BrowserStack AI Evals
EvaluationScenarios

Scenarios

A conversation-shape dataset type for the Conversation Simulator — each row is a persona, a goal, and an opening message.

Scenarios

Scenarios are a dataset type in the Dataset Library, alongside Datasets and Evaluation Datasets. A scenarios dataset is conversation-shaped: each row is a test case the Conversation Simulator drives against your agent. Every row carries a persona, a goal, and an opening message — the simulator roleplays the persona, opens with that message, and works toward the goal over a multi-turn conversation.

What a scenario contains

FieldDescription
PersonaThe user profile the simulator roleplays for this row. See Personas.
GoalWhat the simulated user is trying to achieve, in plain language — for example Place an order for next-day delivery.
Opening messageThe first thing the simulated user says, which starts the conversation.
Expected output (optional)The outcome you expect from the agent, used as a reference when scoring.

Personas

A persona is a reusable user profile the simulator roleplays — a name, a short description, behavioural traits, an example opening utterance, and a system prompt that governs how it behaves across a conversation. Every scenario row carries one persona. Personas fall into three groups:

  • Cooperative — collaborative, on-topic users (for example Helpful User, Confused Novice, Domain Expert, Terse / Power User).
  • Challenging-but-realistic — users that stress the agent (for example Topic Drifter, Indecisive User, Anxious User).
  • Your own — custom personas you create for your product, each with its own name, description, and behaviour.

Personas are attached to scenario rows when you build a dataset — each row carries its own persona. You do not pick personas when you create an experiment.

Working with scenarios