Voice Agent
Reference

Glossary

The canonical terms used throughout Voice Agent documentation.

Agent Under Test (AUT)

The voice agent being evaluated. The run detail uses the label Agent Under Testing.

Test Agent (TA)

The counterparty that interacts with the AUT in a generated test. UI labels can include TA → AUT, TA Role, and Calling From (TA Number).

Scenario

The configured expected interaction for a test. A scenario is not the same as Call Driver, which is one classification field in a run’s configuration.

Test suite

A group of related runs and their aggregate results.

Test run

One execution or uploaded-recording evaluation and its evidence. The UI can call it a Call or Recording when naming a local control.

Evaluation

The metrics, evidence, and results associated with a run. It is not a separate item you create.

Rubric and metric

A rubric is a grouped evaluation dimension. A metric is an individual evaluated check within that rubric.

Robustness and bias

Audio Robustness compares clean and noisy variants. Language Bias compares paired accent variants, shown as Accent A and Accent B. When configuring a test, both sit under Real-World Reliability. Language Bias appears only for phone and local simulation runs that have at least two accents. Suite results still label the comparisons Robustness Breakdown and Bias Breakdown.

Artifact

A generated or uploaded evidence file, such as a recording, transcript, or report. Use the concrete UI name when following an instruction.

Workspace

The selected, membership-scoped boundary for agents, scenarios, suites, runs, and results.

On this page