Discovery
Find a restaurant pair worth checking, within your call limits.
On this page
Choose what to test
Select your saved task and restaurant list. Add two fictional diner profiles, each with one to five public music, film or brand interests. Both keep the same practical needs. Choose your assistant’s version label and how many times to repeat each request.
The optional two-restaurant follow-up focuses on the interesting pair. The task, practical needs and eligible restaurant list stay unchanged once discovery starts.
What happens next
- You confirm the public-interest matches.
- Qloo ranks the restaurants for each diner.
- TasteCase finds a pair whose order changes and selects a case to inspect.
- Your runner asks your assistant the saved requests and returns its answers.
- You open the case and decide what is worth keeping.
Needs confirmation means your input is needed. Waiting for runner means you need to run the connected program on your machine.
Keep the run small
Set call limits and a deadline before starting. Begin with a small restaurant list and few repeats. Qloo calls, case-selection calls and assistant answers each use their own allowance; your provider’s account limits apply too.
| Setting | Default | Allowed range |
|---|---|---|
| Qloo calls | 8 | 1–20 |
| Case-selection calls | 2 | 1–8 |
| Assistant runs | 6 | 2–12 |
| Investigation rounds | 3 | 1–6 |
| Time limit | 15 minutes | 30 seconds–1 hour |
| Output size | 64,000 bytes | 1,024–262,144 bytes |
| Retries per operation | 1 | 0–2 |
A sent request may use your allowance even if its answer never arrives. TasteCase does not retry a rejected API key or silently switch to a paid model.
Understand the stopping point
Ready for review means there is a case to open. No useful contrast is a valid outcome, not a broken run. Partial means some work was saved but the full plan did not finish.
Other stopping points include an unavailable provider, used-up budget, deadline, cancellation or failure. Open any saved cases and read the gaps before deciding what to do next.
Stop a run
Cancel to stop new work. A request that was already sent may still finish; its result stays labelled with the cancelled run.