Skip to content

Metrics Selection

The Metrics tab lets you choose which evaluation metrics to run against your API responses.

Metrics Tab

Metric Categories

Metrics are organized into five categories, accessible via sub-tabs:

RAG

Metrics for evaluating Retrieval-Augmented Generation systems:

Metric Description Required Fields
AnswerRelevancyMetric Multi-step relevancy assessment input, actual_output
AnswerPrecisionMetric Precision vs expected output input, actual_output, expected_output
FaithfulnessMetric Factuality vs retrieval context actual_output, retrieval_context
ContextualRelevancyMetric Context relevance to the query input, retrieval_context
ContextualPrecisionMetric Precision of retrieved context input, actual_output, expected_output, retrieval_context
ContextualRecallMetric Completeness of context expected_output, retrieval_context
BiasMetric Bias detection in responses actual_output
ToxicityMetric Toxicity level detection actual_output
RestrictedRefusalMetric Proper refusal of harmful requests actual_output
GEval Custom G-Eval scoring input, actual_output
CustomEvalMetric Custom verdict-based evaluation input, actual_output

Agent

Metrics for evaluating AI agents and tool use:

Metric Description
ToolCorrectnessMetric Verifies correct tool selection
TaskSuccessMetric Evaluates task completion
RoleAdherenceMetric Checks role compliance

Security

Metrics for safety and security evaluation:

Metric Description
PromptInjectionMetric Detects prompt injection attacks
JailbreakMetric Detects jailbreak attempts
PIILeakageMetric Checks for PII in responses

Deterministic

Non-LLM metrics using exact matching and pattern rules.

Vector

Vector-based similarity metrics.

How to Select Metrics

  1. Click the checkbox next to each metric you want to enable
  2. The required fields (shown as grey tags beneath each metric) indicate what data is needed
  3. Ensure your Column Mapping and Response Mapping provide the required fields

Warning

Selecting metrics that require fields not available in your dataset or API response will result in evaluation errors. Check the required fields tags carefully.