What makes a legal AI field test useful?
A field test should let readers inspect the evidence behind the conclusion: the actual source, task, prompt, configuration, unedited responses, comparison criteria, observations, and limits. A fluent demonstration alone does not show whether a draft is supported by the record.
Our evidence policy
Record the method before generating the responses. Retain every planned run. Link observations to source passages. Identify who prepared the comparison and who completed the legal review. Leave cost, review time, and publication dates unset until they are actually documented. Distinguish observed outputs from authored teaching examples.
Use the method on your own workflow
Start with a bounded task and a source you can check. Define unacceptable errors before testing, review the complete result, and measure the work required to produce an accepted answer. Use the legal AI evaluation guide →
