AI evaluation base metrics

Base metrics from AI evaluation framework providers for configuring compliance evaluations and monitoring AI system controls.

Agentic AI base metrics

These base metrics apply to the Agentic AI system type and are captured at the span evaluation level. Span evaluation assesses individual interactions or execution steps within an agentic workflow.

Note:

AI evaluation metrics provide automated assessment but require human review and validation. Metric results support decision-making and should not be the sole basis for compliance determinations.

Table 1. Agentic AI base metrics by provider

Metric	Description	Provider	Scoring format
Prompt injection	Detects when malicious instructions are embedded in AI agent inputs to hijack the agent's behavior or override its original directives.	Traceloop	Binary
Agent goal accuracy	Measures how accurately the agent achieved the stated user goal.	Traceloop	Percentage
Toxicity	Detects when AI agent interactions contain harmful, offensive, or toxic content in inputs or outputs.	Traceloop	Binary
Answer relevancy	Measures how relevant the answer is to the question asked.	Traceloop	Binary
Context relevance	Evaluates whether the retrieved context is relevant to the user's query.	Traceloop	Percentage
Answer completeness	Measures how completely the answer covers all aspects of the question.	Traceloop	Percentage
PII detection	Detects when AI agent interactions contain personally identifiable information (PII) in inputs or outputs.	Traceloop	Binary
Agent tool trajectory	Compares actual tool calls against expected reference tool calls.	Traceloop	Percentage