:
- Instructions and output requested:
- First-pass output location:
- Material errors and omissions:
- Source checks performed:
- Corrections and unresolved questions:
- Reviewer and feedback (only identify with permission):
- Time preparing inputs:
- Time producing first pass:
- Time verifying and correcting:
- Total time to reviewed result:
- Acceptance or stop decision:
- Change to make before next run:

## 4. Compare fairly
Use the same source set, scope and acceptance standard for the comparison. Count review and rework. Separate repeated practice effects from tool effects. A small exercise is not a general benchmark and cannot establish an employer-wide productivity claim.

## 5. Portfolio handoff
Attach a short statement of your contribution, source provenance, whether the case is synthetic, what the tool did, what you corrected and what remains unresolved. Share only material you are permitted to share. Never turn a simulated outcome into a resume claim about paid client work.

## Source framework
- O*NET Lawyers: https://www.onetonline.org/link/details/23-1011.00
- O*NET Paralegals: https://www.onetonline.org/link/details/23-2011.00
- ABA Formal Opinion 512 (29 July 2024): https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf
These inform the attention to task, source and supervision. They do not endorse this kit.