---
name: agent-evals
description: Build repeatable evaluations for an agent from production failures, deterministic verifiers, cost, latency, and trace evidence.
---

Turn representative failures into isolated tasks with objective checks and comparable runs.