Frequently asked questions
Are the personas real people?
No. The personas are synthetic and should always be labeled that way.
What can teams test?
The stated MVP covers sales objections, support escalations, and vulnerable-user safety scenarios.
What does a simulation produce?
A run can produce a trace, rubric score, risk flags, suggested fixes, and a reusable evaluation set.
Can a simulation replace testing with real users?
No. Synthetic scenarios support repeatable evaluation, but a team decides what additional research and human review its launch requires.
How are sensitive scenarios handled?
Sensitive packs require review, clear usage policies, trace provenance, and restrictions against deceptive or impersonation uses.
Can results be compared over time?
Saved evaluation sets can be reused to compare future agent runs when the team keeps the scenario and rubric controlled.
Does the platform guarantee an agent is safe?
No. Scorecards and flags support review; they do not prove safety or remove human accountability.
Who is this for?
It is aimed at AI product teams, research groups, training organizations, game studios, agencies, creators, and educators.
What should a team prepare?
Prepare an agent goal, the workflow being tested, relevant constraints, a risk boundary, and the decision the evaluation must support.
What should we verify before integration?
Verify the availability and fit of the API, test-harness, prompt-repository, analytics, SSO, webhook, and export connections your workflow needs.