Core concepts and terminology
The shared vocabulary for creating, running, and interpreting voice-agent evaluations.
Use these terms consistently when working in Voice Agent.
Workspace
A workspace is the area where your team’s agents, scenarios, recordings, and results are organized. Select the appropriate workspace before creating or reviewing work.
Agent Under Test (AUT)
The AUT is the voice agent being evaluated. It has a role and vertical; it may also have a phone number and behavior definition that give applicable tests more context.
Test Agent (TA)
The TA is the counterparty in a generated test. It takes the opposite role to the AUT: a customer for a business AUT, or a business for a customer AUT.
Scenario
A scenario describes the intended interaction for a generated test: who the TA is, what should happen in the call, and which metrics matter. You can generate, refine, and save scenarios for reuse.
Test suite
A test suite groups related work and its aggregate results. A suite lets you see overall progress, average score, and performance patterns across its runs.
Test run
A test run is one execution attempt or uploaded-recording evaluation. When generated and available, its recording, transcript, score, metric results, and evidence describe that specific interaction.
Metric and evidence
A metric evaluates one aspect of the interaction, such as whether a required behavior happened. Evidence connects the result to relevant parts of the call, helping you understand why the metric received its result.
Evaluation
An evaluation is the result of applying metrics to a run: its score, outcomes, and evidence. It is useful language for discussing results, rather than a separate product object.
Logical definitions and physical runs
Selected driving metrics define the logical tests you ask Voice Agent to perform. Each eligible definition may expand into physical runs when you add robustness or bias variants. The product shows the expected run count before you submit a test.