# AI evidence review with each draft and its evidence side by side

## What an engagement can include

- A written map of your current review practice, agreed before any build.
- An evaluation set from reports you have already reviewed, with the miss rate reported before use and rerun after model or prompt changes.
- A decision record for each statement, with your staff approving the report.
- Private models on your own hardware, so drafts stay on your systems.

[How Justin evaluates models](https://justj.dev/evaluation)