4.0 Domain 4 Roadmap: Evaluating and Deploying Workflows (22% Weight)

Core Architectural Topics in Domain 4 (~22%)

1. Trajectory Evaluation vs. Final Response Evaluation (Module 4.1)

Build golden evalset datasets to test tool-call sequences (EXACT, IN_ORDER, ANY_ORDER trajectory match modes) alongside final response rubrics and LLM Autoraters.

2. Production Runtime Selection: Agent Runtime vs. Cloud Run vs. GKE (Module 4.2)

Select the optimal hosting platform based on managed session/memory integration, serverless scale-to-zero, or kernel-isolated code execution (Agent Sandbox on GKE).

3. Agent Starter Pack & Automated CI/CD Release Gates (Module 4.3)

Bootstrap Terraform infrastructure, automated evaluation gates, and safe traffic-splitting rollouts using the Agent Starter Pack.

4. OpenTelemetry Tracing, Monitoring & Cost/Latency Optimization (Module 4.4)

Inspect multi-agent execution spans in Cloud Trace, track token consumption and loop anomalies in Cloud Monitoring, and tune production resilience.