FAQ
Short answers to common questions.
On this page
Do you see my model keys?
No. Your provider key stays on the machine running your assistant. The runner sends back the answer and a limited activity record, not the key.
Can a taste finding fail a release?
An advisory finding cannot. You must first choose a supported rule, such as “for Profile B, never recommend these restaurants”, and approve it. Then a new run can pass or fail that rule.
What if Qloo doesn’t know a restaurant?
You will see a gap in the evidence. That does not mean the diner dislikes it, and your assistant may still recommend it if it is allowed by your task.
Can I test restaurants outside my catalog?
The task uses your allowed restaurant list. If your assistant suggests something outside it, TasteCase keeps that answer so an approved check can flag it.
Do I need two different models?
No. Baseline is your current assistant; Candidate is the version you want to try. Both can use the same model with different prompts. Configure those differences locally — changing a release label alone does not change the assistant.