evaluation

To evaluate agent outputs against custom rubrics, define a multi-dimensional rubric with weights, provide the expected ground truth and tool-use logs, and the pipeline will compute per-dimension scores and an overall pass/fail judgment.

A multi-dimensional rubric for agent monitoring is a defined set of weighted criteria, such as factual accuracy, citation accuracy, and tool efficiency, used to systematically score agent performance against ground truth.

Yes, you can use rubric evaluation for multi-agent configurations by running automated evaluations against context histories and tool logs to generate scores and pass/fail verdicts across single-agent and multi-agent setups.

You need a defined rubric with weights, ground truth data, and access to agent tool-use logs and context histories to generate structured dimension scores and overall pass/fail judgments.

Automated agent evaluation works for continuous monitoring in production by tracking quality over time, comparing configurations, and enforcing quality gates within development and production pipelines.