What we’ve tested
Completed checks, real runs and remaining gaps.
On this page
These are completed checks, not promises about every future run. Provider availability and free-tier limits can change.
Code checks
The deployed release passed 668 automated tests, code-style checks, type checks and the production build. Its dependency audit reported no known vulnerabilities.
Saved work and access checks
Tests covered sign-in, account isolation, runner connections, saving decisions, publishing test collections, reopening exact saved versions and retrying uncertain saves without creating duplicate work. Separate database checks used the dedicated hosted Supabase project.
Some automated tests use simulated provider results to check errors and edge cases. Those are not the real results shown in the public sample.
Real Qloo and model runs
A hosted browser run completed discovery, review, publication, an 8-of-8 comparison, downloads and reopening. It used four Qloo calls and eleven model requests: one case-selection request plus ten assistant requests. There were no retries or paid fallbacks. The ten local assistant responses reported zero cost.
A downloaded test pack was also replayed through the local runner before deployment, without new Qloo calls. The newly downloaded hosted pack was not separately replayed.
The hosted app
TasteCase is hosted on Vercel, with its API and discovery worker on one free Render service and saved data in Supabase. Private-route and second-account access checks passed in the hosted browser run.
Recorded evidence
What remains untested
- Successful signup and password-recovery email links.
- A deliberate hosted cold-start test, cancellation and objective-rule failure branches.
- Replay of the newly downloaded hosted pack, or a runner on a different physical machine or Windows.
- Whether Qloo-guided discovery finds more useful cases than ordinary generation.