Current limitations
Known gaps and boundaries, stated plainly.
What you can test
- Dining recommendations from your fixed restaurant list, not booking or live availability.
- Up to ten matched restaurants per discovery job and ten selected cases per comparison.
- Only the supported checks you approved. A written reason is not executable test code.
- Comparisons allow 120 planned runs, 24 reclaims and 144 charged dispatches.
What the evidence cannot prove
- Qloo rankings suggest questions to inspect, not what an individual should like.
- Missing evidence is not a pass or a dislike. Partial activity records cannot prove a tool was skipped.
- A release label identifies your local setup; it does not prove which model actually ran.
- Restaurant facts are saved as you supplied them. TasteCase does not verify their freshness.
- A sent provider call may count against your allowance even if its outcome is unknown.
Hosting and connection limits
- The free Render backend sleeps after 15 idle minutes and may need about a minute to wake. Background discovery may pause; saved work remains in Supabase.
- Render’s free hours are shared with other services in the workspace. Free model quotas and provider availability also limit how many tests you can run.
- A runner’s saved handshake proves a past connection, not that it is online now.
- History loads in pages of 20, up to an offset of 10,000. A loading failure or exceeded limit is shown, not silently treated as the end.
- The public sample is read-only and recorded. It shows an evidence gap and partial activity records, not a fully verified recommendation verdict.