Policy Engine
Verdicts are decided only by these rules — never by an LLM. AI is used solely to translate a rule into plain English; it cannot change any verdict, reason code, or receipt.
Identity Verification
P-IDENTITY
Identity
actor.verified == true
block
Business Purpose
P-PURPOSE
Governance
request.business_purpose != null
hold
Operator Authority
P-AUTHORITY
Authority
actor.authority_tier >= action.required_tier
escalate
Consent Scope Match
P-CONSENT
Consent
recipient in consent.allowed_recipients
hold
Recipient Authorization
P-RECIPIENT
Recipient
recipient in policy.allow_list[data_class]
hold
Data Classification Disclosure
P-DATACLASS
Data
data_class in {PHI,PII,FIN} && disclosure == external -> require_authorization
hold
Financial Delegation Limit
P-FINANCE
Finance
amount <= actor.spend_limit && !self_approval
block
Least-Privilege Tool Access
P-LEASTPRIV
Agent
requested_scopes subset_of agent.baseline_scopes
escalate
