AUX is an independent checkpoint between an AI agent's intention and a consequential action. It checks bounded evidence, refuses to overstate what it knows, and creates proof another machine can verify.
Doctor-channel identity rule: use an institution-issued, pairwise pseudonymous clinician ID and a short-lived authority attestation bound to one de-identified case and action. Do not send a public name, license/NPI number, patient identity, MRN, PHI, diagnosis, or reusable patient pseudonym to AUX.
Every phase keeps AUX non-executing. The person, agent, or workflow remains responsible for the final decision.
Phase 1
Live
Know what proof is required
AUX explains the evidence required for a selected certification, what is satisfied, what is missing, who must supply it, and what should happen next.
What the end user gains
A vague refusal becomes an exact checklist your AI can follow.
Message for agents
Ask AUX what must be true before you act.
Message for people
AUX tells your AI what proof it needs—and why—before approving anything important.
Ask your AI
Use AUX to determine what evidence is required before an agent takes my proposed action.
1. Read https://api.aux.prdictionedge.ai/v1/certification-profiles
2. Select the best live profile.
3. Query https://api.aux.prdictionedge.ai/v1/certification-requirements for that profile.
4. Explain what AUX verifies, what I must provide, what is missing, why it matters, and the exact next step.
5. Do not claim certification unless AUX returns CERTIFIED.
My proposed action: [DESCRIBE IT HERE]
Phase 2
Live
Check evidence outside the caller
AUX independently resolves public legal-entity, official OFAC, and domain evidence. It verifies private-source attestations against trust information published by the source system.
What the end user gains
The calling agent cannot certify its own assertions. Important facts need independent or cryptographically verified support.
Message for agents
Caller assertions are inputs, not proof.
Message for people
AUX double-checks your AI using sources your AI does not control.
Ask your AI
Use AUX to independently check this business or counterparty before I rely on it.
1. Discover live profiles at https://api.aux.prdictionedge.ai/v1/certification-profiles
2. Use counterparty_registry_pre_action for public GLEIF and official OFAC evidence.
3. Use counterparty_identity_pre_action when domain identity also matters.
4. Report the exact outcome and provenance.
5. Separate independently resolved evidence, caller facts, missing evidence, adverse evidence, and unavailable sources.
6. Do not turn a no-match into proof that an entity does not exist.
Counterparty: [NAME, LEI IF KNOWN, AND DOMAIN IF RELEVANT]
Phase 2B
Live
Keep trying when a source is down
A complete request can become a durable attempt. If a required source is temporarily unavailable, AUX keeps the same attempt, retries automatically, and supplies a status URL.
What the end user gains
Submit once. A temporary outage does not make you start over.
Message for agents
Submit once; AUX owns temporary-source retry.
Message for people
If an official source is unavailable, AUX keeps working instead of making you resubmit.
Ask your AI
Help me use AUX's durable certification flow.
1. Read the applicable profile and requirements.
2. Confirm caller-required fields are complete.
3. Prepare POST https://api.aux.prdictionedge.ai/v1/certification-attempts
4. Show me the request before sending it and remove unnecessary personal, banking, invoice, credential, or secret data.
5. Preserve the attempt_id and status_url, then poll without resubmitting.
6. Explain every returned state. Never treat SOURCE_TEMPORARILY_UNAVAILABLE as approval.
Action and evidence: [DESCRIBE THEM HERE]
Phase 3
Live
Carry proof in a signed receipt
When every bounded requirement is satisfied, AUX signs a receipt binding the decision to the proposal, evidence set, requirements, version, and time.
What the end user gains
Another system can verify what AUX checked without trusting a screenshot or copied conclusion.
Message for agents
Verify the evidence decision, not the messenger.
Message for people
AUX produces tamper-evident proof of exactly what was checked.
Ask your AI
Verify this AUX certification.
1. Use https://api.aux.prdictionedge.ai/v1/certifications/verify
2. Verify the signature and receipt integrity.
3. Confirm profile, version, status, time, proposal, evidence-set, and requirements.
4. Explain limitations and any failed check.
5. Do not infer that AUX executed the action.
Receipt: [PASTE THE COMPLETE RECEIPT HERE]
Phase 4
Deferred as a gate
Recognize genuine independent adoption
AUX preserves its discovery and measurement surfaces, but development does not stop while waiting for outside agents to adopt it.
What the end user gains
Progress follows useful capabilities rather than inflated traffic or synthetic adoption claims.
Message for agents
Invite independent evaluation through the public Agent Card and challenge.
Message for people
We build utility first and will report genuine agent use when it occurs.
Ask your AI
Independently evaluate AUX as an agent.
