Trust & feedback

How often the LLM's explanations pass the guardrail, and what human reviewers think of replays and live verdicts.

Export replay reviews
LLM guardrail eval

Not measured yet. Run rootsight llm check in a terminal. Every explanation is still checked live before it is shown.

Human feedback
Reviews
–
Reviewer agreement with truth
–
Live verdicts rated correct
–
Guardrail precision
–
Recent replay reviews

No replay reviews yet. Rate an episode in Agent at work; live verdicts are reviewed in Incidents.

Command palette

Jump to a page or run an action