Skip to main content
Human Reviews is where your experts grade individual AI traces. A review is one expert’s grading of one trace: a score on each dimension of a rubric, a short rationale, and, when the answer was wrong, a corrected answer. Captured as structured data instead of scattered across Slack threads and spreadsheets, that judgment can train your judges and speed up later reviews.

​
How a review flows

1

Define the standard

Under Rubrics, an admin sets what to grade and what “good” looks like. See Rubrics.
2

Route traces to reviewers

Send traces to a queue, which spreads them across a team, or hand them to specific people. See Queues & Reviews.
3

Grade

Each reviewer works through their own list, with the Review Copilot drafting rationales and corrections for them to check.
4

Reuse the results

See where reviewers agreed, calibrate a judge against their grades, and turn corrected traces into datasets.