Agent Evaluation Framework

Metric Selection

Accuracy Correctness of outputs against ground truth

Latency Response time in milliseconds

Cost Efficiency API/token cost per task completion

Safety Adherence to safety constraints and guardrails

Hallucination Rate Percent of false/unsupported claims

Consistency Reproducibility across multiple runs

Completeness Coverage of all required output elements

Selected Metrics:

Framework Name

Metric	Weight	Excellent (5)	Good (3)	Poor (1)
Select metrics and click Generate to build rubric

Configure metrics to generate test templates

Metrics

Total Weight

Max Score