Audit the recommendation, not the rhetoric.
TasteProof is cultural QA for AI agents. Give it an agent recommendation and explicit taste anchors; Qloo supplies grounded candidates and ranks the proposal against them in one comparable evidence context.
Fixture mode demonstrates UX only. Live mode uses server-side Qloo evidence and fails closed if Qloo is unavailable.
1. Define the audit
Use explicit named cultural anchors only. No inferred identity or sensitive traits.
Ready.
2. Taste Passport
The conclusion is relative to the candidates tested in this run—not a universal cultural truth.
3. Reproducible evidence path
Inspect evidence record
No run yet.
Why Qloo is necessary
Without Qloo, TasteProof can collect a proposal but cannot perform its central cultural-grounding test. Qloo creates the comparison set and evaluates the proposal against explicit taste signals.
- Canonical workflows: recommend → rank
- Proposal and alternatives ranked together
- Evidence separated from interpretation
- Digest-backed run identity
Responsible boundary
Qloo describes aggregate taste patterns. TasteProof does not claim to predict one person's behavior, infer sensitive identity, or make eligibility decisions.
Ambiguous or unresolved entities remain visible. Live failures never silently become fixture results.