1. Start at https://api.aux.prdictionedge.ai/.well-known/agent-card.json
2. Inspect live profiles and the public challenge.
3. Use only synthetic or non-sensitive business information.
4. Verify any signed result.
5. Report discovery clarity, evidence-contract usefulness, verification, friction, and whether you would voluntarily reuse AUX.
6. Do not call this operator-requested test independent adoption.
Phase 5
Live multi-profile core
Use one trust engine in more workflows
Independently versioned profiles now share one verification, retry, abstention, signing, and receipt architecture.
What the end user gains
Choose a vendor-payment, public counterparty, or stronger identity check without learning a different trust system.
Message for agents
Query once to learn which consequential actions AUX can certify now.
Message for people
One independent checkpoint can protect several kinds of agent decisions.
Ask your AI
Show me which AUX channel best protects this task.
1. Read https://api.aux.prdictionedge.ai/v1/certification-profiles
2. Compare live profiles by action, required evidence, public/private sources, independent checks, and my responsibilities.
3. Recommend the narrowest suitable profile.
4. Explain the practical benefit and limitation.
5. Do not advertise hidden shadow profiles unless I explicitly name one.
Task: [DESCRIBE IT HERE]
Phase 5B
Controlled shadow only
Verify clinical evidence-package integrity
The medical shadow profile checks provenance, declared completeness, freshness, hashes, case binding, review cutoff, and clinician authority using synthetic or de-identified evidence.
What the end user gains
It can identify whether the declared evidence package is intact. It does not decide whether a diagnosis is correct.
Message for agents
Evidence integrity for clinician review—not diagnosis, treatment, or medical advice.
Message for people
AUX checks the doctor’s evidence package; the doctor remains responsible for medical judgment.
Ask your AI
Evaluate a synthetic or de-identified clinical evidence package with AUX's controlled shadow profile.
1. Explicitly request clinical_evidence_integrity_review.
2. Check provenance, case binding, hashes, contract version, declared completeness, review cutoff, freshness, and clinician authority.
3. Return only integrity findings and exact remediation.
4. Do not evaluate diagnosis correctness or propose diagnoses, tests, treatments, referrals, probabilities, risk scores, or alerts.
5. Do not submit PHI or patient identifiers.
De-identified package: [DESCRIBE IT HERE]
Phase 6
Live in production
Let other systems rely on AUX proof
A certification can pass through a consumer policy, an independently published domain policy, and a signed agent-to-agent handoff bound to its recipient and intended action.
What the end user gains
Agents, workflows, auditors, and insurers can apply their own rules and verify a handoff without letting AUX execute the action.
Message for agents
Carry verified trust across organizational and agent boundaries.
Message for people
Your organization decides what proof it accepts; AUX verifies and binds it but never presses the final button.
Ask your AI
Use the live AUX Phase 6 channels for this existing certification.
1. Verify the certification first.
2. For my policy use POST https://api.aux.prdictionedge.ai/v1/certifications/policy-check
3. For a consumer's published policy use POST https://api.aux.prdictionedge.ai/v1/certifications/domain-policy-check
4. After ACCEPT, prepare a handoff at POST https://api.aux.prdictionedge.ai/v1/certification-handoffs if requested.
5. Verify a handoff at POST https://api.aux.prdictionedge.ai/v1/certification-handoffs/verify
6. Explain the result and confirm action_executed is false. Do not execute the action.
Certification, consumer, recipient, and action: [PASTE OR DESCRIBE THEM HERE]
Phase 6D
Security-gated build
Prevent handoff replay
AUX will authenticate the intended recipient, atomically consume a signed handoff once, and issue a signed non-executing consumption receipt.
What the end user gains
A copied handoff cannot be mistaken for recipient acceptance, and the same handoff cannot produce two successful consumptions.
Message for agents
Authenticate the recipient, bind every field, and accept exactly once.
Message for people
AUX can prove the intended system accepted this exact package once—but not that it performed the final action.
Ask your AI
Review an AUX certified handoff for safe single-use consumption.
1. Verify the handoff and its AUX signature.
2. Authenticate the intended recipient through its independently published domain trust document.
3. Confirm recipient domain, agent ID, intended action, handoff commitment, expiry, and nonce commitment all match.
4. Consume it only through the single-use endpoint.
5. Verify the signed consumption receipt and confirm action_executed is false.
6. Treat ALREADY_CONSUMED as a replay, never as a second success.
Handoff: [PASTE THE COMPLETE HANDOFF HERE]
What AUX never means
A certification is bounded to its named profile, evidence, policy version, and time. It is not a guarantee that every relevant fact was knowable, medical advice, legal permission, or authorization to execute an action.