Generic AI
A plausible suggestion. It can sound personal without grounded cultural evidence.
Judge walkthrough · no login
Qloo understands what feels like Margaret. TastePolicy decides whether her AI may act on that understanding. Three short acts show the difference with real Qloo data.
Act 1 · The person
She still knows exactly what she enjoys. Her circumstances changed; her taste did not.
Request: “Somewhere for a lovely evening out, nothing too far, and not too loud.” AI permission: suggest, compare and shortlist — nothing more.
Act 2 · Cultural intelligence
Qloo resolves Margaret's five cultural anchors to real entities, discovers cross-domain places from those signals and provides candidate-specific cultural evidence. TastePolicy keeps that evidence separate from accessibility, walking time, noise, price and every other real-world requirement.
A plausible suggestion. It can sound personal without grounded cultural evidence.
The same choice becomes culturally grounded: this candidate fits Margaret's cross-domain taste, with evidence for why.
Now ask the question recommendation alone cannot answer: does it work for her real needs, and may the AI go this far with it?
Place photos, maps, street sequences and 360° media are presentation. They never become factual evidence merely because they look relevant.
Act 3 · Agency
Something important is still unknown, so your AI stops and asks instead of guessing.
This fits you and what works today. Your AI may go as far as you allowed.
A known hard violation becomes REJECT. If required evidence cannot be obtained, the result is UNAVAILABLE. Two concise variants show the same contract beyond Margaret: sensory-aware and low-energy.
The thesis
Qloo answers what fits this person? TastePolicy answers may this agent go this far with that fit, for this person, now?
Optional deeper track
The product idea is complete above. Open this section only if you want the underlying authority contracts, developer surfaces and evidence trail.
TastePolicy resolves cultural anchors through Qloo, verifies the identities being used, discovers cross-domain candidates and requests candidate-specific affinity and explainability. It does not recreate Qloo's graph or substitute generic LLM guesses for cultural evidence.
Open the Assisted Choice experience · Inspect the engineering route
Known hard violation → REJECT. Critical fact missing → ASK. Required evidence unavailable → UNAVAILABLE. Requirements satisfied and the requested action within the person's ceiling → ALLOW.
Qloo evidence, place facts, requirements, rules and the final decision remain separately attributable in Cultural Receipts.
The same authority contract is exposed through REST, the TypeScript SDK, remote MCP, A2A and OAuth. Deeper controls include versioned policies, bounded leases, drift checks and portable proofs.
Developer platform · Engineering demo · Proof verifier · Trust boundary · Live readiness
Hosted, live, controlled, synthetic and blocked evidence are labelled separately. The public repository includes release, acceptance, benchmark and human-pilot records rather than turning unproved integrations or test fixtures into product claims.
ALLOW grants preference-choice authority only. It never grants payment, booking, messaging, account changes, private-data disclosure or legal authority. Recommendation is not authorization, and verifying a proof does not make a separate execution permissible.